Advancing drug discovery with accurate and reliable protein-ligand binding free energy calculations.

The exponential growth of biomedical literature presents a major challenge for systematically identifying disease-related genes and therapeutic targets. We introduce MaBEL (Mapping of Biological Entities from Literature), a scalable and adaptable text-mining platform that unifies literature retrieval, entity recognition, and data integration into a single framework. In contrast to single-source text-mining systems, MaBEL retrieves publications from PubMed, Scopus, ScienceDirect, SciELO, and major preprint servers (bioRxiv, medRxiv, arXiv, ChemRxiv), consolidating them through DOI-based deduplication to ensure comprehensive and nonredundant coverage. The platform employs modular natural language processing pipelines, combining SciSpaCy for gene and protein recognition, BioSyn for rapid alias normalization, and PubTator 3.0 for enriched semantic and relational annotation. Built on a distributed architecture using Flask, Celery, and Docker, MaBEL supports asynchronous, large-scale text processing with near real-time performance. Applied to seven major diseases, MaBEL processed over 14,000 unique articles, achieving accurate identification of high-frequency, disease-salient genes and strong concordance with Open Targets Platform association scores. This demonstrates its reliability for uncovering biologically meaningful disease-gene relationships. By integrating multi-source retrieval, scalable computation, and modular adaptability, MaBEL represents a novel, extensible framework that advances biomedical text mining beyond static, single-database approaches, facilitating rapid hypothesis generation and accelerating the discovery of molecular targets in translational research. The source codes can be accessed at https://github.com/omixlab/Mabel.

Background:

The exponential growth of biomedical literature presents a major challenge for systematically identifying disease-related genes and therapeutic targets. Current approaches often rely on single-source text-mining systems, which may miss relevant publications or introduce redundancy due to overlapping content across databases. There is a need for a unified, scalable platform that integrates literature retrieval, entity recognition, and data processing to support comprehensive and efficient discovery of biologically meaningful disease-gene relationships.

Methods:

MaBEL (Mapping of Biological Entities from Literature) is a scalable and adaptable text-mining platform that unifies literature retrieval, entity recognition, and data integration into a single framework. It retrieves publications from multiple sources—including PubMed, Scopus, ScienceDirect, SciELO, and major preprint servers (bioRxiv, medRxiv, arXiv, ChemRxiv)—and consolidates them using DOI-based deduplication to ensure comprehensive and nonredundant coverage. The platform employs modular natural language processing pipelines: SciSpaCy for gene and protein recognition, BioSyn for rapid alias normalization, and PubTator 3.0 for enriched semantic and relational annotation. Built on a distributed architecture using Flask, Celery, and Docker, MaBEL supports asynchronous, large-scale text processing with near real-time performance.

Results:

Applied to seven major diseases, MaBEL processed over 14,000 unique articles and achieved accurate identification of high-frequency, disease-salient genes. The results showed strong concordance with Open Targets Platform association scores, demonstrating the platform’s reliability in uncovering biologically meaningful disease-gene relationships. The integration of multi-source retrieval and modular NLP components enabled robust and reproducible extraction of gene-disease associations across diverse biomedical literature.

Data Summary:

MaBEL processed more than 14,000 unique articles across seven major diseases. The platform demonstrated high accuracy in identifying disease-salient genes, with results showing strong concordance with established association scores from the Open Targets Platform. This quantitative validation confirms the effectiveness of MaBEL’s unified approach in extracting reliable gene-disease relationships from large-scale literature.

Conclusions:

MaBEL represents a novel, extensible framework that advances biomedical text mining beyond static, single-database approaches. By integrating multi-source retrieval, scalable computation, and modular adaptability, it enables comprehensive and efficient identification of disease-related genes and therapeutic targets. The platform’s performance across seven diseases validates its utility for hypothesis generation and molecular target discovery in translational research.

Practical Significance:

MaBEL facilitates rapid hypothesis generation and accelerates the discovery of molecular targets in translational research by providing a reliable, scalable, and adaptable tool for mining biomedical literature. Its open-source availability (https://github.com/omixlab/Mabel) allows researchers to apply it to diverse disease contexts, supporting data-driven prioritization of candidate genes and enhancing the efficiency of target validation pipelines in drug development and precision medicine.