EviCYP: In Silico Prediction of Cytochrome P450 Substrates Based on Vector Quantization and Evidential Deep Learning

Area of Science:

• Biochemistry
• Pharmacology
• Computational Biology

Background:

• Cytochrome P450 (CYP) enzymes are critical in drug metabolism, making accurate substrate identification essential for drug discovery and safety.
• Existing computational models for CYP substrate prediction face limitations due to data quality and lack of uncertainty quantification, impacting their reliability.
• There is a need for robust computational tools that can reliably predict CYP substrates and assess prediction confidence.

Purpose of the Study:

• To develop a novel computational framework, EviCYP, for accurate prediction of cytochrome P450 (CYP) substrates.
• To integrate evidential deep learning and vector quantization (VQ) to enhance prediction accuracy and quantify uncertainty.
• To provide a trustworthy tool for researchers in drug discovery and safety assessment.

Main Methods:

• Curated a high-quality dataset of 10,996 samples (4388 substrates, 2880 nonsubstrates, 3728 pseudonegatives) for nine major CYP isoforms.
• Developed the EviCYP architecture using multimodal encoders for molecular representations and enzyme sequences, incorporating VQ for feature compression.
• Implemented an evidential layer to output both class probabilities and a reliable uncertainty estimate for predictions.

Main Results:

• EviCYP achieved a high average AUROC of 0.9500 on an internal test set.
• The model demonstrated reliable uncertainty quantification, where high-uncertainty predictions strongly correlated with classification errors.
• The framework effectively processes multimodal molecular and enzymatic data, reducing redundancy through VQ.

Conclusions:

• EviCYP provides a robust and trustworthy computational tool for predicting CYP substrates.
• The integration of evidential deep learning and VQ addresses key limitations of existing models, particularly in uncertainty quantification.
• This framework has significant implications for improving the efficiency and reliability of drug discovery and safety assessment processes.