Neural interaction explainable AI predicts drug response across cancers.

Area of Science:

• Computational Biology
• Artificial Intelligence in Medicine
• Cancer Genomics

Background:

• Cancer treatment efficacy varies significantly between patients.
• Current therapies often rely on population averages, not individual molecular profiles.
• Actionable mutations guide some treatment choices, but most therapies lack personalization.

Purpose of the Study:

• To develop a scalable deep learning framework (NeurixAI) for modeling drug-gene interactions and predicting cancer drug response.
• To identify transcriptomic patterns associated with treatment outcomes.
• To leverage explainable AI (xAI) for uncovering mechanisms of drug response and resistance.

Main Methods:

• Trained a deep learning framework (NeurixAI) on a large dataset of drug perturbation experiments (546,646) and tumor molecular profiles (476).
• Modeled drug-gene interactions and transcriptomic patterns linked to drug sensitivity.
• Applied explainable AI (xAI) to identify key genes and mechanisms influencing individual tumor response.

Main Results:

• NeurixAI accurately predicted treatment responses for 272 targeted and 30 chemotherapeutic drugs in unseen tumor samples (Spearman's rho > 0.2).
• The framework demonstrated high performance on an external validation set.
• Identified 160 repurposed non-cancer drugs with anticancer potential and elucidated known/novel drug resistance mechanisms.

Conclusions:

• Integrating transcriptomics with explainable AI (xAI) offers a powerful approach for optimizing personalized cancer treatment.
• NeurixAI can enhance precision oncology by predicting drug efficacy and identifying novel therapeutic targets.
• The framework facilitates drug repurposing and provides insights into individual tumor biology for improved treatment strategies.