peerj_json_files/PeerJ_Json_116.json

{
    "v1_Abstract": "BD-Func (BiDirectional FUNCtional enrichment) is an algorithm that calculates functional enrichment by comparing lists of pre-defined genes that are known to be activated versus inhibited in a pathway or by a regulatory molecule. This paper shows that BD-Func can correctly predict cell line alternations and patient characteristics with accuracy comparable to popular algorithms, with a significantly faster run-time. BD-Func can compare scores for individual samples across multiple groups as well as provide predictive statistics and receiver operating characteristic (ROC) plots to quantify the accuracy of the signature associated with a binary phenotypic variable. BD-Func facilitates collaboration and reproducibility by encouraging users to share novel molecular signatures in the BD-Func discussion group, which is where the novel progesterone receptor and LBH589 signatures from this paper can be found. The novel LBH589 signature presented in this paper also serves as a case study showing how a custom signature using cell line data can accurately predict activity in vivo. This software is available to download at https://sourceforge.net/projects/bdfunc/. 2",
    "v1_col_introduction": "introduction  : Systems-level analysis of the combined expression pattern of multiple genes can be more\ninformative than the expression pattern of an individual gene, and there are a number of tools to calculate functional enrichment of differentially expressed genes (Huang et al. 2009; Naeem et al. 2012; Nam & Kim 2008). However, many functional annotations merely list membership in a pathway or ontology without explicitly modelling genes that should show activation or inhibition. For example, consider the KEGG canonical Wnt signalling pathway (Figure 1) (Kanehisa & Goto 2000). This gene list includes molecules that both activate and inhibit the pathway, resulting in different phenotypes (Dellinger et al. 2012; Logan & Nusse 2004). However, many functional enrichment tools would expect all the members of the pathway to behave similarly (Figure 1C), such that up-regulation of a mix of activators and inhibitors can receive a higher score than selective up-regulation of only activators within the pathway. For example, the most standard method for functional enrichment is to calculate over-representation of one gene list in another gene list, possibly using a Fisher\u2019s exact test or hypergeometric test; in the example described above, this sort of statistical test would ask if a list of differentially expressed genes shows a higher number of Wnt signalling genes than expected by chance. This sort of test cannot differentiate the behavior of those genes unless more detailed gene lists are defined (such as Wnt-inhibitors versus Wnt-agonists). This is a basic problem that BD-Func (BiDirectional FUNCtional enrichment) is designed to overcome.\nMost functional enrichment tools either require upstream filtering of gene lists (FuncAssociate\n(Berriz et al. 2009), GATHER (Chang & Nevins 2006), DAVID (Huang et al. 2008), Connectivity Map (Lamb et al. 2006), WebGestalt (Zhang et al. 2005), GoMiner (Zeeberg et al. 2003), ErmineJ (Lee et al. 2005)) and/or a comparison of signal intensities between two groups (T-profiler (Boorsma et al. 2005), GSVA (Hanzelmann et al. 2013), PAGE (Kim & Volsky 2005), GSEA (Subramanian et al. 2005), ErmineJ (Lee et al. 2005)). However, BD-Func compares the relative expression levels between activated and inhibited genes, and we show that BD-Func can successfully analyze either fold-change values between populations or raw intensity / expression values (for both microarray and RNA-Seq data). Signalling Pathway Impact Analysis (Draghici et al. 2007; Tarca et al. 2008) can be used to model activation and inhibition within a graph, but that algorithm primarily focuses on network topology (which is not always known); in contrast, BD-Func uses a simpler assumption of binning genes into two categories (activation or inhibition). Additionally, the ability to analyze absolute expression values in a single sample is a unique\n21\n22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37\n38 39 40 41 42 43 44 45 46 47 48 49\nPeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013)\nR ev ie w in g M an\nus cr ip t\nfeature only present in a limited number of functional genomic tools. For example, ASSESS (Edelman et al. 2006) can theoretically predict functional enrichment in a single sample, but that score is not taken into context amongst other samples: in other words, BD-Func uniquely uses a single sample enrichment score as a classifier, which may be useful for personalized medicine research or for the development of novel molecular signatures where the user may need to quantify the utility of the signature as a classifier. Additionally, correlations between a single sample and various samples within a database (such as SPIED (Williams 2012)) can provide information about a single sample, but this requires having a database of samples for comparison (so, this strategy will not work without the presence of an external database). Finally, BD-Func users are encouraged to share their lists of activated and inhibited genes in a simple file format. This allows easy application of models that may not be in an existing database for molecular signatures.\nThis study tests the accuracy of BD-Func on datasets that were used to define gene sets in\nMSigDB (Molecular Signatures DataBase, (Subramanian et al. 2005)), in comparison to GSEA (Gene Set Enrichment Analysis, (Mootha et al. 2003; Subramanian et al. 2005)) and IPA (Ingenuity Pathway Analysis, Ingenuity\u00ae Systems, www.ingenuity.com). IPA was selected for comparison because the upstream regulator function utilizes a similar principle as BD-Func (activation and inhibition is predicted by comparing the proportion of activated or inhibited targets, based upon annotations in the proprietary IPA database). GSEA was selected for comparison for two reasons: 1) GSEA was specifically designed to analyze MSigDB signatures, thus serving as a good baseline for positive control datasets and 2) MSigDB contains some signatures for both up- and down-regulated genes, so it is useful to compare separate analysis of these signatures (using GSEA) versus a direct comparison of up-regulated to down-regulated gene expression (using BD-Func). Different models for TGF\u03b2 and progesterone receptor (PGR) activity are also tested for robustness by application to other datasets. Finally, the utility of BD-Func to study custom gene signatures is tested with a novel signature associated with progesterone receptor status in breast cancer patients as well as a novel LBH589 signature that was defined using previously published cell line data and is validated using novel in vivo data presented in this study.",
    "v2_Abstract": "BD-Func (BiDirectional FUNCtional Enrichment) is an algorithm that calculates functional enrichment by comparing lists of genes that are known to be activated versus inhibited in a pathway or by a regulatory molecule. This paper shows that BD-Func can correctly predict cell line alternations and patient characteristics with accuracy comparable to popular algorithms, with a run-time that is up to 20 times faster. BD-Func is currently the only such algorithm which can make functional predictions on a single sample (and compare scores for individual samples across multiple groups), and BD-Func can provide predictive statistics and receiver operating characteristic (ROC) plots to quantify the accuracy of the signature associated with a binary phenotypic variable. BD-Func facilitates collaboration and reproducibility by encouraging users to share novel molecular signatures in the BD-Func discussion group, which is where the novel progesterone receptor and LBH589 signatures from this paper can be found. The novel LBH589 signature presented in this paper also serves as a case study showing how users can define a custom signature using cell line data that accurately predicts activity in vivo. This software is available to download at https://sourceforge.net/projects/bdfunc/. 1 PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t",
    "v2_col_introduction": "introduction  : Systems-level analysis of the combined expression pattern of multiple genes can sometimes\nbe more informative than the expression pattern of an individual gene, and there are a number of tools to calculate functional enrichment of differentially expressed genes (Ashburner et al. 2000; Berriz et al. 2009; Boorsma et al. 2005; Chang & Nevins 2006; Draghici et al. 2007; Edelman et al. 2006; Hanzelmann et al. 2013; Huang et al. 2008; Huang et al. 2009; Kanehisa & Goto 2000; Kim et al. 2007; Kim & Volsky 2005; Lamb et al. 2006; Lee et al. 2005; Liberzon et al. 2011; McCormack et al. 2013; Naeem et al. 2012; Nam & Kim 2008; Sartor et al. 2009; Sartor et al. 2010; Subramanian et al. 2005; Tarca et al. 2008; Vazquez et al. 2010; Williams 2012; Xu et al. 2008; Zeeberg et al. 2003; Zhang et al. 2005). However, many functional annotations merely list membership in a pathway or ontology without explicitly modelling genes that should show activation or inhibition. For example, consider the KEGG canonical Wnt signalling pathway (Figure 1) (Kanehisa & Goto 2000). This pathway captures both exhibits molecules that either activate or inhibit the pathway, resulting in different phenotypes (Dellinger et al. 2012; Logan & Nusse 2004). However, many functional enrichment tools would expect all the members of the pathway to behave similarly (Figure 1C), such that up-regulation of a mix of activators and inhibitors can receive a higher score than selected up-regulation of only activators within the pathway. This is a basic problem that BD-Func (BiDirectional FUNCtional Enrichment) is designed to overcome.\nMost functional enrichment tools either require upstream filtering of gene lists (Berriz et al.\n2009; Chang & Nevins 2006; Huang et al. 2008; Lamb et al. 2006; Vazquez et al. 2010; Zhang et al. 2005) or a comparison of signal intensities between two groups (Boorsma et al. 2005; Kim & Volsky 2005; Subramanian et al. 2005). However, BD-Func compares the relative expression levels between activated and inhibited genes, and we show that BD-Func can successfully analyze either fold-change values between populations or raw intensity / expression values (for both microarray and RNA-Seq data). In fact, the ability to analyze absolute expression values in a single sample is a unique feature only present in a limited number of functional genomic tools (Edelman et al. 2006; Williams 2012). This may be useful for personalized medicine research or for the development of novel molecular signatures where the user may need to quantify the utility of the signature as a classifier. Additionally, BD-Func users are encouraged to share their lists of activated and inhibited genes in a simple file\n2PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013)\nR ev ie w in g M an\nus cr ip t\nformat. This allows easy application of models that may not be in an existing database for molecular signatures.\nThis study tests the accuracy of BD-Func on datasets that were used to define gene sets in\nMSigDB (Molecular Signatures DataBase, (Subramanian et al. 2005)), in comparison to GSEA (Gene Set Enrichment Analysis, (Mootha et al. 2003; Subramanian et al. 2005)) and IPA (Ingenuity Pathway Analysis, Ingenuity\u00ae Systems, www.ingenuity.com). Different models for TGF\u03b2 and progesterone receptor (PGR) activity are also tested for robustness by application to other datasets. Finally, the utility of BD-Func to study custom gene signatures is tested with a novel signature associated with progesterone receptor status in breast cancer patients as well as a novel LBH589 signature that was defined using previously published cell line data and is validated using novel in vivo data presented in this study.",
    "v1_text": "results : acknowledgements : We would like to thank Christine Brown, Mike Barish, and Thanh Dellinger for discussions that led to the creation of this algorithm. We would like to thank Xiwei Wu, Zheng Liu, and two anonymous reviewers for discussions regarding the BD-Func algorithm. We would like to thank the City of Hope Functional Genomics Core for processing the microarray data. 268 269 270 271 272 273 274 275 276 277 PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t discussion : Comparison of BD-Func to GSEA indicated that BD-Func can provide similar oncogenic signature predictions with much shorter run time (Table S3) and a more direct comparison of genes that are expected to be up- or down-regulated by the oncogenic regulators. One limitation to BD-Func is that it can only conduct functional enrichment for regulators with genes that are both up- and down-regulated, so there are many gene lists in MSigDB that cannot be analyzed using BD-Func (which instead should be 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t analyzed using a tool such as GSEA). It is also worth noting that BD-Func can work with a wide range of sizes of gene lists for activated and inhibited genes (Table S4), but we would recommend using at least a few dozen genes when defining custom signatures. Comparison of the BD-Func oncogenic signatures to the IPA upstream regulators also showed that both programs could provide similar performance, which is not surprising giving the design of that module in IPA. One benefit to utilizing IPA is that IPA has a curated database which lists with a wider variety of regulators than the MSigDB oncogenic signatures that can be analyzed in BD-Func. In contrast, one major benefit to using BD-Func is the greater theoretical range of applications. For example, BD-Func provides an enrichment file for Gene Ontology (GO (Ashburner et al. 2000)) categories, but IPA does utilize this same strategy of analyzing functional ontologies by comparing the expression of positively and negatively regulated genes. The Connectivity Map is a commonly used tool to study gene expression profiles for drugs and other chemical perturbations (Lamb et al. 2006). There are no LBH589 / panobinostat treatments in the Connectivity Map database (although this database can certainly provide other useful information), so BDFunc provides a unique opportunity to test for gene signatures that show a strong positive or negative correlation with novel drug treatments (such as LBH589). Additionally, BD-Func is compatible with any gene mapping (in this case, gene symbol), whereas the Connectivity Map requires users to define their signatures in terms of HG-U133A probes. For example, affected gene symbols had to be converted to HG-U133A probes for this analysis. We believe that being able to define signatures based upon gene symbol is a substantial practical benefit. BD-Func also calculates a test statistic to represent functional activation or inhibition for each individual sample in a dataset, and this study shows how this statistic can be directly used as a classifier that can be used to quantify the predictive power of a given functional model. More specifically, this study shows the utility of using BD-Func for applying two novel predictive models (for progesterone receptor status in patients and for LBH589 drug treatment). The LBH589 signature provided biological confirmation that the results from an in vitro model can indeed apply to validation experiments in vivo. This is important because our hope is that the streamlined analysis, simple input file design, and BD-Func discussion board can be used to help scientists quickly and easily share novel predictive models. In short, BD-Func 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t provides a novel framework for functional enrichment (by comparing the relative expression of activated versus inhibited genes) that is freely available with a user interface that is accessible to biologists without any coding experience. The results of this paper show that BD-Func provides accurate predictions matched by other popular tools, which make it a useful complement to standard analysis using tools like GSEA or IPA. in this table, bd-func is used to analyze fold-change values between pr+ and pr- patients. msigdb : = CLAUS_PGR_POSITIVE_MENINGIOMA signatures. COH = novel PGR signature developed in this study. PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t Table 3: Prediction of Progesterone Receptor Status in Patient Samples Cohort BD-Func (MSigDB) BD-Func (COH) GSEA (MSigDB) IPA GSE9438 (N=31) Yes Yes No (UP) No (DN) No Huang et al. 2003 (N=88) No Yes No (UP) No (DN) No Chin et al. 2006 (N=117) No Yes No (UP) No (DN) No Anders et al. 2008 (N=73) No Yes No (UP) No (DN) No Finak et al. 2008 (N=53) No Yes No (UP) No (DN) No expO (N=256) No Yes No (UP) No (DN) No TCGA (N=739) No Yes No (UP) No (DN) No 1 2 3 4 PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t material and methods : BD-Func Algorithm 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t The basic principle behind BD-Func is to treat activated and inhibited genes as replicate observations in two populations. BD-Func is agnostic towards the type of signal used for analysis. For example, this paper uses BD-Func to study both fold-change values between two populations as well as raw signal intensities for a single column of signal values. In the paper, the t-test statistic is used to compare the expression patterns of activated and inhibited genes. However, BD-Func also allows users to compare the activated and inhibited distributions using a Mann-Whitney U test or Kolmogorov-Smirnov (K-S) test. BD-Func users have the option to calculate False Discovery Rates (FDR) using the method of Benjamini and Hochberg (Benjamini & Hochberg 1995) or the Storey q-value (Storey & Tibshirani 2003). BD-Func comes with four enrichment files: c2, c5, and c6 from MSigDB (Liberzon et al. 2011) and a list of functions defined directly from the human Gene Ontology (GO) annotations (Ashburner et al. 2000). All of these lists were created by searching for functions with both \u201cup\u201d and \u201cdown\u201d (or \u201cpositive regulation\u201d and \u201cnegative regulation\u201d) gene lists. For the human GO file, a functional annotation needed to contain at least 10 positively regulated genes and 10 negatively regulated genes. We also encourage users to share their own custom models on the BD-Func discussion group: http://sourceforge.net/p/bdfunc/discussion/ There are three different input files that can be analyzed using BD-Func (Figure S1): 1-D Input File: If the user supplies expression values for a single column of data, a density plot is created for the signal for the activated and inhibited genes. In this case, the BD-Func algorithm works exactly as described above. 2-D Input File: If the input file contains multiple columns of data, BD-Func is first run separately for each sample (represented by a column in the data matrix), as described above. Next, box-plots are created for test-statistic scores for each group (labelled in the header of the input file; Figure S2). Finally, an ANOVA p-value is provided to compare the test statistics between groups. Theoretically, test statistics could be used to make functional predictions for each sample in isolation. There are some examples in this paper where this strategy works OK. However, comparing test statistics across all samples within different groups is the only strategy that consistently works well for all the analysis presented in this paper. 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t 2-D Input File for Classifier: If the input file contains multiple columns of data and two groups (called \u201cpositive\u201d and \u201cnegative\u201d), BD-Func will create a receiver operating characteristic (ROC) plot using the test-statistic as the classification score using the ROCR package (Sing et al. 2005). This is in addition to all the calculations and output files for a normal 2-D input file containing multiple columns for BD-Func to analyze (in this paper, this represents normalized signal intensity values). Sample Acquisition and Processing Microarray datasets were downloaded from GEO (Edgar et al. 2002) or Array-Express (Parkinson et al. 2009). When raw .CEL files were available, samples were RMA normalized (Irizarry et al. 2003). Otherwise, processed intensity values were used for microarray analysis. Fold-Change values for all of the cell line datasets (TGF\u03b2 (Padua et al. 2008; Qin et al. 2009; Renzoni et al. 2004; Sartor et al. 2010; Scandura et al. 2004), mTOR (Wei et al. 2006), p53 (Elkon et al. 2005), and BRCA1(Furuta et al. 2006)) and the MSigDB progesterone receptor dataset (Claus et al. 2008) were calculated using the method of the least-squares mean using Partek\u00ae Genomics SuiteTM (Partek Inc. 2012). All other clinical samples (Anders et al. 2008; Bild et al. 2006; Chin et al. 2006; Finak et al. 2008; Huang et al. 2003; Ivshina et al. 2006; Sotiriou et al. 2003; The Cancer Genome Atlas Network 2012) were downloaded and analyzed for differential expression using BRAVO (http://bravo.coh.org/) (Deng et al. 2013, unpublished data). The novel progesterone receptor gene signature presented in this paper was produced by identifying genes in the expO dataset (GEO Series GSE2109) with a |fold-change| > 2 and an False Discovery Rate (FDR) < 0.05 (where the FDR is calculated using the method of Benjamini and Hochberg (Benjamini & Hochberg 1995) to analyze the distribution of t-test p-values). This is how the samples in this particular paper were processed, but users are not required to use this particular set of tools for preparing BD-Func input files and/or creating gene lists for custom signatures. Reads Per Kilobase per Million mapped reads (RPKM (Mortazavi et al. 2008)) values for RNA- Seq data was downloaded from the TCGA web portal (The Cancer Genome Atlas Network 2012). RPKM values were transformed by addition of 0.1 (to avoid large fold-change values for low coverage reads) followed by a log2 transformation (to normalize the signal distribution). GSEA Comparison 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t With the exception of the progesterone receptor signature (which utilized the CLAUS_PGR_POSITIVE_MENINGIOMA (Claus et al. 2008) signatures from MSigDB-c2, version 3.1), oncogenic signatures were defined using the following gene lists from MSigDB-c6 ((Liberzon et al. 2011), version 3.1) for GSEA analysis: TGFB_UP.V1 (Padua et al. 2008), MTOR_UP.N4.V1 (Wei et al. 2006), P53_DN.V2 (Elkon et al. 2005), and BRCA_DN.V1 (Furuta et al. 2006). GSEA ((Subramanian et al. 2005) ,version 2.0) calculated p-values by permutation over phenotypes whenever possible (Anders et al. 2008; Chin et al. 2006; Claus et al. 2008; Elkon et al. 2005; Finak et al. 2008; Huang et al. 2003; Padua et al. 2008; Sartor et al. 2010; Scandura et al. 2004; The Cancer Genome Atlas Network 2012; Wei et al. 2006), although there were a few datasets with less than 3 replicates for which gene sets had to be permuted instead of phenotypes (Furuta et al. 2006; Qin et al. 2009; Renzoni et al. 2004). For recovery of known perturbations of oncogenic regulators, GSEA results must either show a FWER p-value < 0.25 or a NOM p-value < 0.05, which are the default cut-offs IPA Comparison Ingenuity Pathway Analysis (IPA, Ingenuity\u00ae Systems, www.ingenuity.com) contains an \u201cUpstream Regulator\u201d module that compares the enrichment of activated and inhibited genes among up- and downregulated genes. So, the underlying principle is similar to BD-Func except it utilizes Ingenuity\u2019s propriety database of regulatory interactions and uses a z-score to calculate statistical significance between activated and inhibited genes. Gene lists in IPA were filtered for those genes showing |fold-change| > 1.5 while the entire gene list is used to define background enrichment. For recovery of known perturbations of oncogenic regulators, the upstream regulator must be identified as \u201cactivated\u201d or \u201cinhibited\u201d (|z-score| > 2), which are the default cut-offs LBH589 Signature Activated and inhibited genes were defined using overlapping gene lists defined from 3 cell line treatments that have been previously published (Kubo et al. 2013). That same study showed that LBH589 treatment significantly decreased tumor volume in exemestane (EXE) resistant MCF-7aro xenografts in mice. This study analyzes novel microarray data from EXE-resistant tumors treated with (EXE + LBH589) or without (EXE only) treatment of LBH589. All animal research procedures were approved by the City of 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t Hope Institutional Animal Care and Use Committee. This novel microarray data is available in GEO series GSE47346. In order to be included in the novel BD-Func signature, genes must show differential expression in all 3 cell lines. Genes were defined as differentially expressed if they showed a |fold-change| > 1.5 and pvalue < 0.05 , and the BD-Func signature genes had to meet these conditions for each of the 3 LBH589 cell line treatments (with consistent direction of fold-change). P-values were calculated via 1-way ANOVA with appropriate linear contrast was used to compare data sets using Partek\u00ae Genomics SuiteTM (Partek, Inc., St. Louis, MO). Fold-change values were calculated based upon the least-squares mean, and data was normalized using robust multichip average (RMA) normalization (Irizarry et al. 2003). a bd-func density plot for fold-change values for activated and inhibited genes for exe + lbh589 : vs. Exe alone tumors. At a population level, the Exe + LBH589 tumors show higher expression of activated genes whereas the Exe tumors show increased expression of inhibited genes (p= 2.0 x 10-15). 1 qualitatively detected at 2 hours but not 4 hours, but activity is not significant with p-value < 0.05 for : either time-point PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t Table 2: Prediction of TGF\u03b2 Activity in Novel Datasets BD-Func (Fold-Change) BD-Func (Intensity) GSEA IPA GSE1724 Yes Yes Yes (UP) No (DN) Yes GSE1805 No No1 No (UP) Yes (DN) No GSE6653 Yes Yes Yes (UP) Yes (DN) Yes GSE17708 Yes Yes Yes (UP) No (DN) Yes PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t Table 3(on next page) b bd-func box-plot for single-sample signature scores. exe alone shows the greatest inhibition of : LBH-related gene expression whereas EXE + LBH shows less inhibition of LBH-related gene expression, as PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t Table 1(on next page) bd-func shows equal or greater performance to gsea and ipa for functional enrichment : Given the relative ease by which samples can be classified as having positive or negative activity for an individual biomarker, the accuracy of BD-Func was first tested by applying several MSigDB oncogenic signatures (Liberzon et al. 2011) to the datasets from which the signatures were defined (Claus et al. 2008; Elkon et al. 2005; Furuta et al. 2006; Padua et al. 2008; Wei et al. 2006). BD-Func was able to detect the activation or inhibition of all of these oncogenic signatures (Table 1, Figure 2). GSEA could detect all of the signatures except the Claus et al. (Claus et al. 2008) progesterone receptor signature. Among these 5 test datasets, IPA could only detect the activity of 2 of these genes; however, this is not a completely fair comparison because we would expect some over fitting of the MSigDB signatures for the GSEA and BD-Func analysis. Nevertheless, the significance of this analysis is that BD-Func can accurately detect perturbation of all of these biomarkers on datasets where we know that these specific genes will be altered. In order to test the performance of BD-Func, GSEA, and IPA on novel datasets (which were not used by MSigDB to define gene lists), we applied the 3 algorithms to four datasets with TGF\u03b2 treatments (Table 2). All 3 algorithms showed roughly equal performance for predicting TGF\u03b2 treatment in the appropriate samples. Overall, analyses of these nine datasets indicate that BD-Func can provide similar quality results as GSEA and IPA. 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t Six breast cancer datasets were also used to test the robustness of the progesterone receptor (PGR) signature ((Anders et al. 2008; Chin et al. 2006; Finak et al. 2008; Huang et al. 2003; The Cancer Genome Atlas Network 2012)). Unfortunately, neither BD-Func, GSEA, nor IPA could predict progesterone status in all seven patient populations (Table 3). To be fair, the original Claus et al. (Claus et al. 2008) dataset was used to define progesterone receptor status in meningioma samples whereas the novel datasets tested were all breast cancer samples (where testing for over-expression of progesterone receptor is common (Bardou et al. 2003)). Nevertheless, BD-Func is designed to be able to utilize custom gene signatures with activated and inhibited, so we defined a novel progesterone receptor signature using the expO dataset (GEO Series GSE2109). This signature can identify progesterone receptor positive and negative patients for all 7 cohorts (1 meningioma and 6 breast cancer datasets), so it is robust enough for application to multiple cancer types. Another unique feature of BD-Func is the ability to use the activation versus inhibition test statistic as a classifier to define a predictive model. If a t-test statistic of 2 is used for the cut-off of distinguishing the positive and negative populations (roughly corresponding to a p-value < 0.05), it is clear that the MSigDB signature is extremely accurate at predicting PGR status in the original dataset but not in the breast cancer datasets (Table S1). Likewise, the TGF\u03b2 signature could differentiate between the treated and untreated groups if the test statistic of 2 was used as the threshold to distinguish the groups (Figure S2). However, this threshold does not work well in all circumstances: unlike the analysis of fold-change values, the p-value (for any statistical method) is not always the ideal statistic for assessment of functional enrichment on intensity values. For example, the mTOR and BRCA1 signatures (Figure 2B) show appropriate changes in test statistics that clearly distinguish treated and untreated groups, but activation and inhibition can\u2019t be defined based upon a pre-defined cut-off for the test statistic value (e.g. 2 or -2). For this reason, we provide an ANOVA p-value to quantify the difference in test-statistic between groups, where the test statistic serves as a score for a second calculation of statistical significance. Additionally, we believe that predictive statistics are a useful method for accessing BD-Func scores for individual samples within large patient populations. In order to quantify the accuracy of the model without depending on a pre-defined cut-off, BD-Func produces receiver operating characteristic 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t (ROC) plots for each cohort and the area under the curve (AUC) is calculated for each ROC plot (where a 100% accurate model would have an AUC=1.00) (Figure S3, Table S1). The superior performance of the novel PGR signature on the breast cancer cohorts becomes even clearer when these predictive statistics are compared for the two models (Figure 3, Figure S4, and Table S2). Importantly, the novel signature showed the same level of accuracy on the TCGA breast cancer dataset as the original expO dataset. This is significant because the TCGA dataset is over twice as large as the expO dataset, and the TCGA dataset utilizes RNA-Seq while the expO dataset utilizes microarrays to quantify gene expression. In other words, this shows that BD-Func is capable of producing very robust predictions that translate across different genomic technologies. LBH589 (panobinostat) is a histone deacetylase inhibitor that has been previously shown to suppress the proliferation of aromatase inhibitor resistant breast cancer cells, which was a conclusion supported in part by functional enrichment analysis of commonly affected genes from 3 cell line experiments (Kubo et al. 2013). Gene lists derived from these cell line experiments can be easily used to define a BD-func signature, so we hypothesized that the results from this previous in vitro cell culture study could be used to predict LBH589 activity in vivo in an animal study. Specifically, we asked if the signature defined based upon LBH589 treatment in 3 cell lines (H295R, MCF-7her2, HeLa) could detect LBH589 activity in a mouse xenograft from a different cell line (MCF-7aro xenograft treated with EXE). Indeed, BD-Func correctly used the cell line LBH589 signature to identify common gene expression changes in the tumours treated with LBH589 and EXE compared to the mice that were only exposed to EXE treatment (Figure 4). This figure shows the output figures from BD-Func for the 5 MSigDB signatures tested on their original datasets. A. Density plots for fold-change values for activated genes (colored red) and inhibited genes (colored green). These plots are used to illustrate BD-Func analysis on a single column of data (in this case, fold-change values between the positive and negative populations). B. Box-plots of activation versus inhibition test statistics for all relevant samples in each of the 5 MSigDB datasets. Note that each box-plot describes the distribution of test statistics for each group \u2013 it does not represent the expression of an individual gene or a metagene. In each of these five examples, the test statistic shows very clear differences among the different groups. If the median t-test statistic is greater than 2, the box is colored red. If the median t-test statistic is less than -2, the box is colored green. PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t Figure 3 many pathways are characterized by a mix of activation and inhibition : This figure shows the initial signalling steps in the Wnt signalling pathway, as defined by KEGG ( Kanehisa & Goto 2000 ) . Up-regulated genes are shown in red and down-regulated genes are shown in green. A. Complete Activation All genes activating the Wnt signalling pathway are up-regulated and all inhibitors are down-regulated. B. Complete Inhibition All genes activating the Wnt signalling pathway are down-regulated and all inhibitors are up-regulated. C. Mixed Pattern All genes in the figure are up-regulated. It is unclear what the downstream expression levels should be, but one may hypothesize a mixed result from Figure 1A and Figure 1B. However, most functional enrichment tools would predict this as the pattern with the strongest up-regulation. Underneath each diagram is the expected signal distribution that would be produced by BD-Func. PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t Figure 2 custom progesterone signature outperforms msigdb signature on breast cancer patients : AUC (Area Under the Curve) values from the ROC plots for each patient cohort are displayed. Although the MSigBD gene set shows extremely high accuracy for the original meningioma dataset, it shows essentially random predictive power for the 6 larger breast cancer datasets. On the other hand, a custom progesterone receptor signature defined using the expO dataset shows high accuracy for all 7 cohorts, and the high accuracy is maintained even in the largest cohort (TCGA). PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t Figure 4 Novel Cell Line LBH589 Signature Can Accurately Detect Drug Activity In Vivo recoveryof known perturbations for selected msigdb oncogenic genes : BD-Func \u201cFold-Change\u201d corresponds to analysis of fold-change values calculated between the perturbed and unperturbed groups. BD-Func \u201cIntensity\u201d corresponds to calculation of activity vs. inhibition score for each sample in the dataset followed by a comparison in the distribution of test statistics for all of the samples. 1p53 signal changes with sign matching p53 expression (in this case P53_DN indicates genes down-regulated by knock-down of p53, not genes negatively related by p53) PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t Table 1: Recovery of Known Perturbations for Selected MSigDB Oncogenic Genes BD-Func (Fold-Change) BD-Func (Intensity) GSEA IPA TGF\u03b2 (E-TABM-420) Yes Yes Yes (UP) Yes (DN) Yes mTOR \u2013 N4 (GSE5824) Yes Yes Yes (UP) Yes (DN) No P53 \u2013 V21 (GSE1676 ) Yes Yes Yes (UP) No (DN) Yes BRCA1 (GSE4754) Yes Yes Yes (UP) Yes (DN) No PGR (GSE9438) Yes Yes No (UP) No (DN) No 1 2 PeerJ reviewing PDF | (v2013:06:581:1:0:NEW 12 Aug 2013) R ev ie w in g M an us cr ip t Table 2(on next page) prediction of tgf\u03b2 activity in novel datasets :",
    "v2_text": "results : b bd-func box-plot for single-sample signature scores. exe alone shows the greatest inhibition of : LBH-related gene expression whereas EXE + LBH shows less inhibition of LBH-related gene expression, as PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t Table 1(on next page) acknowledgements : We would like to thank Christine Brown, Mike Barish, and Thanh Dellinger for discussions that led to the creation of this algorithm. We would like to thank Xiwei Wu and Zheng Liu for discussions regarding the BD-Func algorithm. We would like to thank the City of Hope Functional Genomics Core for processing the microarray data. 10PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t discussion : Comparison of BD-Func to GSEA indicated that BD-Func can provide similar oncogenic signature predictions with much shorter run time (Table S3) and a more direct comparison of genes that are expected to be up- or down-regulated by the oncogenic regulators. One limitation to BD-Func is that it can only conduct functional enrichment for regulators with genes that are both up- and down-regulated, so there are many gene lists in MSigDB that cannot be analyzed using BD-Func. It is also worth noting that BD-Func can work with a wide range of sizes of gene lists for activated and inhibited genes (Table S4), but we would recommend using at least a few dozen genes when defining custom signatures. Comparison of the BD-Func oncogenic signatures to the IPA upstream regulators also showed that both programs could provide similar performance, which is not surprising giving the design of that module in IPA. One benefit to utilizing IPA is that IPA has a curated database which lists with a wider variety of regulators than the MSigDB oncogenic signatures that can be analyzed in 9PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t BD-Func. In contrast, one major benefit to using BD-Func is the greater theoretical range of applications. For example, BD-Func provides an enrichment file for Gene Ontology (GO (Ashburner et al. 2000)) categories, but IPA does utilize this same strategy of analyzing functional ontologies by comparing the expression of positively and negatively regulated genes. BD-Func also calculates a test statistic to represent functional activation or inhibition for each individual sample in a dataset, and this study shows how this statistic can be directly used as a classifier that can be used to quantify the predictive power of a given functional model. More specifically, this study shows the utility of using BD-Func for applying two novel predictive models (for progesterone receptor status in patients and for LBH589 drug treatment). The LBH589 signature provided biological confirmation that the results from an in vitro model can indeed apply to validation experiments in vivo. This is important because our hope is that the streamlined analysis, simple input file design, and BD-Func discussion board can be used to help scientists quickly and easily share novel predictive models. in this table, bd-func is used to analyze fold-change values between pr+ and pr- patients. msigdb : = CLAUS_PGR_POSITIVE_MENINGIOMA signatures. COH = novel PGR signature developed in this study. PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t Table 3: Prediction of Progesterone Receptor Status in Patient Samples Cohort BD-Func (MSigDB ) BD-Func (COH) GSEA (MSigDB) IPA GSE9438 (N=31) Yes Yes No (UP) No (DN) No Huang et al. 2003 (N=88) No Yes No (UP) No (DN) No Chin et al. 2006 (N=117) No Yes No (UP) No (DN) No Anders et al. 2008 (N=73) No Yes No (UP) No (DN) No Finak et al. 2008 (N=53) No Yes No (UP) No (DN) No expO (N=256) No Yes No (UP) No (DN) No TCGA (N=739) No Yes No (UP) No (DN) No PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t a bd-func density plot for fold-change values for activated and inhibited genes for exe + lbh589 : vs. Exe alone tumors. At a population level, the Exe + LBH589 tumors show higher expression of activated genes whereas the Exe tumors show increased expression of inhibited genes (p= 2.0 x 10-15). material and methods : BD-Func Algorithm The basic principle behind BD-Func is to treat activated and inhibited genes as replicate observations in two populations. BD-Func is agnostic towards the type of signal used for analysis. For example, this paper uses BD-Func to study both fold-change values between two populations as well as raw signal intensities for an individual sample. In the paper, the t-test statistic is used to compare the expression patterns of activated and inhibited genes. However, BD-Func also allows users to compare the activated and inhibited distributions using a Mann-Whitney U test or Kolmogorov-Smirnov (K-S) test. BD-Func users have the option to calculate False Discovery Rates (FDR) using the method of Benjamini and Hochberg (Benjamini & Hochberg 1995) or the Storey q-value (Storey & Tibshirani 2003). BD-Func comes with four enrichment files: c2, c5, and c6 from MSigDB (Liberzon et al. 2011) and a list of functions defined directly from the human Gene Ontology (GO) annotations (Ashburner et al. 2000). All of these lists were created by searching for functions with both \u201cup\u201d and \u201cdown\u201d (or \u201cpositive regulation\u201d and \u201cnegative regulation\u201d) gene lists. For the human GO file, a functional annotation needed to contain at least 10 positively regulated genes and 10 negatively regulated 3PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t genes. We also encourage users to share their own custom models on the BD-Func discussion group: http://sourceforge.net/p/bdfunc/discussion/ There are three different output figures that can be created by BD-Func. If the user supplies expression values for a single column of data, a density plot is created for the signal for the activated and inhibited genes. If the input file contains multiple columns of data, box-plots are created for test-statistic scores for each group (labelled in the header of the input file; Figure S1) and an ANOVA p-value is provided to compare the test statistics between groups. If the input file contains multiple columns of data and two groups (called \u201cpositive\u201d and \u201cnegative\u201d), BD-Func will create a receiver operating characteristic (ROC) plot using the test-statistic as the classification score using the ROCR package (Sing et al. 2005). Sample Acquisition and Processing Microarray datasets were downloaded from GEO (Edgar et al. 2002) or Array-Express (Parkinson et al. 2009). When raw .CEL files were available, samples were RMA normalized (Irizarry et al. 2003). Otherwise, processed intensity values were used for microarray analysis. Fold-Change values for all of the cell line datasets (Elkon et al. 2005; Furuta et al. 2006; Padua et al. 2008; Qin et al. 2009; Renzoni et al. 2004; Sartor et al. 2010; Scandura et al. 2004; Wei et al. 2006) and the MSigDB progesterone receptor dataset (Claus et al. 2008) were calculated using the method of the least-squares mean using Partek\u00ae Genomics SuiteTM (Partek Inc. 2012). All other clinical samples (Anders et al. 2008; Bild et al. 2006; Chin et al. 2006; Finak et al. 2008; Huang et al. 2003; Ivshina et al. 2006; Sotiriou et al. 2003; The Cancer Genome Atlas Network 2012) were downloaded and analyzed for differential expression using BRAVO (http://bravo.coh.org/) (Deng et al.). The novel progesterone receptor gene signature presented in this paper was produced by identifying genes in the expO dataset (GEO Series GSE2109) with a |fold-change| > 2 and an False Discovery Rate (FDR) < 0.05 (where the FDR is calculated using the method of Benjamini and Hochberg (Benjamini & Hochberg 1995) to analyze the distribution of t-test p-values), again calculated within BRAVO (Deng et al.). Reads Per Kilobase per Million mapped reads (RPKM (Mortazavi et al. 2008)) values for RNA-Seq data was downloaded from the TCGA web portal (The Cancer Genome Atlas Network 4PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t 2012). RPKM values were transformed by addition of 0.1 (to avoid large fold-change values for low coverage reads) followed by a log2 transformation (to normalize the signal distribution). GSEA Parameters With the exception of the progesterone receptor signature (which utilized the CLAUS_PGR_POSITIVE_MENINGIOMA (Claus et al. 2008) signatures from MSigDB-c2, version 3.1), oncogenic signatures were defined using the following gene lists from MSigDB-c6 ((Liberzon et al. 2011), version 3.1) for GSEA analysis: TGFB_UP.V1 (Padua et al. 2008), MTOR_UP.N4.V1 (Wei et al. 2006), P53_DN.V2 (Elkon et al. 2005), and BRCA_DN.V1 (Furuta et al. 2006). GSEA ((Subramanian et al. 2005) ,version 2.0) calculated p-values by permutation over phenotypes whenever possible (Anders et al. 2008; Chin et al. 2006; Claus et al. 2008; Elkon et al. 2005; Finak et al. 2008; Huang et al. 2003; Padua et al. 2008; Sartor et al. 2010; Scandura et al. 2004; The Cancer Genome Atlas Network 2012; Wei et al. 2006), although there were a few datasets with less than 3 replicates for which gene sets had to be permuted instead of phenotypes (Furuta et al. 2006; Qin et al. 2009; Renzoni et al. 2004). For recovery of known perturbations of oncogenic regulators, GSEA results must either show a FWER p-value < 0.25 or a NOM p-value < 0.05. IPA Parameters Ingenuity Pathway Analysis (IPA, Ingenuity\u00ae Systems, www.ingenuity.com) contains an \u201cUpstream Regulator\u201d module that compares the enrichment of activated and inhibited genes among up- and down-regulated genes. So, the underlying principal is similar to BD-Func except it utilizes Ingenuity\u2019s propriety database of regulatory interactions and uses a z-score to calculate statistical significance between activated and inhibited genes. Gene lists in IPA were filtered for those genes showing |fold-change| > 1.5 while the entire gene list is used to define background enrichment. For recovery of known perturbations of oncogenic regulators, the upstream regulator must be identified as \u201cactivated\u201d or \u201cinhibited\u201d (|z-score| > 2). LBH589 Signature Activated and inhibited genes were defined using overlapping gene lists defined from 3 cell line treatments that have been previously published (Kubo et al. 2013). That same study showed that LBH589 treatment significantly decreased tumor volume in exemestane (EXE) resistant MCF-7aro 5PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t xenografts in mice. This study analyzes novel microarray data from EXE-resistant tumors treated with (EXE + LBH589) or without (EXE only) treatment of LBH589. All animal research procedures were approved by the City of Hope Institutional Animal Care and Use Committee. This novel microarray data is available in GEO series GSE47346. In order to be included in the novel BD-Func signature, genes must show differential expression in all 3 cell lines. Genes were defined as differentially expressed if they showed a |fold-change| > 1.5 and p-value < 0.05 , and the BD-Func signature genes had to meet these conditions for each of the 3 LBH589 cell line treatments (with consistent direction of fold-change). P-values were calculated via 1-way ANOVA with appropriate linear contrast was used to compare data sets using Partek\u00ae Genomics SuiteTM (Partek, Inc., St. Louis, MO). Fold-change values were calculated based upon the least-squares mean, and data was normalized using robust multichip average (RMA) normalization (Irizarry et al. 2003). For Connectivity Map analysis, genes were defined as differentially expressed if they showed |fold-change| > 1.5 and p-value < 0.05 (the same as for an individual cell line treatment). NetAffx was used to map gene symbols to HG-U133A probes (Liu et al. 2003). 1 qualitatively detected at 2 hours but not 4 hours, but activity is not significant with p-value < 0.05 for : either time-point PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t Table 2: Prediction of TGF\u03b2 Activity in Novel Datasets BD-Func (Fold-Change) BD-Func (Intensity) GSEA IPA GSE1724 Yes Yes Yes (UP) No (DN) Yes GSE1805 No No1 No (UP) Yes (DN) No GSE6653 Yes Yes Yes (UP) Yes (DN) Yes GSE17708 Yes Yes Yes (UP) No (DN) Yes PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t Table 3(on next page) recovery of known perturbations for selected msigdb oncogenic genes : BD-Func \u201cFold-Change\u201d corresponds to analysis of fold-change values calculated between the perturbed and unperturbed groups. BD-Func \u201cIntensity\u201d corresponds to calculation of activity vs. inhibition score for each sample in the dataset followed by a comparison in the distribution of test statistics for all of the samples. 1p53 signal changes with sign matching p53 expression (in this case P53_DN indicates genes down-regulated by knock-down of p53, not genes negatively related by p53) PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t Table 1: Recovery of Known Perturbations for Selected MSigDB Oncogenic Genes BD-Func (Fold-Chang e) BD-Func (Intensity) GSEA IPA TGF\u03b2 (E-TABM-420) Yes Yes Yes (UP) Yes (DN) Yes mTOR \u2013 N4 (GSE5824) Yes Yes Yes (UP) Yes (DN) No P53 \u2013 V21 (GSE1676 ) Yes Yes Yes (UP) No (DN) Yes BRCA1 (GSE4754) Yes Yes Yes (UP) Yes (DN) No PGR (GSE9438) Yes Yes No (UP) No (DN) No PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t Table 2(on next page) bd-func shows equal or greater performance to gsea and ipa for functional enrichment : Given the relative ease by which samples can be classified as having positive or negative activity for an individual biomarker, the accuracy of BD-Func was first tested by applying several MSigDB oncogenic signatures (Liberzon et al. 2011) to the datasets from which the signatures were defined (Elkon et al. 2005; Furuta et al. 2006; Padua et al. 2008; Renzoni et al. 2004; Wei et al. 2006). BD-Func was able to detect the activation or inhibition of all of these oncogenic signatures (Table 1, Figure 2). GSEA could detect all of the signatures except the Claus et al. (Elkon et al. 2005; Furuta et al. 2006; Renzoni et al. 2004; Wei et al. 2006) progesterone receptor signature. Among these 5 test datasets, IPA could only detect the activity of 2 of these genes; however, this is not a completely fair comparison because we would expect some over fitting of the MSigDB signatures for the GSEA and BD-Func analysis. Nevertheless, the significance of this analysis is that BD-Func can detect 6PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t accurately detect perturbation of all of these biomarkers on datasets where we know that these specific genes will be altered. In order to test the performance of BD-Func, GSEA, and IPA on novel datasets (which were not used by MSigDB to define gene lists), we applied the 3 algorithms to four datasets with TGF\u03b2 treatments (Table 2). All 3 algorithms showed roughly equal performance for predicting TGF\u03b2 treatment in the appropriate samples. Overall, analyses of these nine datasets indicate that BD-Func can provide similar quality results as GSEA and IPA. Six breast cancer datasets were also used to test the robustness of the progesterone receptor (PGR) signature ((Anders et al. 2008; Chin et al. 2006; Finak et al. 2008; Huang et al. 2003; The Cancer Genome Atlas Network 2012)). Unfortunately, neither BD-Func, GSEA, nor IPA could predict progesterone status in these two patient populations (Table 3). To be fair, the original Claus et al. (Claus et al. 2008) dataset was used to define progesterone receptor status in meningioma samples whereas the novel datasets tested were all breast cancer samples (where testing for over-expression of progesterone receptor is common (Bardou et al. 2003)). Nevertheless, BD-Func is designed to be able to utilize custom gene signatures with activated and inhibited, so we defined a novel progesterone receptor signature using the expO dataset (GEO Series GSE2109). This signature can identify progesterone receptor positive and negative patients for all 7 cohorts (1 meningioma and 6 breast cancer datasets), so it is robust enough for application to multiple cancer types. Another unique feature of BD-Func is the ability to use the activation versus inhibition test statistic as a classifier to define a predictive model. If a t-test statistic of 2 is used for the cut-off of distinguishing the positive and negative populations (roughly corresponding to a p-value < 0.05), it is clear that the MSigDB signature is extremely accurate at predicting PGR status in the original dataset but not in the breast cancer datasets (Table S1). Likewise, the TGF\u03b2 signature could differentiate between the treated and untreated groups if the test statistic of 2 was used as the threshold to distinguish the groups (Figure S1). However, this threshold does not work well in all circumstances: unlike the analysis of fold-change values, the p-value (for any statistical method) is not always the ideal statistic for assessment of functional enrichment on intensity values. For example, the mTOR and BRCA1 signatures (Figure 2B) show appropriate changes in test statistics that clearly distinguish 7PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t treated and untreated groups, but activation and inhibition can\u2019t be defined based upon a pre-defined cut-off for the test statistic value (e.g. 2 or -2). For this reason, we provide an ANOVA p-value to quantify the difference in test-statistic between groups, where the test statistic serves as a score for a second calculation of statistical significance. Additionally, we believe that predictive statistics are a useful method for accessing BD-Func scores for individual samples within large patient populations. In order to quantify the accuracy of the model without depending on a pre-defined cut-off, BD-Func produces receiver operating characteristic (ROC) plots for each cohort and the area under the curve (AUC) is calculated for each ROC plot (where a 100% accurate model would have an AUC=1.00) (Figure S2). The superior performance of the novel PGR signature on the breast cancer cohorts becomes even clearer when these predictive statistics are compared for the two models (Figure 3, Figure S3). Importantly, the novel signature showed the same level of accuracy on the TCGA breast cancer dataset as the original expO dataset. This is significant because the TCGA dataset is over twice as large as the expO dataset, and the TCGA dataset utilizes RNA-Seq while the expO dataset utilizes microarrays to quantify gene expression. In other words, this shows that BD-Func is capable of producing very robust predictions that translate across different genomic technologies. LBH589 (panobinostat) is a histone deacetylase inhibitor that has been previously shown to suppress the proliferation of aromatase inhibitor resistant breast cancer cells, which was a conclusion supported in part by functional enrichment analysis of commonly affected genes from 3 cell line experiments (Kubo et al. 2013). Gene lists derived from these cell line experiments can be easily used to define a BD-func signature, so we hypothesized that the results from this previous in vitro cell culture study could be used to predict LBH589 activity in vivo in an animal study. Specifically, we asked if the signature defined based upon LBH589 treatment in 3 cell lines (H295R, MCF-7her2, HeLa) could detect LBH589 activity in a mouse xenograft from a different cell line (MCF-7aro xenograft treated with EXE). Indeed, BD-Func correctly used the cell line LBH589 signature to identify common gene expression changes in the tumours treated with LBH589 and EXE compared to the mice that were only exposed to EXE treatment (Figure 4). 8PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t Next, we compared the BD-Func results to those profiles defined using the Connectivity Map, which is a commonly used tool to study gene expression profiles for drugs and other chemical perturbations (Lamb et al. 2006). Unfortunately, there are no LBH589 / panobinostat treatments in the Connectivity Map database. To be fair, this database can provide useful information: for example, the Connectivity Map indicates that LBH589 shows a gene expression signature that is correlated with fulvestrant treatment (an estrogen receptor antagonist used for metastatic breast cancer in postmenopausal women) and anti-correlated with estradiol and alpha-estradiol treatment. Likewise, the upstream regulator function in IPA predicts inhibition of beta-estradiol due to LBH589 treatment. All of these results match the expectation that LBH589 is an inhibitory modulator of aromatase (estrogen synthetase), resulting in the reduction of estrogen levels (57). Nevertheless, it should be emphasized that BD-Func can uniquely allow users to compare their own experiments to the LBH589 signature presented in this paper, which is not possible to achieve using these other very popular tools. Additionally, BD-Func is compatible with any gene mapping (in this case, gene symbol), whereas the Connectivity Map requires users to define their signatures in terms of HG-U133A probes. For example, affected gene symbols had to be converted to HG-U133A probes for this analysis. We believe that being able to define signatures based upon gene symbol is a substantial practical benefit. This figure shows the output figures from BD-Func for the 5 MSigDB signatures tested on their original dataset. A. Density plots for fold-change values for activated genes (colored red) and inhibited genes (colored green). These plots are used to illustrate BD-Func analysis on a single sample (in this case, fold-change values between the positive and negative populations). B. Box-plots of activation versus inhibition test statistics for all relevant samples in each of the 5 MSigDB datasets. Note that each box-plot describes the distribution of test statistics for each group \u2013 it does not represent the expression of an individual gene or a metagene. In each of these five examples, the test statistic shows very clear differences among the different groups. If the median t-test statistic is greater than 2, the box is colored red. If the median t-test statistic is less than -2, the box is colored green. PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t Figure 3 many pathways are characterized by a mix of activation and inhibition : This figure shows the initial signalling steps in the Wnt signalling pathway, as defined by KEGG ( Kanehisa & Goto 2000 ) . Up-regulated genes are shown in red and down-regulated genes are shown in green. A. Complete Activation All genes activating the Wnt signalling pathway are up-regulated and all inhibitors are down-regulated. B. Complete Inhibition All genes activating the Wnt signalling pathway are down-regulated and all inhibitors are up-regulated. C. Mixed Pattern All genes in the figure are up-regulated. It is unclear what the downstream expression levels should be, but one may hypothesize a mixed result from Figure 1A and Figure 1B. However, most functional enrichment tools would predict this as the pattern with the strongest up-regulation. Underneath each diagram is the expected signal distribution that would be produced by BD-Func. PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t Figure 2 custom progesterone signature outperforms msigdb signature on breast cancer patients : AUC (Area Under the Curve) values from the ROC plots for each patient cohort are displayed. Although the MSigBD gene set shows extremely high accuracy for the original meningioma dataset, it shows essentially random predictive power for the 6 larger breast cancer datasets. On the other hand, a custom progesterone receptor signature defined using the expO dataset shows high accuracy for all 7 cohorts, and the high accuracy is maintained even in the largest cohort (TCGA). PeerJ reviewing PDF | (v2013:06:581:0:0:NEW 13 Jun 2013) R ev ie w in g M an us cr ip t Figure 4 Novel Cell Line LBH589 Signature Can Accurately Detect Drug Activity In Vivo prediction of tgf\u03b2 activity in novel datasets :",
    "url": "https://peerj.com/articles/160/reviews/",
    "review_1": "William Jungers \u00b7 Aug 23, 2013 \u00b7 Academic Editor\nACCEPT\nThe authors have responded quickly and thoroughly to reviewer suggestions, and I believe the ms. has been improved as a result of these additions and emendations. This study provides a fascinating window into the function of an evolutionary novelty, and I believe it merits publication in its revised form.",
    "review_2": "William Jungers \u00b7 Aug 21, 2013 \u00b7 Academic Editor\nMINOR REVISIONS\nBoth reviewers believe this is an interesting and well-written article that combines 3DGM with FEM. Both reviewers request that the authors provide an illustration with attachment sites for the muscle(s) included in the models, and I encourage the authors to do so.\nOne reviewer has no other substantive comments, but the second challenges some of the specific conclusions -- and these should be addressed in the revision. I tend to agree that the unloaded model unduly impacts the ordination space, and could be excluded without loss of information. I also agree with the reviewer's comment about Figure 4. Please detail pecisely how you have responded to the various suggestions and queries in the revision response files (including rebuttals).",
    "review_3": "Reviewer 1 \u00b7 Aug 13, 2013\nBasic reporting\nNo comments\nExperimental design\nNo comments\nValidity of the findings\nNo comments\nAdditional comments\nThis is an excellent, well written article on a fascinating newly discovered species of rodent. I think that it would be very useful to include an additional figure that shows attachment sites and lines of action (or vectors) of the jaw adductor muscles used in the analyses.\n\nOne minor editorial note - line 137: Following Cox et al.........\nCite this review as\nAnonymous Reviewer (2013) Peer Review #1 of \"Masticatory biomechanics of the Laotian rock rat, Laonastes aenigmamus, and the function of the zygomaticomandibularis muscle (v0.1)\". PeerJ https://doi.org/10.7287/peerj.160v0.1/reviews/1",
    "review_4": "Reviewer 2 \u00b7 Aug 8, 2013\nBasic reporting\nNo comments\nExperimental design\nNo comments\nValidity of the findings\nNo comments\nAdditional comments\nThis is a well written and interesting paper that uses a 3D analytical approach, combining both geometric morphometrics and finite element analysis, to understand the function of the Zygomaticomandibularis muscle.\n\nI have some suggestions below that the authors should consider (particularly interpretation of the results in the discussion section), and I especially think the ms would benefit from an additional figure that shows precisely the origins and insertions of the muscles used to make the FEM \u2013 this would make the results entirely more easy to interpret for the reader.\n\nLn17: \u201cspecific diversity\u201d should be \u201ctaxonomic diversity\u201d\nLn78: insert, \u201ce.g.\u201d before \u201cDumont\u201d ?\nLn104: \u201cCox et al.\u201d typo, and same for the next ref.\nLn153: A ratio is expressed as two numbers separated by a colon e.g. \u201c2:3\u201d. It\u2019s not the same as a fraction, percentage or proportion. The values expressed in Figure 3 relating to mechanical efficiency are incorrectly described as a ratio.\nLn185: Table 4 shows not demonstrates\nLn204: it is not clear why you included the unloaded model here \u2013 this clearly causes all of the variance in all other models to be clumped together on the right hand side of the plot, as a consequence of those all being very different from the unloaded model. Would it not be more informative to produce a plot excluding the unloaded model?\nLn209: maybe you should include a reference to Fig. 2 here (e.g. also shown in Fig. 2) after \u201cregion\u201d\nLn209: insert \u201cby PC2\u201d at the end of the sentence\nLn214: looking at Fig. 2 what appears to actually be happening is dorsal movement (causing stress) of the area of the skull above the molars, i.e. between the orbits at the beginning of the calvaria surface. With procrustes superimposition, you spread the deformation across all your landmarks.\nLn221: \u201c\u201dindicating that even more deformation is occurring in these models\u201d \u2013 so we conclude that the position of the IOZM, or at least the reason it\u2019s there at all, is to limit the amount of stress in the skull (by limiting the amount of strain/displacement etc)\nLn222: this is not inferred\nLn226: insert \u201crestrained\u201d after loaded\nLn236: or the geometry of the zygomatic is adapted to restrain this?\nLn239: I wonder if this is likely? A system is adapted whereby expenditure from the temporalis would be required every time the masseter is used, otherwise the zygomatic gets snapped.\nLn244: should read \u201cbut is\u201d\nLn248: this conclusion is difficult to evaluate without a figure showing us the FEM \u2013 i.e. the muscle insertions and origins\nLn254: I think this is an incorrect conclusion: look at Fig. 4, the position of attachment (again we need a figure with this) has little effect. BUT the presence or absence of IOZM is having an effect. Look at the orange symbols (no IOZM) in Fig 4, these are separate from the others because the absence of the IOZM is causing more PC shape change (which we know is a resultant of more strain = more stress). Mean VM stress might not pick this up because it's so vague and the pattern maybe similar (this depends again on where the other muscle attachments are, which we don't know) but this is likely controlled by skull morphology (mostly).\nLn259: But the reason it\u2019s there, as shown in Fig. 4 is that it reduces landmark displacement/strain (and hence stress magnitude)\nLn263: but again, I think the data show that despite the IOZM being relatively small and having long muscle fibres, it is very effective at reducing the strain in the skull.\nLn264: this is because stress distribution is largely controlled by skull morphology in this case\nLn267: \u201creduces the bit force generated at all teeth\u201d - but increases strain (Landmark displacement in the PCA, orange dots) i.e. removing the IOZM leads to a weaker bite AND a more strained/stressed skull. The mean stress and stress distribution may not appear different in Fig. 2, but the magnitudes of the stresses within this similar distribution must be higher with IOZM removed. This appears correct in Fig. 2 along the zygomatic arch, although images are inconclusive...\nLn268: again need a figure with muscle insertions\u2026\nLn277: not sure that is correct - your data show that the presence of the IOZM evolved to increase bite force and decrease strain. Moving the IOZM has no result in Fig 4 on strain.\n\nFigure 4: the small images of the skulls showing somehow deformation on the PC axes, are difficult to read \u2013 it\u2019s hard to quickly and clearly see which regions of the skull are deforming \u2013 could the authors not add vectors to these graphics to show the direction of change, or a wireframe?\nCite this review as\nAnonymous Reviewer (2013) Peer Review #2 of \"Masticatory biomechanics of the Laotian rock rat, Laonastes aenigmamus, and the function of the zygomaticomandibularis muscle (v0.1)\". PeerJ https://doi.org/10.7287/peerj.160v0.1/reviews/2",
    "pdf_1": "https://peerj.com/articles/160v0.2/submission",
    "pdf_2": "https://peerj.com/articles/160v0.1/submission",
    "all_reviews": "Review 1: William Jungers \u00b7 Aug 23, 2013 \u00b7 Academic Editor\nACCEPT\nThe authors have responded quickly and thoroughly to reviewer suggestions, and I believe the ms. has been improved as a result of these additions and emendations. This study provides a fascinating window into the function of an evolutionary novelty, and I believe it merits publication in its revised form.\nReview 2: William Jungers \u00b7 Aug 21, 2013 \u00b7 Academic Editor\nMINOR REVISIONS\nBoth reviewers believe this is an interesting and well-written article that combines 3DGM with FEM. Both reviewers request that the authors provide an illustration with attachment sites for the muscle(s) included in the models, and I encourage the authors to do so.\nOne reviewer has no other substantive comments, but the second challenges some of the specific conclusions -- and these should be addressed in the revision. I tend to agree that the unloaded model unduly impacts the ordination space, and could be excluded without loss of information. I also agree with the reviewer's comment about Figure 4. Please detail pecisely how you have responded to the various suggestions and queries in the revision response files (including rebuttals).\nReview 3: Reviewer 1 \u00b7 Aug 13, 2013\nBasic reporting\nNo comments\nExperimental design\nNo comments\nValidity of the findings\nNo comments\nAdditional comments\nThis is an excellent, well written article on a fascinating newly discovered species of rodent. I think that it would be very useful to include an additional figure that shows attachment sites and lines of action (or vectors) of the jaw adductor muscles used in the analyses.\n\nOne minor editorial note - line 137: Following Cox et al.........\nCite this review as\nAnonymous Reviewer (2013) Peer Review #1 of \"Masticatory biomechanics of the Laotian rock rat, Laonastes aenigmamus, and the function of the zygomaticomandibularis muscle (v0.1)\". PeerJ https://doi.org/10.7287/peerj.160v0.1/reviews/1\nReview 4: Reviewer 2 \u00b7 Aug 8, 2013\nBasic reporting\nNo comments\nExperimental design\nNo comments\nValidity of the findings\nNo comments\nAdditional comments\nThis is a well written and interesting paper that uses a 3D analytical approach, combining both geometric morphometrics and finite element analysis, to understand the function of the Zygomaticomandibularis muscle.\n\nI have some suggestions below that the authors should consider (particularly interpretation of the results in the discussion section), and I especially think the ms would benefit from an additional figure that shows precisely the origins and insertions of the muscles used to make the FEM \u2013 this would make the results entirely more easy to interpret for the reader.\n\nLn17: \u201cspecific diversity\u201d should be \u201ctaxonomic diversity\u201d\nLn78: insert, \u201ce.g.\u201d before \u201cDumont\u201d ?\nLn104: \u201cCox et al.\u201d typo, and same for the next ref.\nLn153: A ratio is expressed as two numbers separated by a colon e.g. \u201c2:3\u201d. It\u2019s not the same as a fraction, percentage or proportion. The values expressed in Figure 3 relating to mechanical efficiency are incorrectly described as a ratio.\nLn185: Table 4 shows not demonstrates\nLn204: it is not clear why you included the unloaded model here \u2013 this clearly causes all of the variance in all other models to be clumped together on the right hand side of the plot, as a consequence of those all being very different from the unloaded model. Would it not be more informative to produce a plot excluding the unloaded model?\nLn209: maybe you should include a reference to Fig. 2 here (e.g. also shown in Fig. 2) after \u201cregion\u201d\nLn209: insert \u201cby PC2\u201d at the end of the sentence\nLn214: looking at Fig. 2 what appears to actually be happening is dorsal movement (causing stress) of the area of the skull above the molars, i.e. between the orbits at the beginning of the calvaria surface. With procrustes superimposition, you spread the deformation across all your landmarks.\nLn221: \u201c\u201dindicating that even more deformation is occurring in these models\u201d \u2013 so we conclude that the position of the IOZM, or at least the reason it\u2019s there at all, is to limit the amount of stress in the skull (by limiting the amount of strain/displacement etc)\nLn222: this is not inferred\nLn226: insert \u201crestrained\u201d after loaded\nLn236: or the geometry of the zygomatic is adapted to restrain this?\nLn239: I wonder if this is likely? A system is adapted whereby expenditure from the temporalis would be required every time the masseter is used, otherwise the zygomatic gets snapped.\nLn244: should read \u201cbut is\u201d\nLn248: this conclusion is difficult to evaluate without a figure showing us the FEM \u2013 i.e. the muscle insertions and origins\nLn254: I think this is an incorrect conclusion: look at Fig. 4, the position of attachment (again we need a figure with this) has little effect. BUT the presence or absence of IOZM is having an effect. Look at the orange symbols (no IOZM) in Fig 4, these are separate from the others because the absence of the IOZM is causing more PC shape change (which we know is a resultant of more strain = more stress). Mean VM stress might not pick this up because it's so vague and the pattern maybe similar (this depends again on where the other muscle attachments are, which we don't know) but this is likely controlled by skull morphology (mostly).\nLn259: But the reason it\u2019s there, as shown in Fig. 4 is that it reduces landmark displacement/strain (and hence stress magnitude)\nLn263: but again, I think the data show that despite the IOZM being relatively small and having long muscle fibres, it is very effective at reducing the strain in the skull.\nLn264: this is because stress distribution is largely controlled by skull morphology in this case\nLn267: \u201creduces the bit force generated at all teeth\u201d - but increases strain (Landmark displacement in the PCA, orange dots) i.e. removing the IOZM leads to a weaker bite AND a more strained/stressed skull. The mean stress and stress distribution may not appear different in Fig. 2, but the magnitudes of the stresses within this similar distribution must be higher with IOZM removed. This appears correct in Fig. 2 along the zygomatic arch, although images are inconclusive...\nLn268: again need a figure with muscle insertions\u2026\nLn277: not sure that is correct - your data show that the presence of the IOZM evolved to increase bite force and decrease strain. Moving the IOZM has no result in Fig 4 on strain.\n\nFigure 4: the small images of the skulls showing somehow deformation on the PC axes, are difficult to read \u2013 it\u2019s hard to quickly and clearly see which regions of the skull are deforming \u2013 could the authors not add vectors to these graphics to show the direction of change, or a wireframe?\nCite this review as\nAnonymous Reviewer (2013) Peer Review #2 of \"Masticatory biomechanics of the Laotian rock rat, Laonastes aenigmamus, and the function of the zygomaticomandibularis muscle (v0.1)\". PeerJ https://doi.org/10.7287/peerj.160v0.1/reviews/2\nReview 5: nan\nReview 6: \n",
    "LLM_extracted_review": "[\"Review 1: The authors have responded quickly and thoroughly to reviewer suggestions, and I believe the manuscript has been improved as a result of these additions and emendations. This study provides a fascinating window into the function of an evolutionary novelty, and I believe it merits publication in its revised form.\\n\\nReview 2: Both reviewers believe this is an interesting and well-written article that combines 3DGM with FEM. Both reviewers request that the authors provide an illustration with attachment sites for the muscle(s) included in the models. One reviewer has no other substantive comments, but the second challenges some of the specific conclusions -- and these should be addressed in the revision. I tend to agree that the unloaded model unduly impacts the ordination space, and could be excluded without loss of information. I also agree with the reviewer's comment about Figure 4. \\n\\nReview 3: This is an excellent, well-written article on a fascinating newly discovered species of rodent. I think that it would be very useful to include an additional figure that shows attachment sites and lines of action (or vectors) of the jaw adductor muscles used in the analyses.\\n\\nReview 4: This is a well-written and interesting paper that uses a 3D analytical approach, combining both geometric morphometrics and finite element analysis, to understand the function of the Zygomaticomandibularis muscle. I have some suggestions below that the authors should consider, particularly interpretation of the results in the discussion section. I especially think the manuscript would benefit from an additional figure that shows precisely the origins and insertions of the muscles used to make the FEM. \\n\\nReview 5: nan\\n\\nReview 6: \"]"
}