Drug target assessments: classifying target modulation and associated health effects using multi-level BERT-based...

Area of Science:

• Biomedical Informatics
• Computational Biology
• Drug Discovery

Background:

• Drug target selection is critical for successful drug development.
• Manual risk/benefit analyses are time-consuming and prone to bias.
• Large Language Models (LLMs) offer a systematic and efficient alternative for literature curation.

Purpose of the Study:

• To develop and evaluate BERT-based LLMs for classifying drug target-health effect relationships.
• To provide mechanistic insights into target modulation's impact on health and disease.
• To create an AI-assisted tool for efficient drug target identification and evaluation.

Main Methods:

• Developed BERT-models for multi-level classification of PubMed-indexed relationships.
• Classified relationships based on causality, target modulation, and health effect direction.
• Validated model performance with F1 scores ranging from 0.86 to 0.92.

Main Results:

• Achieved competitive performance with F1 scores of 0.86-0.92.
• Demonstrated applicability through case studies (ADAM33, OSM).
• The developed pipeline is the first to offer detailed classification of these relationships.

Conclusions:

• The LLMs provide mechanistic insights into drug target-health effects.
• This AI-driven approach enhances drug target identification and evaluation efficiency.
• The TargetTri platform offers a novel resource for artificial intelligence-assisted drug discovery.