Prediction of new bioactive molecules using a Bayesian belief network

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Natural products and synthetic compounds are crucial for novel drug development.
• Identifying biologically active small molecules remains a significant challenge in drug discovery.

Purpose of the Study:

• To introduce a novel activity prediction approach using Bayesian belief network for classification (BBNC).
• To evaluate the performance of BBNC against classical machine learning algorithms.

Main Methods:

• Developed a Bayesian belief network for classification (BBNC) where compound fragments form the network roots.
• Predicted biological activities by calculating similarity between unknown compounds and known activity classes.
• Applied BBNC to eight diverse datasets and compared results with three traditional machine learning algorithms.

Main Results:

• BBNC achieved high prediction accuracy, ranging from 79% for diverse datasets to 99% for homogeneous ones.
• The method demonstrated efficient calculation times.
• BBNC performed optimally on homogeneous datasets but showed reduced efficacy on structurally heterogeneous sets.

Conclusions:

• BBNC is an effective addition to computational chemistry tools for predicting small molecule activity.
• The approach is particularly useful for homogeneous datasets, offering high accuracy and speed.
• Combining BBNC with other prediction methods can provide a more comprehensive understanding of compound activity.