Literature-derived, context-aware gene regulatory networks improve biological predictions and mathematical modeling
This repository provides reproducible code for the study by Tsutsui M. et al. (2025), focused on constructing and analyzing context-aware gene regulatory networks (GRNs) derived from biomedical literature.
This directory contains scripts and notebooks corresponding to each main figure of the paper:
-
fig2
Detection of disease-relevant PubMed literature using BERT-based embedding similarity, and evaluation of retrieval metrics (e.g., AUROC). -
fig3
Construction and visualization of context-dependent gene regulatory networks (GRNs). -
fig4
Quantitative analysis of the relationship between context-dependent GRNs and differentially expressed genes (DEGs). -
fig5
Generation of cell-type-specific gene embeddings and benchmarking their predictive performance for drug target prediction. -
fig7
Automation of mathematical model construction using context-dependent GRNs and large language models (LLMs).
- Supplementary analyses related to fig3.
- Evaluates the predictive performance of drug repurposing for unseen diseases using features derived from context-dependent GRNs.
- To reproduce results, refer to the TxGNN repository for training data, test splits, and environment setup:
👉 https://github.com/mims-harvard/TxGNN
- Code for training deep learning models using fig5’s cell-type-specific gene embeddings.
- To reproduce results, refer to the FRoGS repository for dataset splits, model architecture, and environment setup:
👉 https://github.com/chenhcs/FRoGS
Due to storage constraints, large datasets are hosted on Zenodo and must be downloaded as needed.
The automation of mathematical model generation (fig7) using LLMs is handled in a separate project:
👉 BioMathForge
Please refer to that repository for installation instructions and usage details.
This code is provided for academic use only under the MIT License.
Please cite our paper if you find this work useful.