Exploring the Potential of Spherical Harmonics and PCVM for Compounds Activity Prediction

Area of Science:

• Computational Chemistry
• Cheminformatics
• Machine Learning in Drug Discovery

Background:

• Biologically active compounds offer therapeutic potential for various diseases.
• Machine learning accelerates drug discovery by predicting molecular activity more efficiently than traditional methods.
• Developing accurate molecular descriptors and classification models is crucial for effective drug discovery.

Purpose of the Study:

• To investigate a novel representation technique, Spherical Harmonics (SH), for predicting compound activity.
• To evaluate the performance of a classifier, Probabilistic Classification Vector Machines (PCVM), when combined with SH features.
• To assess the efficacy of the SH-PCVM approach for G protein-coupled receptors (GPCRs) activity prediction.

Main Methods:

• Utilized representation learning to generate precise molecular features.
• Employed Spherical Harmonics (SH) for molecular representation.
• Integrated SH features with a Probabilistic Classification Vector Machines (PCVM) classifier, termed SH-PCVM.
• Performed ten-fold cross-validation on twenty-one GPCR datasets.

Main Results:

• The SH-PCVM approach achieved a classification accuracy of 0.86.
• Performance was evaluated using accuracy, precision, recall, Matthews' Correlation Coefficient, and Cohen's kappa.
• The SH-based representation proved effective and relatively concise for GPCR activity prediction.

Conclusions:

• The novel SH-PCVM method demonstrates highly satisfactory performance in predicting GPCR activity.
• This approach offers a promising, efficient, and accurate tool for machine learning-driven drug discovery.
• The findings highlight the potential of Spherical Harmonics as a powerful molecular descriptor for pharmacological applications.