page_content stringlengths 10 9.29k | metadata dict |
|---|---|
Vol. 06, No. 02, pp. 342 –357 (2025)
ISSN: 2708-0757
JOURNAL OF APPLIED SCIENCE AND TECHNOLOGY TRENDS
www.jastt.org
342
doi: 10.38094/jastt62404
A Hybrid LLM–Knowledge Graph Framework for Accurate
Biomedical Question Answering
Havraz Y. Omar1,²... | {
"author": "",
"creationDate": "D:20251020163736+03'00'",
"creationdate": "2025-10-20T16:37:36+03:00",
"creator": "Microsoft® Word for Microsoft 365",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/JASTTytm_A+Hybrid+LLM–Knowledge+Graph_.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D... |
Omar & Mohammed / Journal of Applied Science and Technology Trends Vol. 06, No. 02, pp. 342 –357 (2025)
343
for visualizing and validating LLM outputs [16], and MedKA for
KG-enhanced question answering [17].
To address these challenges, several recent works have
explored the integration of large language models ... | {
"author": "",
"creationDate": "D:20251020163736+03'00'",
"creationdate": "2025-10-20T16:37:36+03:00",
"creator": "Microsoft® Word for Microsoft 365",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/JASTTytm_A+Hybrid+LLM–Knowledge+Graph_.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D... |
Omar & Mohammed / Journal of Applied Science and Technology Trends Vol. 06, No. 02, pp. 342 –357 (2025)
344
evaluated using Exact Match, Precision, Recall, F1, Hits@k,
MRR, and latency across simple, medium, and complex question
sets. Unlike prior template-based methods, our approach enables
traceable, outcome-l... | {
"author": "",
"creationDate": "D:20251020163736+03'00'",
"creationdate": "2025-10-20T16:37:36+03:00",
"creator": "Microsoft® Word for Microsoft 365",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/JASTTytm_A+Hybrid+LLM–Knowledge+Graph_.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D... |
Omar & Mohammed / Journal of Applied Science and Technology Trends Vol. 06, No. 02, pp. 342 –357 (2025)
345
record-high accuracy, while open-source models achieved
impressive gains through prompt optimization.
Feng et al.[22] developed the Knowledge Graph-based
Thought (KGT) framework that integrated LLMs with a... | {
"author": "",
"creationDate": "D:20251020163736+03'00'",
"creationdate": "2025-10-20T16:37:36+03:00",
"creator": "Microsoft® Word for Microsoft 365",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/JASTTytm_A+Hybrid+LLM–Knowledge+Graph_.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D... |
Omar & Mohammed / Journal of Applied Science and Technology Trends Vol. 06, No. 02, pp. 342 –357 (2025)
346
scientific literature, clinical records, genomic databases, and
experimental findings [19, 31].
Such integration creates a comprehensive biomedical
knowledge base that supports advanced analytics and disco... | {
"author": "",
"creationDate": "D:20251020163736+03'00'",
"creationdate": "2025-10-20T16:37:36+03:00",
"creator": "Microsoft® Word for Microsoft 365",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/JASTTytm_A+Hybrid+LLM–Knowledge+Graph_.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D... |
Omar & Mohammed / Journal of Applied Science and Technology Trends Vol. 06, No. 02, pp. 342 –357 (2025)
347
In biomedical research and clinical settings, LLMs help
translate natural language questions from doctors, researchers, or
patients into precise, structured queries that can be executed on
biomedical knowl... | {
"author": "",
"creationDate": "D:20251020163736+03'00'",
"creationdate": "2025-10-20T16:37:36+03:00",
"creator": "Microsoft® Word for Microsoft 365",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/JASTTytm_A+Hybrid+LLM–Knowledge+Graph_.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D... |
Omar & Mohammed / Journal of Applied Science and Technology Trends Vol. 06, No. 02, pp. 342 –357 (2025)
348
drugs, diseases, symptoms) and execution returns structured data
(tuples) relevant to the question.
Step 5. Answer Synthesis: The structured tuples flow to
Answer Synthesis, which aggregates and formats th... | {
"author": "",
"creationDate": "D:20251020163736+03'00'",
"creationdate": "2025-10-20T16:37:36+03:00",
"creator": "Microsoft® Word for Microsoft 365",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/JASTTytm_A+Hybrid+LLM–Knowledge+Graph_.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D... |
Omar & Mohammed / Journal of Applied Science and Technology Trends Vol. 06, No. 02, pp. 342 –357 (2025)
349
entities such as Diseases and Drugs are uploaded much faster,
generally under 2 seconds.
TABLE II. DATA UPLOAD TIMES FOR DIFFERENT ENTITY AND RELATIONSHIP
TYPES IN NEO4J
Entity / Relationship Type
Uploa... | {
"author": "",
"creationDate": "D:20251020163736+03'00'",
"creationdate": "2025-10-20T16:37:36+03:00",
"creator": "Microsoft® Word for Microsoft 365",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/JASTTytm_A+Hybrid+LLM–Knowledge+Graph_.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D... |
Omar & Mohammed / Journal of Applied Science and Technology Trends Vol. 06, No. 02, pp. 342 –357 (2025)
350
documentation, indexes “enable quicker and more efficient
pattern matching” by allowing the query planner to rapidly locate
nodes by label and property.
With the schema in place, data was imported using Cy... | {
"author": "",
"creationDate": "D:20251020163736+03'00'",
"creationdate": "2025-10-20T16:37:36+03:00",
"creator": "Microsoft® Word for Microsoft 365",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/JASTTytm_A+Hybrid+LLM–Knowledge+Graph_.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D... |
Omar & Mohammed / Journal of Applied Science and Technology Trends Vol. 06, No. 02, pp. 342 –357 (2025)
351
This query searches for a disease node whose name contains
’alzheimer’ and follows HAS_SYMPTOM edges to list related
symptom names. The system then executes this cypher to retrieve
answers. The prompts (su... | {
"author": "",
"creationDate": "D:20251020163736+03'00'",
"creationdate": "2025-10-20T16:37:36+03:00",
"creator": "Microsoft® Word for Microsoft 365",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/JASTTytm_A+Hybrid+LLM–Knowledge+Graph_.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D... |
Omar & Mohammed / Journal of Applied Science and Technology Trends Vol. 06, No. 02, pp. 342 –357 (2025)
352
The dataset contains 60 questions divided into three difficulty
levels based on how complex the language is and how deep the
biomedical reasoning needs to be:
• Level 1: 25 simple questions focusing mostly... | {
"author": "",
"creationDate": "D:20251020163736+03'00'",
"creationdate": "2025-10-20T16:37:36+03:00",
"creator": "Microsoft® Word for Microsoft 365",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/JASTTytm_A+Hybrid+LLM–Knowledge+Graph_.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D... |
Omar & Mohammed / Journal of Applied Science and Technology Trends Vol. 06, No. 02, pp. 342 –357 (2025)
353
Difficulty Level
Total Questions
Correct Queries
Cypher Exact Match (EM) (%)
Simple
25
24
96%
Medium
20
19
95%
Complex
15
13
86.7%
avg
92.6%
To better understand the quality of the res... | {
"author": "",
"creationDate": "D:20251020163736+03'00'",
"creationdate": "2025-10-20T16:37:36+03:00",
"creator": "Microsoft® Word for Microsoft 365",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/JASTTytm_A+Hybrid+LLM–Knowledge+Graph_.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D... |
Omar & Mohammed / Journal of Applied Science and Technology Trends Vol. 06, No. 02, pp. 342 –357 (2025)
354
P@1 refers to the accuracy of the very first retrieved item, P@5
evaluates correctness within the top three results, and P@10
considers the top ten. Higher values indicate that relevant items
tend to appea... | {
"author": "",
"creationDate": "D:20251020163736+03'00'",
"creationdate": "2025-10-20T16:37:36+03:00",
"creator": "Microsoft® Word for Microsoft 365",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/JASTTytm_A+Hybrid+LLM–Knowledge+Graph_.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D... |
Omar & Mohammed / Journal of Applied Science and Technology Trends Vol. 06, No. 02, pp. 342 –357 (2025)
355
For the question ’What are the symptoms of brain cancer?’
The system generated a Cypher query that correctly followed the
HAS_SYMPTOM relationship from disease nodes to symptoms
nodes, filtering by the spe... | {
"author": "",
"creationDate": "D:20251020163736+03'00'",
"creationdate": "2025-10-20T16:37:36+03:00",
"creator": "Microsoft® Word for Microsoft 365",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/JASTTytm_A+Hybrid+LLM–Knowledge+Graph_.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D... |
Omar & Mohammed / Journal of Applied Science and Technology Trends Vol. 06, No. 02, pp. 342 –357 (2025)
356
Fig. 7. Side effects of drugs that treat epilepsy
Executing this query returns drugs associated with epilepsy
and their corresponding side effects. For example, the query
identifies Pregabalin as a tre... | {
"author": "",
"creationDate": "D:20251020163736+03'00'",
"creationdate": "2025-10-20T16:37:36+03:00",
"creator": "Microsoft® Word for Microsoft 365",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/JASTTytm_A+Hybrid+LLM–Knowledge+Graph_.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D... |
Omar & Mohammed / Journal of Applied Science and Technology Trends Vol. 06, No. 02, pp. 342 –357 (2025)
357
[4] Z. Ji, N. Lee, R. Frieske, T. Yu, D. Su, Y. Xu, E. Ishii, Y. J. Bang, A. Madotto,
and P. Fung, “Survey of hallucination in natural language generation,”
ACM Computing Surveys, vol. 55, no. 12, pp. 1–38,... | {
"author": "",
"creationDate": "D:20251020163736+03'00'",
"creationdate": "2025-10-20T16:37:36+03:00",
"creator": "Microsoft® Word for Microsoft 365",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/JASTTytm_A+Hybrid+LLM–Knowledge+Graph_.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D... |
DALK: Dynamic Co-Augmentation of LLMs and KG to answer
Alzheimer’s Disease Questions with Scientific Literature
Dawei Li1*, Shu Yang2∗, Zhen Tan1, Jae Young Baik2, Sunkwon Yun3, Joseph Lee2, Aaron Chacko2,
Bojian Hou2, Duy Duong-Tran2,4, Ying Ding5, Huan Liu1†, Li Shen2†, Tianlong Chen3†
1School of Computing, and Augme... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
the sizes of domain-specific LLMs, consequently
affecting their performances.
To tackle these limitations, here we propose
a Dynamic Co-Augmentation of LLMs and KG
(DALK) framework that facilitates mutual benefits
between LLMs and knowledge graphs (KG) for
the AD domain. Initially, our framework addresses
the data qual... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
vice versa, let alone the potential for mutual en-
hancement between the two as we propose here.
3
Our Methodology
This section elaborates on our dynamic co-
augmentation framework of LLMs and KG. Sec-
tion 3.1 presents the details of augmenting an AD-
specific evolving KG with LLMs and literature
corpus in a time-slic... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
… Successful application of
AD preventative approaches
may hinge on an accurate and
comprehensive view of
comorbidities, including
cardiovascular disease,
diabetes, and head trauma.
Literature Corpus
LLMs for KG
KG for LLMs
Head
Relation
Tail
Diabetes
AD
Head trauma
…
…
…
risk factor
AD
risk factor
Extracted Trip... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
Moreover, many recent works (Yu et al., 2023; Li
et al., 2023d; Chen et al., 2024; Wu et al., 2024)
have demonstrated LLMs can indeed be influenced
by such noisy information. To address this chal-
lenge, we borrow insights from the recent self-
powered LLMs (Wang et al., 2022; Pan et al., 2023;
Li et al., 2023a; Yuan e... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
MedQA
MedMCQA
MMLU
QA4MRE
AVG
Biomedical LLMs
ChatDoctor (Yunxiang et al., 2023)
25.7
36.4
46.9
51.4
40.1
Med-Alpaca (Shu et al., 2023)
41.4
42.8
44.9
57.1
46.5
BiomedGPT (Zhang et al., 2023a)
38.8
41.9
48.9
42.6
43.1
Meditron (Chen et al., 2023)
27.6
31.4
36.7
25.7
30.4
Biomistral (Labrak et al., 2024)
44.7
49.5
53.1
... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
MedQA
MedMCQA
MMLU
QA4MRE
AVG
AVG Length
107.4
23.8
342.9
17.6
122.9
GPT-3.5-turbo
50.0
71.9
83.6
62.9
67.1
DALK
57.9
75.2
85.4
71.4
72.6
DALK
w/o self-aware
knowledge retrieval
56.5
71.0
77.6
77.1
70.6
Table 3: Ablation study results with and without our
proposed self-aware knowledge retrieval.
knowledge retrieval mod... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
Path-based Sub-graph
Answer
Baseline
-
C %
DALK
-w/o self-aware
knowledge retrieval
neurofibrillary tangles->FORM BY->microtubule-associated protein tau...
...
entorhinal cortex->is a part of->brain->ASSOCIATES->mouse with
Alzheimer’s disease->brain region->temporal lobe
C %
DALK
Reranked Triples1: entorhinal cortex ->... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
provides an innovative dynamic co-augmentation
framework for the refinement of large language
models and knowledge graphs. Initially, our ap-
proach extracts structural insights from the unstruc-
tured scientific literature, crafting a specialized
knowledge graph for AD. Subsequently, we employ
a coarse-to-fine samplin... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
Geesa Daluwatumulle, Rupika Wijesinghe, and Ruvan
Weerasinghe. 2023. In silico drug repurposing us-
ing knowledge graph embeddings for alzheimer’s
disease. In Proceedings of the 9th International Con-
ference on Bioinformatics Research and Applications,
ICBRA ’22, page 61–66, New York, NY, USA. As-
sociation for Comput... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
of the 2023 Conference on Empirical Methods in
Natural Language Processing, pages 6342–6353.
Alejandro Lozano, Scott L Fleming, Chia-Chun Chiang,
and Nigam Shah. 2023. Clinfo. ai: An open-source
retrieval-augmented large language model system for
answering medical questions using scientific litera-
ture. In PACIFIC SYM... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
Zhen Tan, Alimohammad Beigi, Song Wang, Ruocheng
Guo, Amrita Bhattacharjee, Bohan Jiang, Mansooreh
Karami, Jundong Li, Lu Cheng, and Huan Liu. 2024.
Large language models for data annotation: A survey.
arXiv preprint arXiv:2402.13446.
Jiabin Tang, Yuhao Yang, Wei Wei, Lei Shi, Lixin Su,
Suqi Cheng, Dawei Yin, and Chao ... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
Hongming Zhang, Xin Liu, Haojie Pan, Yangqiu Song,
and Cane Wing-Ki Leung. 2020. Aser: A large-scale
eventuality knowledge graph. In Proceedings of the
web conference 2020, pages 201–211.
Kai Zhang, Jun Yu, Zhiling Yan, Yixin Liu, Eashan Ad-
hikarla, Sunyang Fu, Xun Chen, Chen Chen, Yuyin
Zhou, Xiang Li, et al. 2023a. ... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
A
Details of LLMs for KG
Table 7 and 8 present examples of our two KG construction methods respectively. For both methods, we
adopt a select-or-generate prompt to instruct the LLM whether to choose a relation from hetionet (Him-
melstein et al., 2017), a well-built general medical KG, or generate a new one to describe ... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
Input
Read the following abstract, answer the following question. Abstract: Thiamine
pyrophosphate (TPP) and the activities of thiamine-dependent enzymes are reduced
in Alzheimer’s disease (AD) patients. In this study, we analyzed the relationship
between thiamine deficiency (TD) and amyloid precursor protein (APP) pro... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
B
Details of KG for LLMs
In this section, we provide detailed input and output for adopting KG to augment LLMs, including
path-based and neighbor-based sub-graph sampling results (Table 11), self-aware knowledge retrieval
(Table 12), describing sub-graphs with LLMs (Table 13) and inference with sampled knowledge (Table... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
Input
There are some knowledge graph paths. They follow entity->relationship->entity format.
Reranked Triples1: entorhinal cortex ->is a part of ->brain
Reranked Triples2: entorhinal cortex ->associates ->mouse with Alzheimer’s disease
Reranked Triples3: temporal lobe ->affected by ->Alzheimer’s disease
Use the knowled... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
Dataset
MedQA
MedMCQA
MMLU
QA4MRE
Total
Number
152
210
49
35
446
Table 16: Statistics of our ADQA benchmark.
D
Further Experiment for RAG
MedQA
MedMCQA
NMMLU
QA4MRE
AVG
Almanac w/ 256 chunk size
50.0
69.0
67.3
62.9
62.3
Almanac w/ top 10 docuemnt
48.7
68.6
65.3
62.9
61.4
Almanac w/ CoT
50.0
65.7
77.6
65.7
64.7
Clinfo.a... | {
"author": "",
"creationDate": "D:20240509001837Z",
"creationdate": "2024-05-09T00:18:37+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2405.04819v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240509001837Z",
"moddate": "2024-05-09T00... |
Databases and ontologies
BioThings Explorer: a query engine for a federated
knowledge graph of biomedical APIs
Jackson Callaghan
1,†, Colleen H. Xu
1,†, Jiwen Xin1,†, Marco Alvarado Cano1,
Anders Riutta
2, Eric Zhou1, Rohan Juneja1, Yao Yao1, Madhumita Narayan1,
Kristina Hanspers2, Ayushi Agrawal
2, Alexander R. Pico2,... | {
"author": "",
"creationDate": "D:20230927155028+05'30'",
"creationdate": "2023-09-27T15:50:28+05:30",
"creator": "Arbortext Advanced Print Publisher 9.0.114/W",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/btad570.pdf",
"format": "PDF 1.3",
"keywords": "",
"modDate": "D:20260111011054+00'0... |
2 A registry of semantically annotated APIs
The first step in creating a network of interoperable APIs is to
annotate each API in a semantically precise way. We built this
API annotation system on the OpenAPI specification, the de
facto standard for documenting API metadata in a human-
and machine-readable format. Open... | {
"author": "",
"creationDate": "D:20230927155028+05'30'",
"creationdate": "2023-09-27T15:50:28+05:30",
"creator": "Arbortext Advanced Print Publisher 9.0.114/W",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/btad570.pdf",
"format": "PDF 1.3",
"keywords": "",
"modDate": "D:20260111011054+00'0... |
5 Discussion
Integration of existing data from multiple disparate sources is
a key step in assessing the state of current knowledge. There
are many existing efforts to create biomedical knowledge
graphs by integrating locally downloaded data and standard-
izing it using a common data model (Himmelstein et al. 2017;
Fec... | {
"author": "",
"creationDate": "D:20230927155028+05'30'",
"creationdate": "2023-09-27T15:50:28+05:30",
"creator": "Arbortext Advanced Print Publisher 9.0.114/W",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/btad570.pdf",
"format": "PDF 1.3",
"keywords": "",
"modDate": "D:20260111011054+00'0... |
format. Second, because the entire federated KG is never in-
stantiated in a single place, reasoning and scoring methods
that rely on having the entire knowledge graph in memory
cannot be used with BioThings Explorer.
In sum, we believe that knowledge graphs enable many ex-
citing use cases in biomedical research (Nich... | {
"author": "",
"creationDate": "D:20230927155028+05'30'",
"creationdate": "2023-09-27T15:50:28+05:30",
"creator": "Arbortext Advanced Print Publisher 9.0.114/W",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/btad570.pdf",
"format": "PDF 1.3",
"keywords": "",
"modDate": "D:20260111011054+00'0... |
GigaScience , 2025, 14 , 1–12
DOI: 10.1093/gigascience/giae082
RESEARCH
Kno wledge gr aph–based thought: a kno wledge
graph–enhanced LLM framework for pan-cancer
question ans w ering
Yichun Feng
1 ,2 ,‡ , Lu Zhou 2 ,‡ , Chao Ma 3 ,‡ , Yikai Zheng 2 , Ruikun He 4 ,5 , * , and Yixue Li 1 ,2 , *
1 Hangzhou Institu... | {
"author": "Feng Yichun, Zhou Lu, Ma Chao, Zheng Yikai, He Ruikun, Li Yixue",
"creationDate": "D:20250106105103Z",
"creationdate": "2025-01-06T10:51:03+00:00",
"creator": "OUP",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/giae082.pdf",
"format": "PDF 1.4",
"keywords": "",
"modDate": "D:202... |
2 | GigaScience , 2025, Vol. 14
LLaMA with medical liter atur e. Additionall y, Med-P aLM [ 9 ] shows
promising performance on the MedQA exam based on clinical
cor por a and human feedback. Meanwhile, aiming at the Chinese
medical domain, LLMs such as BenTsao [ 10 ], DoctorGLM [ 11 ], and
HuatuoGPT [ 12 ] are deve... | {
"author": "Feng Yichun, Zhou Lu, Ma Chao, Zheng Yikai, He Ruikun, Li Yixue",
"creationDate": "D:20250106105103Z",
"creationdate": "2025-01-06T10:51:03+00:00",
"creator": "OUP",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/giae082.pdf",
"format": "PDF 1.4",
"keywords": "",
"modDate": "D:202... |
Knowledge gr a ph–based thought | 3
A
B
C
Figure 1: Illustr ativ e examples contr asting our work with pr e vious efforts. (A) LLM-onl y-based infer ence , ans wering questions solel y thr ough the
inherent knowledge of LLMs. (B) Subgraph-based inference, enhancing LLMs by retrieving the knowledge from KGs based on t... | {
"author": "Feng Yichun, Zhou Lu, Ma Chao, Zheng Yikai, He Ruikun, Li Yixue",
"creationDate": "D:20250106105103Z",
"creationdate": "2025-01-06T10:51:03+00:00",
"creator": "OUP",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/giae082.pdf",
"format": "PDF 1.4",
"keywords": "",
"modDate": "D:202... |
4 | GigaScience , 2025, Vol. 14
Table 2. Four differ ent r easoning types of task. Eac h r easoning type ma y include o v erla pping questions, so the sum acr oss the 4 differ ent
reasoning types of the task may exceed the total number of questions
Reasoning type
Claim example
Graph
Question number
One-hop
What... | {
"author": "Feng Yichun, Zhou Lu, Ma Chao, Zheng Yikai, He Ruikun, Li Yixue",
"creationDate": "D:20250106105103Z",
"creationdate": "2025-01-06T10:51:03+00:00",
"creator": "OUP",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/giae082.pdf",
"format": "PDF 1.4",
"keywords": "",
"modDate": "D:202... |
Knowledge gr a ph–based thought | 5
F igure 2: F r ame work of KGT. (A) Question anal ysis. Decompose the question and extr act its k e y information. (B) Gr a ph sc hema–based infer ence. Input
the types of the head and tail entities into the gr a ph sc hema of the knowledge gr a ph, complete the gr a ph r easoning,... | {
"author": "Feng Yichun, Zhou Lu, Ma Chao, Zheng Yikai, He Ruikun, Li Yixue",
"creationDate": "D:20250106105103Z",
"creationdate": "2025-01-06T10:51:03+00:00",
"creator": "OUP",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/giae082.pdf",
"format": "PDF 1.4",
"keywords": "",
"modDate": "D:202... |
6 | GigaScience , 2025, Vol. 14
Inference
Subgraph inference
Based on the relational chains and attribute data in the subgraph,
determine the r ele v ance to the question text. Prune an y err oneous
information, r etaining onl y the corr ect r elational c hains.
Natural language output
The LLM divides the subgr ... | {
"author": "Feng Yichun, Zhou Lu, Ma Chao, Zheng Yikai, He Ruikun, Li Yixue",
"creationDate": "D:20250106105103Z",
"creationdate": "2025-01-06T10:51:03+00:00",
"creator": "OUP",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/giae082.pdf",
"format": "PDF 1.4",
"keywords": "",
"modDate": "D:202... |
Knowledge gr a ph–based thought | 7
Table 3. Comparison of results between KGT and other commonly used methods based on the Code-Llama-13B. The best results are
displayed in bold for each indicator
ROUGE (%)
Method
GPT-4 Eval (%)
BERTScore (%)
Recall
Precision
F1 score
Base
46.6
85.3
25.3
28.5
24.5
CoT&... | {
"author": "Feng Yichun, Zhou Lu, Ma Chao, Zheng Yikai, He Ruikun, Li Yixue",
"creationDate": "D:20250106105103Z",
"creationdate": "2025-01-06T10:51:03+00:00",
"creator": "OUP",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/giae082.pdf",
"format": "PDF 1.4",
"keywords": "",
"modDate": "D:202... |
8 | GigaScience , 2025, Vol. 14
Table 5. Ablation study of the KGT fr ame work under Code-Llama-13B
ROUGE (%)
Method
GPT-4 Eval (%)
BERTScore (%)
Recall
Precision
F1 score
KGT (ours)
92.4
97.7
87.4
87.7
86.8
Without GSBI
71.8
95.5
68.1
69.8
66.8
Without QA&GSBI
69.7
94.7
55.0
66.3
58.2
Withou... | {
"author": "Feng Yichun, Zhou Lu, Ma Chao, Zheng Yikai, He Ruikun, Li Yixue",
"creationDate": "D:20250106105103Z",
"creationdate": "2025-01-06T10:51:03+00:00",
"creator": "OUP",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/giae082.pdf",
"format": "PDF 1.4",
"keywords": "",
"modDate": "D:202... |
Knowledge gr a ph–based thought | 9
A
B
C
D
Figure 4: (A), (B), (C), and (D) r espectiv el y r epr esent the r elational dia gr ams of drug r epositioning, drug r esistance r esearc h, individualized tr eatment,
and selection and understanding of biomarkers.
wer e specificall y designed to v alidate the effectiv ene... | {
"author": "Feng Yichun, Zhou Lu, Ma Chao, Zheng Yikai, He Ruikun, Li Yixue",
"creationDate": "D:20250106105103Z",
"creationdate": "2025-01-06T10:51:03+00:00",
"creator": "OUP",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/giae082.pdf",
"format": "PDF 1.4",
"keywords": "",
"modDate": "D:202... |
10 | GigaScience , 2025, Vol. 14
Abbreviations
APE: automatic prompt engineer; BFS: breadth-first search; CF:
catastr ophic for getting; CoT: c hain of thought; GPT: gener ativ e
pr etr ained tr ansformer; ICL: in-context learning; KG: knowledge
gr a ph; KGQA: knowledge gr a ph question answering; LLM: large
lang... | {
"author": "Feng Yichun, Zhou Lu, Ma Chao, Zheng Yikai, He Ruikun, Li Yixue",
"creationDate": "D:20250106105103Z",
"creationdate": "2025-01-06T10:51:03+00:00",
"creator": "OUP",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/giae082.pdf",
"format": "PDF 1.4",
"keywords": "",
"modDate": "D:202... |
Knowledge gr a ph–based thought | 11
arXiv:230515075. 24 May 2023. https:// doi.org/ 10.48550/arXiv.2
305.15075 .
13. Yang S, Zhao H, Zhu S, et al. Zhongjing: enhancing the Chi-
nese medical capabilities of large language model through
expert feedback and real-world multi-turn dialogue. In: Pro-
ceedings of the AA... | {
"author": "Feng Yichun, Zhou Lu, Ma Chao, Zheng Yikai, He Ruikun, Li Yixue",
"creationDate": "D:20250106105103Z",
"creationdate": "2025-01-06T10:51:03+00:00",
"creator": "OUP",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/giae082.pdf",
"format": "PDF 1.4",
"keywords": "",
"modDate": "D:202... |
12 | GigaScience , 2025, Vol. 14
pers), Toronto, Canada: Association for Computational Linguis-
tics; 2023:16190–206.https:// doi.org/ 10.48550/arXiv.2305.06590 .
38. Achiam J, Adler S, Agarwal S, et al. GPT-4 Technical Report
(Mar 14 version) [large language model]. 2023. arXiv preprint
arXiv:230308774. 14 Mar 202... | {
"author": "Feng Yichun, Zhou Lu, Ma Chao, Zheng Yikai, He Ruikun, Li Yixue",
"creationDate": "D:20250106105103Z",
"creationdate": "2025-01-06T10:51:03+00:00",
"creator": "OUP",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/giae082.pdf",
"format": "PDF 1.4",
"keywords": "",
"modDate": "D:202... |
An LLM-based Knowledge Synthesis and Scientific Reasoning Framework
for Biomedical Discovery
Oskar Wysocki1,2, Magdalena Wysocka2, Danilo S. Carvalho2, Alex Bogatu2,
Danilo Gusicuma1, Maxime Delmas1, Harriet Unsworth2, André Freitas1,2,3
1Idiap Research Institute, Switzerland
2National Biomarker Centre, CRUK-MI, Univ. ... | {
"author": "",
"creationDate": "D:20240628000344Z",
"creationdate": "2024-06-28T00:03:44+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2406.18626v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240628000344Z",
"moddate": "2024-06-28T00... |
to support biological analyses. We demonstrate the
key functionalities of the platform contextualised
within a real-use case in the context of molecular-
level evidence enrichment for biomarker discovery
in oncology.
2
BioLunar
BioLunar enables the creation of LLM-based
biomedical scientific workflows using software
co... | {
"author": "",
"creationDate": "D:20240628000344Z",
"creationdate": "2024-06-28T00:03:44+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2406.18626v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240628000344Z",
"moddate": "2024-06-28T00... |
Figure 1: BioLunar interface. An exemplary workflow of Gene Enrichment with an input gene set, knowledge base
query and LLM interpretation components.
a molecular score indicating evidence quality, as-
sessed by human annotators. The Gene Ontology6
(GO) offered gene function insights, and the Hu-
man Protein Atlas7 sup... | {
"author": "",
"creationDate": "D:20240628000344Z",
"creationdate": "2024-06-28T00:03:44+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2406.18626v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240628000344Z",
"moddate": "2024-06-28T00... |
Gene Enrichment
Sort and
select top N
Upload set
of genes
Interpret
results
(LLM)
Save results
Human Protein Atlas
Compute
overlap and
statistics
Upload set
of genes
Interpret
results
(LLM)
Save results
Query
Human
Protein Atlas
Run Gene
Enrichment
Provide
'Context'
Provide
'Context'
A)
B)
Figure 2: A) Gene Enrichment ... | {
"author": "",
"creationDate": "D:20240628000344Z",
"creationdate": "2024-06-28T00:03:44+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2406.18626v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240628000344Z",
"moddate": "2024-06-28T00... |
Workflow - CivicDB analysis
Discovery / new knowledge
Get genes
details and
molecular
profiles
Upload set
of genes
Save results
Provide
'Context'
Interpret genes
in the context
(LLM)
Identify well
known
molecular
profiles
Identify
molecular
profiles without
evidence
Query
PubMed
and select
top N
Prepare
PubMed
subquer... | {
"author": "",
"creationDate": "D:20240628000344Z",
"creationdate": "2024-06-28T00:03:44+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2406.18626v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240628000344Z",
"moddate": "2024-06-28T00... |
Provide
'Context'
CIVIC subworkflow
Conclusion (LLM)
Results (table)
Report
Overall conclusion
(LLM)
HPA subworkflow
Conclusion (LLM)
Results (table)
COSMIC subworkflow
Conclusion (LLM)
Results (table)
OncoKB subworkflow
Conclusion (LLM)
Results (table)
Conclusion (LLM)
Gene Enrichment
subworkflow
Conclusion (L... | {
"author": "",
"creationDate": "D:20240628000344Z",
"creationdate": "2024-06-28T00:03:44+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2406.18626v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240628000344Z",
"moddate": "2024-06-28T00... |
Cancer type: Breast Cancer
Sample size: 27
Cancer type: breast cancer
Biopsy site: primary site
Center:
Informed consent:
Sample sent:
Export report
Analysis run date: 2024-02-24
Pipeline version: v2.0. details
Context: Breast cancer (BC) presents a significant global health challenge,
wi... | {
"author": "",
"creationDate": "D:20240628000344Z",
"creationdate": "2024-06-28T00:03:44+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2406.18626v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240628000344Z",
"moddate": "2024-06-28T00... |
ological sequences, and perform gene expression
analysis by including modules supported by various
bioinformatics tools. These workflow systems are
currently centred around the composition of spe-
cialised bioinformatics software, configuration pa-
rameters and supporting datasets, facilitating reuse
and reproducibilit... | {
"author": "",
"creationDate": "D:20240628000344Z",
"creationdate": "2024-06-28T00:03:44+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2406.18626v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240628000344Z",
"moddate": "2024-06-28T00... |
A
Appendix | {
"author": "",
"creationDate": "D:20240628000344Z",
"creationdate": "2024-06-28T00:03:44+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2406.18626v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240628000344Z",
"moddate": "2024-06-28T00... |
Scenario 1
Context:
The analysis focuses on HER2-low breast cancer (HLBC), a subtype that
challenges traditional classifications based on HER2 expression and ERBB2
amplification. Despite being operationally defined, HLBCs constitute a
significant portion of breast cancers, particularly among estrogen receptor-
positive... | {
"author": "",
"creationDate": "D:20240628000344Z",
"creationdate": "2024-06-28T00:03:44+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2406.18626v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240628000344Z",
"moddate": "2024-06-28T00... |
Cancer type: Breast Cancer
Sample size: 5
Cancer type: breast cancer
Biopsy site: primary site
Center:
Informed consent:
Sample sent:
Export report
Analysis run date: 2024-02-24
Pipeline version: v2.0. details
Context:
The analysis focuses on HER2-low breast cancer (HLBC), a subtype
that ... | {
"author": "",
"creationDate": "D:20240628000344Z",
"creationdate": "2024-06-28T00:03:44+00:00",
"creator": "LaTeX with hyperref",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2406.18626v1.pdf",
"format": "PDF 1.5",
"keywords": "",
"modDate": "D:20240628000344Z",
"moddate": "2024-06-28T00... |
Nucleic Acids Research , 2025, 53 , D730–D737
https://doi.org/10.1093/nar/gkae1113
Advance access publication date: 18 November 2024
Database issue
The STRING database in 2025: protein networks with
directionality of regulation
Damian Szklarczyk 1 , 2 , Katerina Nastou
3 , Mik aela K outrouli
3 , Rebecca Kirsch... | {
"author": "Szklarczyk Damian, Nastou Katerina, Koutrouli Mikaela, Kirsch Rebecca, Mehryary Farrokh, Hachilif Radja, Hu Dewei, Peluso Matteo E., Huang Qingyao, Fang Tao, Doncheva Nadezhda T., Pyysalo Sampo, Bork Peer, Jensen Lars J., vonMering Christian",
"creationDate": "D:20241230125932+05'30'",
"creationdate"... |
Nucleic Acids Research , 2025, Vol. 53, Database issue
D 731
Introduction
The function of every living cell is primarily governed by a
complex network of interacting proteins, with each protein’s
role determined not only by its molecular activities but also
by its position within this network ( 1 ,2 ). Connected ... | {
"author": "Szklarczyk Damian, Nastou Katerina, Koutrouli Mikaela, Kirsch Rebecca, Mehryary Farrokh, Hachilif Radja, Hu Dewei, Peluso Matteo E., Huang Qingyao, Fang Tao, Doncheva Nadezhda T., Pyysalo Sampo, Bork Peer, Jensen Lars J., vonMering Christian",
"creationDate": "D:20241230125932+05'30'",
"creationdate"... |
D 732
Nucleic Acids Research , 2025, Vol. 53, Database issue
cific network modes are fully consistent with the functional
network, such that if an interaction is present in either the
physical or regulatory network, it will, by definition, also be
present in the full functional association network (with an
equal ... | {
"author": "Szklarczyk Damian, Nastou Katerina, Koutrouli Mikaela, Kirsch Rebecca, Mehryary Farrokh, Hachilif Radja, Hu Dewei, Peluso Matteo E., Huang Qingyao, Fang Tao, Doncheva Nadezhda T., Pyysalo Sampo, Bork Peer, Jensen Lars J., vonMering Christian",
"creationDate": "D:20241230125932+05'30'",
"creationdate"... |
Nucleic Acids Research , 2025, Vol. 53, Database issue
D 733
Figure 1. Illustration of the new ‘regulatory network’ mode in STRING, where the network edges visually indicate the direction, confidence and sources
of each regulatory interaction. Clicking on an edge within the network brings up a pop-up window with a d... | {
"author": "Szklarczyk Damian, Nastou Katerina, Koutrouli Mikaela, Kirsch Rebecca, Mehryary Farrokh, Hachilif Radja, Hu Dewei, Peluso Matteo E., Huang Qingyao, Fang Tao, Doncheva Nadezhda T., Pyysalo Sampo, Bork Peer, Jensen Lars J., vonMering Christian",
"creationDate": "D:20241230125932+05'30'",
"creationdate"... |
D 734
Nucleic Acids Research , 2025, Vol. 53, Database issue
uses the whole genome / proteome background; however, it
is recommended that users provide a custom background
list representing a more realistic universe of genes / proteins
detected by their assay ( 35 ). Alternatively, the entire sorted
dataset can b... | {
"author": "Szklarczyk Damian, Nastou Katerina, Koutrouli Mikaela, Kirsch Rebecca, Mehryary Farrokh, Hachilif Radja, Hu Dewei, Peluso Matteo E., Huang Qingyao, Fang Tao, Doncheva Nadezhda T., Pyysalo Sampo, Bork Peer, Jensen Lars J., vonMering Christian",
"creationDate": "D:20241230125932+05'30'",
"creationdate"... |
Nucleic Acids Research , 2025, Vol. 53, Database issue
D 735
Figure 2. Enrichment analysis interface from the STRING database website. Lef t: Sc hematic of the enrichment analysis tab highlighting various sections
of the webpage. Top right: A zoomed-in view of the analysis table with two user-highlighted terms (colo... | {
"author": "Szklarczyk Damian, Nastou Katerina, Koutrouli Mikaela, Kirsch Rebecca, Mehryary Farrokh, Hachilif Radja, Hu Dewei, Peluso Matteo E., Huang Qingyao, Fang Tao, Doncheva Nadezhda T., Pyysalo Sampo, Bork Peer, Jensen Lars J., vonMering Christian",
"creationDate": "D:20241230125932+05'30'",
"creationdate"... |
D 736
Nucleic Acids Research , 2025, Vol. 53, Database issue
sure that clusters are easily identifiable. This functionality ex-
tends beyond cluster analysis and is applicable to any gene set.
As such, it is also available through an API, complementing
our suite of other API methods. The new API function, named
g... | {
"author": "Szklarczyk Damian, Nastou Katerina, Koutrouli Mikaela, Kirsch Rebecca, Mehryary Farrokh, Hachilif Radja, Hu Dewei, Peluso Matteo E., Huang Qingyao, Fang Tao, Doncheva Nadezhda T., Pyysalo Sampo, Bork Peer, Jensen Lars J., vonMering Christian",
"creationDate": "D:20241230125932+05'30'",
"creationdate"... |
Nucleic Acids Research , 2025, Vol. 53, Database issue
D 737
networks in all domains of life, supporting directed links and
tissue-specificity. J. Mol. Biol., 433 , 166835.
13. Kim, C.Y. , Baek, S. , Cha, J. , Yang, S. , Kim, E. , Marcotte, E.M. , Hart, T.
and Lee,I. (2022) HumanNet v3: an improved database of huma... | {
"author": "Szklarczyk Damian, Nastou Katerina, Koutrouli Mikaela, Kirsch Rebecca, Mehryary Farrokh, Hachilif Radja, Hu Dewei, Peluso Matteo E., Huang Qingyao, Fang Tao, Doncheva Nadezhda T., Pyysalo Sampo, Bork Peer, Jensen Lars J., vonMering Christian",
"creationDate": "D:20241230125932+05'30'",
"creationdate"... |
Research and Applications
Improving large language model applications in
biomedicine with retrieval-augmented generation: a
systematic review, meta-analysis, and clinical
development guidelines
Siru Liu, PhD,1,2, Allison B. McCoy
, PhD1, Adam Wright
, PhD1,3
1Department of Biomedical Informatics, Vanderbilt Univ... | {
"author": "",
"creationDate": "D:20250322110116+05'30'",
"creationdate": "2025-03-22T11:01:16+05:30",
"creator": "Servigistics Arbortext Advanced Print Publisher 11.1.4667/W",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/ocaf008.pdf",
"format": "PDF 1.5",
"keywords": "large language model; r... |
was asked about medications for peripheral artery disease
patients without increased bleeding risk, it initially omitted
low-dose rivaroxaban. After integrating retrieved text from
the 2024 American College of Cardiology / American Heart
Association Guideline for the Management of Lower Extrem
ity Peripheral Arter... | {
"author": "",
"creationDate": "D:20250322110116+05'30'",
"creationdate": "2025-03-22T11:01:16+05:30",
"creator": "Servigistics Arbortext Advanced Print Publisher 11.1.4667/W",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/ocaf008.pdf",
"format": "PDF 1.5",
"keywords": "large language model; r... |
into
clinical
decision-making
and
medical
question-
answering. These analyses provided insights into how varia
tions in model architecture, retrieval strategies, evaluation
methods, and task types affect system outcomes.
To visualize the meta-analysis outcomes, we generated a
forest plot. This plot displayed t... | {
"author": "",
"creationDate": "D:20250322110116+05'30'",
"creationdate": "2025-03-22T11:01:16+05:30",
"creator": "Servigistics Arbortext Advanced Print Publisher 11.1.4667/W",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/ocaf008.pdf",
"format": "PDF 1.5",
"keywords": "large language model; r... |
Clinical applications of RAG
RAG techniques have been applied across a broad range of
medical specialties, as shown in Table 1. These applications
include clinical decision-making and medical question-
answering. In clinical decision making, RAG has supported
personalized treatment,23,24 emergency triage,25 and dis... | {
"author": "",
"creationDate": "D:20250322110116+05'30'",
"creationdate": "2025-03-22T11:01:16+05:30",
"creator": "Servigistics Arbortext Advanced Print Publisher 11.1.4667/W",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/ocaf008.pdf",
"format": "PDF 1.5",
"keywords": "large language model; r... |
integrates over 40 publicly available biomedical knowledge
sources across separate domains, such as genes, proteins,
drugs, compounds, and diseases, along with their known
relationships.32 Two studies used textbooks, such as Harri
son's Principles of Internal Medicine, while 3 others utilized
electronic health rec... | {
"author": "",
"creationDate": "D:20250322110116+05'30'",
"creationdate": "2025-03-22T11:01:16+05:30",
"creator": "Servigistics Arbortext Advanced Print Publisher 11.1.4667/W",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/ocaf008.pdf",
"format": "PDF 1.5",
"keywords": "large language model; r... |
conducting safety checks by applying 24 predefined rules to
ensure ethical and factual accuracy, and summarizing the
results.30 Glicksberg et al. developed an ensemble model that
combined structured and unstructured data to predict hospi
tal admission probabilities. These predicted probabilities,
along with simila... | {
"author": "",
"creationDate": "D:20250322110116+05'30'",
"creationdate": "2025-03-22T11:01:16+05:30",
"creator": "Servigistics Arbortext Advanced Print Publisher 11.1.4667/W",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/ocaf008.pdf",
"format": "PDF 1.5",
"keywords": "large language model; r... |
requirements. For instance, in question-answering tasks
related to broad medical exams for physicians, clinical
guidelines (eg, StatPearls) and textbooks proved more
useful than PubMed abstracts as external sources.43
Another example from our review involved a task
focused on medical question-answering in internal ... | {
"author": "",
"creationDate": "D:20250322110116+05'30'",
"creationdate": "2025-03-22T11:01:16+05:30",
"creator": "Servigistics Arbortext Advanced Print Publisher 11.1.4667/W",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/ocaf008.pdf",
"format": "PDF 1.5",
"keywords": "large language model; r... |
offer global information based on user query, offering
better performance than naïve RAG on the traditional
vector databases.50
6) Implement few-shot learning with CoT for complex clin
ical reasoning.
Few-shot learning has been shown to enhance LLMs’
reasoning capabilities by teaching specific reasoning that
may ... | {
"author": "",
"creationDate": "D:20250322110116+05'30'",
"creationdate": "2025-03-22T11:01:16+05:30",
"creator": "Servigistics Arbortext Advanced Print Publisher 11.1.4667/W",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/ocaf008.pdf",
"format": "PDF 1.5",
"keywords": "large language model; r... |
Limitations
This study was limited to peer-reviewed publications avail
able in biomedical databases (eg, PubMed, Embase), exclud
ing
preprint
articles
from
repositories
like
ArXiv.
Additionally, only studies in English language were included,
which might have excluded relevant studies in other lan
guages. We... | {
"author": "",
"creationDate": "D:20250322110116+05'30'",
"creationdate": "2025-03-22T11:01:16+05:30",
"creator": "Servigistics Arbortext Advanced Print Publisher 11.1.4667/W",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/ocaf008.pdf",
"format": "PDF 1.5",
"keywords": "large language model; r... |
17. Lefebvre C, Glanville J, Briscoe S, et al. Chapter 4: searching for
and selecting studies. Cochrane Handbook for Systematic Reviews
of Interventions Version, Vol. 6. Cochrane, 2024. https://training.
cochrane.org/handbook/current/chapter-04
18. Chapter 3 Effect Sizes j Doing Meta-Analysis in R. Accessed Octo
be... | {
"author": "",
"creationDate": "D:20250322110116+05'30'",
"creationdate": "2025-03-22T11:01:16+05:30",
"creator": "Servigistics Arbortext Advanced Print Publisher 11.1.4667/W",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/ocaf008.pdf",
"format": "PDF 1.5",
"keywords": "large language model; r... |
prompt-generated rationales. Proceedings of the AAAI Conference
on Artificial Intelligence. Association for the Advancement of Arti
ficial Intelligence; 2024:18417-18425.
53. Du X, Novoa-Laurentiev J, Plasek JM, et al. Enhancing early
detection of cognitive decline in the elderly: a comparative study
utilizing larg... | {
"author": "",
"creationDate": "D:20250322110116+05'30'",
"creationdate": "2025-03-22T11:01:16+05:30",
"creator": "Servigistics Arbortext Advanced Print Publisher 11.1.4667/W",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/ocaf008.pdf",
"format": "PDF 1.5",
"keywords": "large language model; r... |
Joy et. al
1
Federated Knowledge Retrieval Elevates Large Language Model Performance on
Biomedical Benchmarks
Janet Joy1,2, Andrew I. Su1,2
1Department of Integrative Structural and Computational Biology, Scripps Research, La
Jolla, CA, USA.
2Correspondence: Janet Joy (jjoy@scripps.edu) and Andrew I. Su (asu@s... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
2
1 Introduction
Large language models (LLMs) have rapidly advanced the state of natural-language
processing, reaching or surpassing expert performance across a wide range of
biomedical tasks, including cell type annotation, protein-structure prediction and
automated synthesis of clinical-trial resul... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
3
centric (n = 201), and drug-centric (n = 842) question–answer pairs, each explicitly
reflecting the causal flow from drug through intermediate biological nodes to disease
outcomes. Across all three DrugMechDB-derived benchmarks, BTE–RAG robustly
improves factual grounding, accelerates convergence to... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
4
explicit relationships (predicates) between pairs of entities, supplemented by
provenance details indicating the primary knowledge sources.
For each benchmark dataset, targeted queries were structured to retrieve
mechanistically relevant context. Specifically, in the gene-centric benchmark, queries ... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
5
process entities were conducted to retrieve associated chemical entities (drugs). Upon
receiving the structured knowledge graph responses from BTE, both node and edge
information were systematically processed. Nodes were extracted along with their
semantic categories and descriptive names, while edg... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
6
2.2 Datasets from DMDB
2.2 Construction of Mechanistic Question–Answer Benchmarks from
DrugMechDB
DrugMechDB is a rigorously curated biomedical knowledge graph designed to represent
therapeutic mechanisms through explicit stepwise paths. These pathways originate from
drug nodes, traverse biologic... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
7
used in the treatment of Disease 'Y' by targeting Biological Process 'P'?" The
corresponding drug node served as the ground truth. After thorough harmonization and
stringent quality control measures, this benchmark comprised 842 unique QA pairs.
The resulting benchmarks thus offer a robust, multisca... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
8
We first assessed the effect of knowledge graph augmentation on gene-level
mechanistic inference using 798 curated drug–disease pairs from DrugMechDB.
Queries were structured as: "Which gene plays the most significant mechanistic role in
how Drug 'X' treats or impacts Disease 'Y'?" Two models, GPT-4... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
9
similarity scores relative to the embedded query, and those statements falling within the
lowest 10% similarity scores were removed to retain only the most relevant context
lines. This lightweight filtering strategy preserved, and in some cases slightly enhanced
performance across all evaluated accu... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
10
confirming that retrieval largely converts borderline predictions into highly concordant
hits rather than merely redistributing low‑score failures.
Because voluminous context can inflate token budgets, we assessed performance
when progressively discarding lower‑ranked context lines (10th to 90th p... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
11
≥ 40 hits for both models across all pruning levels, demonstrating that a concise subset
of top‑ranked evidence lines is sufficient to drive the bulk of the performance gains.
3.3 Drug–Biological Process Reasoning
We next asked 842 DrugMechDB questions of the form “Which drug can be used in the
tr... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
12
gpt‑4o (blue) are nearly super‑imposable through the first ≈ 600 ranked queries
(cosine < 0.70). Beyond this inflection point, the BTE‑augmented traces bend upward
more steeply, yielding a clear margin in the high‑fidelity zone (cosine ≥ 0.80). Thus,
retrieval does not alter overall parity but sele... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
13
only" approach versus the BTE-augmented strategy (Figure 1A) across three rigorously
constructed DrugMechDB benchmarks (Figure 1B), demonstrates that incorporating
explicit, structured context significantly elevates answer accuracy, enhances
transparency, and allows smaller, more computationally ef... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
14
federated APIs could inadvertently propagate erroneous edges into model-generated
contexts. Although our evaluation leveraged the meticulously curated, high-confidence
knowledge graph of DrugMechDB, real-world applications may require strategies for
managing lower-confidence or conflicting evidence... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
15
Supplementary Figure S8: Cosine-similarity profile for the drug-centric benchmark
using GPT-4o-mini in LLM-only mode.
Supplementary Figure S9: Distribution of answer similarities for the drug-centric
benchmark using GPT-4o-mini in BTE-RAG mode.
Supplementary Figure S10: Distribution of answer simi... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
16
the views of NIA, NCATS, NIH, individual Translator team members, or affiliated
organizations and institutions.
References
1.
Hou, W. & Ji, Z. Assessing GPT-4 for cell type annotation in single-cell RNA-seq
analysis. Nature Methods 2024 21:8 21, 1462–1465 (2024).
2.
Rives, A. et al. Biologica... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
17
Association for Computational Linguistics 1906–1919 (2020)
doi:10.18653/v1/2020.acl-main.173.
13.
Yang Bs, Y., Jin, Q., Huang Phd, F. & Lu, Z. Adversarial Attacks on Large
Language Models in Medicine. (2024).
14.
Luo, R. et al. BioGPT: generative pre-trained transformer for biomedical text
gene... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
18
26.
Evangelista, J. E. et al. Toxicology knowledge graph for structural birth defects.
Communications Medicine 2023 3:1 3, 1–14 (2023).
27.
Callaghan, J. et al. BioThings Explorer: a query engine for a federated knowledge
graph of biomedical APIs. Bioinformatics 39, (2023).
28.
Carbon, S. et al... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
19
42.
Morris, J. H. et al. The scalable precision medicine open knowledge engine
(SPOKE): a massive knowledge graph of biomedical information. Bioinformatics
39, (2023).
43.
Jin, Q. et al. Pubmedqa: A dataset for biomedical research question answering.
arxiv.orgQ Jin, B Dhingra, Z Liu, WW Cohen, X... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Joy et. al
1
Supplementary Figures
Figure S1: Detailed pipeline for BTE-RAG
Supplementary Figure S1 depicts the end-to-end workflow through which the BTE-RAG retrieval module converts
a biomedical question into the evidence snippets ultimately supplied to the language-model reasoner. Beginning
with an... | {
"author": "janet",
"creationDate": "D:20250731211435-07'00'",
"creationdate": "2025-07-31T21:14:35-07:00",
"creator": "Appligent AppendPDF Pro 5.5",
"file_path": "/home/donbr/ag-discovery/gdelt-kb/data/raw/2025.08.01.668022v1.full.pdf",
"format": "PDF 1.7",
"keywords": "",
"modDate": "D:20260110171154... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.