Modeling of compound profiling experiments using support vector machines.

Area of Science:

• Computational chemistry
• Cheminformatics
• Pharmacology

Background:

• Compound profiling against target families is crucial in pharmaceutical research for identifying drug candidates and understanding selectivity.
• Analyzing selectivity and promiscuity patterns helps in drug discovery and development.
• Machine learning approaches are increasingly used to model complex biological data.

Purpose of the Study:

• To model compound activity profiles against a panel of 24 kinases using machine learning techniques.
• To evaluate the performance of Support Vector Machine (SVM) and Naïve Bayesian classification for predicting inhibitor activity.
• To compare the accuracy and characteristics of different classification models in predicting compound-target interactions.

Main Methods:

• Utilized Support Vector Machine (SVM) and Naïve Bayesian classification techniques.
• Modeled profiling experiments involving 429 potential inhibitors against 24 kinases.
• Analyzed prediction accuracy, including false-positive, false-negative, and true-negative rates.

Main Results:

• SVM predictions demonstrated high overall accuracy with low false-positive and high true-negative rates.
• Predictions for promiscuous inhibitors were impacted by false-negative rates.
• Combined target-based SVM classifiers matched or surpassed SVM profile prediction methods and outperformed Bayesian classification.

Conclusions:

• Machine learning, particularly SVM, is effective for modeling compound-kinase profiling data.
• SVM classifiers offer robust prediction of inhibitor activity and selectivity.
• The choice of classifier impacts the prediction of activity patterns, especially for promiscuous compounds.