ADMET-XSpec: A Tool for Systematic Cross-Species Data Integration in ADMET Prediction

Area of Science:

• Computational toxicology
• Machine learning in drug discovery
• Bioinformatics

Background:

• In silico methods are crucial for drug discovery and toxicology, but predicting ADMET endpoints is hindered by limited and inconsistent experimental data.
• Toxicological datasets are often fragmented across species, posing challenges for developing reliable and generalizable machine learning (ML) models.

Purpose of the Study:

• To introduce ADMET-XSpec, a Python package for systematic development, training, and evaluation of ML models for ADMET prediction.
• To enable controlled integration of interspecies and interassay data for improved model robustness and generalizability.

Main Methods:

• Development of a Python-based computational package, ADMET-XSpec.
• Facilitation of controlled incorporation of chemical space from different species and assay types.
• Support for standardized preprocessing, scalable integration of heterogeneous datasets, and rigorous benchmarking.

Main Results:

• ADMET-XSpec allows flexible construction of single-species models and models augmented with cross-species information.
• The framework enables systematic investigation of the impact of additional data from other organisms on model performance.
• Provides quantitative guidance on the conditions under which cross-species and cross-assay data integration improves predictive toxicology.

Conclusions:

• ADMET-XSpec offers a unified environment for studying cross-species effects in ADMET modeling.
• The package is a practical resource for the ADMET modeling community, enhancing model robustness and generalizability.
• It provides the first dedicated framework for controlled interspecies and interassay data integration in ADMET modeling.