BioAct-Het: A Heterogeneous Siamese Neural Network for Bioactivity Prediction Using Novel Bioactivity Representation

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Drug failure due to low bioactivity is a major challenge in experimental procedures.
• Predicting bioactivity classes during lead optimization is essential for enhancing compound bioactivity.
• Existing structure-activity relationship studies often overlook the multifaceted relationship between drugs and bioactivity.

Purpose of the Study:

• To propose the BioAct-Het model for predicting drug bioactivity classes.
• To model the complex relationship between drugs and bioactivity classes using a heterogeneous siamese neural network.
• To address data scarcity by introducing a novel representation for bioactivity classes (Bio-Prof) and enhancing datasets.

Main Methods:

• Developed the BioAct-Het model, a heterogeneous siamese neural network.
• Introduced Bio-Prof, a novel representation for bioactivity classes.
• Enhanced existing bioactivity datasets to mitigate data scarcity.
• Evaluated the model using association-based, bioactivity class-based, and compound-based strategies.

Main Results:

• The BioAct-Het model demonstrated superior performance compared to previous methods.
• The novel Bio-Prof representation and enhanced datasets effectively tackled data scarcity.
• The model's effectiveness was validated through diverse evaluation strategies and a case study.

Conclusions:

• The BioAct-Het model offers a robust approach for predicting drug bioactivity classes.
• The model's ability to integrate complex drug-bioactivity relationships enhances drug discovery pipelines.
• The study provides a valuable tool for mitigating drug failure risks and optimizing lead compounds.