BioMiner: A Multi-modal System for Automated Mining of Protein-Ligand Bioactivity Data from Literature

2026-04-23 · Source: Artificial Intelligence · Field: Science & Research — Life Sciences & Biology, Artificial Intelligence & Machine Learning, Health & Medical Research · Depth: Expert, quick

Summary

BioMiner is a multi-modal extraction framework designed to automate the mining of protein-ligand bioactivity data from scientific literature, addressing the bottleneck of manual curation. It separates bioactivity semantic interpretation from ligand structure construction. The system infers bioactivity semantics through direct reasoning and resolves chemical structures using a chemical-structure-grounded visual semantic reasoning paradigm, where multi-modal large language models process chemically grounded visual representations to infer inter-structure relationships, with exact molecular construction handled by domain chemistry tools. For evaluation, the BioVista benchmark was established, comprising 16,457 bioactivity entries from 500 publications. BioMiner achieved an F1 score of 0.32 for bioactivity triplets on this benchmark. Its utility is shown through applications like building a pre-training database from 82,262 data points, improving downstream model performance by 3.9%, and accelerating protein-ligand complex bioactivity annotation with a 5.59-fold speed increase.

Key takeaway

For AI Scientists and Research Scientists working on drug discovery, BioMiner offers a robust framework to significantly accelerate the extraction of protein-ligand bioactivity data. You should consider integrating such multi-modal extraction systems to build richer pre-training datasets and enhance the efficiency and accuracy of your bioactivity annotation workflows, potentially leading to faster identification of novel drug candidates.

Key insights

BioMiner automates protein-ligand bioactivity data extraction by separating semantic interpretation from chemical structure construction.

Principles

• Separate semantic interpretation from structure construction.
• Ground visual reasoning with chemical structures.

Method

BioMiner uses direct reasoning for bioactivity semantics and multi-modal LLMs on chemically grounded visual representations for structure relationships, delegating exact construction to chemistry tools.

In practice

• Build pre-training databases from literature.
• Accelerate protein-ligand complex annotation.
• Improve QSAR model performance.

Topics

• BioMiner
• Protein-Ligand Bioactivity
• Multi-modal Extraction
• Ligand Structure Reconstruction
• BioVista Benchmark

Best for: AI Scientist, Research Scientist

Related on AIssential

• CP-Agent: Context-Aware Multimodal Reasoning for Cellular Morphological Profiling under Chemical Perturbations — CP-Agent uses an MLLM and context-aware image-metadata embedding to generate interpretable rationales for drug-induced cell changes.
• Tevogen Bio’s Journey to Streamlining Life-Saving Therapies — Integrating a data lakehouse architecture with ML significantly accelerates drug discovery target identification.
• AF2BIND: predicting small-molecule binding sites using the pair representation of AlphaFold2 — AlphaFold2's internal representations can be repurposed to accurately predict de novo small-molecule binding sites.
• AI is reengineering drug discovery by speeding up testing and scanning petabytes of data for connections between diseases — AI significantly enhances drug discovery by analyzing vast data to predict targets, efficacy, and disease interrelationships, improving success rates.
• Designing Protein Binders Using the Generative Model Proteina-Complexa — Proteina-Complexa co-designs protein binders and enzymes using a latent flow-matching generative model with inference-time optimization.
• Contrastive Geometric Learning Unlocks Unified Structure- and Ligand-Based Drug Design — ConGLUDe unifies structure- and ligand-based drug design via contrastive geometric learning, enabling diverse drug discovery tasks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.