Small molecule drug and biotech drug interaction prediction based on multi-modal representation learning

Area of Science:

• Pharmacology
• Computational Biology
• Drug Discovery

Background:

• Drug-drug interactions (DDIs) pose risks, necessitating early detection to prevent medical errors and reduce costs.
• Existing computational methods primarily focus on small molecule drugs (SMDs), with limited tools for biotechnology drugs (BioDs).
• The increasing prevalence of BioDs highlights the urgent need for methods predicting interactions between SMDs and BioDs.

Purpose of the Study:

• To develop a novel computational method for predicting interactions between small molecule drugs and biotechnology drugs.
• To address the limitations of current methods in handling the complexity of heterogeneous drug interactions.
• To improve the accuracy and efficiency of identifying potential adverse drug events.

Main Methods:

• A multi-modal representation learning approach (Multi-SBI) was developed to capture the complex features of both SMDs and BioDs.
• Multi-modal features were employed to represent the heterogeneous structures and relationships of different drug types.
• A Positive-unlabeled (PU) sampling technique was utilized for confident negative sample selection from unlabeled data.
• Deep Neural Network (DNN) classifiers were used to predict interaction events based on learned drug representations.

Main Results:

• The Multi-SBI method demonstrated superior performance in predicting drug-drug interactions compared to state-of-the-art methods.
• The multi-modal approach effectively extracted comprehensive features from heterogeneous drugs.
• PU-sampling successfully minimized noise during the sample selection process.
• A retrospective analysis showed high validation rates for predicted interactions, with 14 out of 20 high-confidence predictions confirmed.

Conclusions:

• The proposed Multi-SBI method offers a robust solution for predicting interactions involving biotechnology drugs.
• The integration of multi-modal features and PU-sampling enhances the accuracy of drug interaction prediction.
• Multi-SBI serves as a valuable tool for identifying novel drug interactions, contributing to improved drug safety and development.