Enhancing ADMET Property Models Performance through Combinatorial Fusion Analysis.
Nan Jiang, Mohammed A. Quazi, Christina Schweikert, D. Frank Hsu, Tudor Ionel Oprea, Suman Sirimulla · ChemRxiv · 2023
Accurate prediction of Absorption, Distribution, Metabolism, Excretion, and Toxicity(ADMET) properties is crucial for drug discovery and development. However, existing computational models for ADMET predictions often lack generalizability and robustness. In this paper, we deployed a Combinatorial Fusion Analysis (CFA) to enhance the performance of ADMET models. Utilizing ADMET benchmark datasets on Therapeutics Data Commons (TDC), we conduct a comprehensive evaluation against traditional and state-of-the-art models. CFA models show superior performance compared to most of the individual models. The CFA model architecture and the performance of CFA models on TDC and other internal datasets are discussed. This significant enhancement suggests that CFA is a viable tool for improving ADMET model performance, promising faster and more cost-effective drug development pipelines. The code and models trained are available on GitHub at https://github.com/FLIDM/CFA4DD.