XAI-ACSM: An Ensemble-Based Explainable Artificial Intelligence Framework for the Accurate Prediction of Anticancer...

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence in oncology

Background:

• Conventional cancer therapies face challenges like toxicity and drug resistance.
• Small-molecule drugs offer advantages in oral bioavailability and systemic efficacy.
• Machine learning (ML) accelerates the identification of anticancer drug candidates.

Purpose of the Study:

• To develop a novel ensemble-based explainable artificial intelligence (XAI) framework, XAI-ACSM.
• To identify and characterize anticancer small molecules (ACSMs) using only SMILES notation.
• To improve the accuracy and efficiency of anticancer drug discovery pipelines.

Main Methods:

• Developed an ensemble XAI framework (XAI-ACSM) integrating ML algorithms and molecular descriptors.
• Evaluated five ML algorithms and 14 molecular descriptors from five encoding schemes.
• Constructed 70 baseline models and integrated the best performing ones using probability averaging.

Main Results:

• XAI-ACSM outperformed baseline models and existing methods in cross-validation and independent tests.
• Achieved high accuracy (0.826), specificity (0.926), and MCC (0.666) on independent test data.
• Identified potential ACSMs among FDA-approved drugs, validated via molecular docking.

Conclusions:

• XAI-ACSM provides a practical and efficient method for screening chemical libraries to find potential ACSMs.
• The framework aids in identifying compounds with limited existing characterization.
• Reduces time and resource requirements in the anticancer drug discovery process.