Crossfeat: a transformer-based cross-feature learning model for predicting drug side effect frequency.

Area of Science:

• Pharmacology
• Computational Biology
• Artificial Intelligence in Medicine

Background:

• Accurate prediction of drug side effect frequencies is crucial for safe medication use.
• Current computational methods struggle with new drugs due to reliance on existing drug-side effect data.
• Existing models often show unreliable performance for novel drugs by omitting critical drug-side effect relationships.

Purpose of the Study:

• To develop a computational model for predicting the occurrence and frequency of drug side effects for new drugs.
• To overcome the limitations of existing methods that require prior drug-side effect information.
• To enhance the safety of drug treatments by providing reliable side effect predictions for novel therapeutics.

Main Methods:

• Proposed CrossFeat, a model utilizing a convolutional neural network-transformer architecture.
• Implemented cross-feature learning for concurrent learning of drug and side effect information.
• Enabled bidirectional learning where drugs learn associated side effects and vice versa.

Main Results:

• CrossFeat accurately predicts drug side effect frequencies for new drugs without prior relationship data.
• Demonstrated superior performance over existing methods in five-fold cross-validation experiments.
• Showcased effective integration of drug and side effect knowledge through bidirectional learning.

Conclusions:

• CrossFeat presents a promising approach for predicting drug side effect frequencies, especially for new drugs with limited data.
• The model's effectiveness is supported by cross-validation, case studies, and ablation experiments.
• CrossFeat enhances drug safety by providing reliable predictions for previously uncharacterized drug-related adverse events.