ProfhEX: AI-based platform for small molecules liability profiling.

Area of Science:

• Computational chemistry and cheminformatics
• Artificial intelligence in drug discovery
• Toxicology and safety pharmacology

Background:

• Off-target drug interactions cause significant failures in drug candidate development.
• Early prediction of adverse effects is crucial to minimize patient risks, animal testing, and costs.
• Increasing chemical library sizes necessitate efficient screening tools for liability estimation.

Purpose of the Study:

• To introduce ProfhEX, an AI-driven platform for profiling small molecules.
• To predict potential liabilities across seven key toxicity groups using machine learning.
• To provide a first-tier screening tool for early-stage drug candidate assessment.

Main Methods:

• Developed 46 OECD-compliant machine learning models using gradient boosting and random forest algorithms.
• Trained models on experimental affinity data for 210,116 unique compounds across 46 targets.
• Validated models rigorously using internal (cross-validation, bootstrap, y-scrambling) and external datasets.

Main Results:

• Champion models achieved an average Pearson correlation coefficient of 0.84 and R2 of 0.68.
• Demonstrated strong hit-detection power across all liability groups (average enrichment factor at 5% of 13.1).
• ProfhEX models showed superior predictive power for large-scale liability profiling compared to existing tools.

Conclusions:

• ProfhEX provides a robust AI-driven solution for early-stage liability profiling in drug discovery.
• The platform effectively predicts potential adverse effects, aiding in the selection of safer drug candidates.
• Future expansion includes new targets and complementary modeling approaches, enhancing its utility.