Development of a Web-Enabled SVR-Based Machine Learning Platform and its Application on Modeling Transgene Expression...

Area of Science:

• Cheminformatics
• Materials Informatics
• Computational Chemistry

Background:

• Support Vector Regression (SVR) is a widely used machine learning algorithm in cheminformatics.
• Specialized SVR algorithms like Fuzzy-SVR and Least Square-SVR (LS-SVR) have emerged but lack accessible software.
• A need exists for user-friendly, downloadable tools for advanced SVR applications in scientific research.

Purpose of the Study:

• To develop an accessible online learning system for predictive cheminformatics and materials informatics.
• To provide a platform for researchers to easily compare different modeling approaches using SVR.
• To address the scarcity of public-domain software for advanced SVR algorithms.

Main Methods:

• Development of the Support vector regression-based Online Learning Equipment (SOLE) web tool.
• Implementation of SVR, Fuzzy-SVR, and LS-SVR algorithms within the SOLE framework.
• Application of the SOLE system to model transgene expression efficacy of polymers derived from aminoglycoside antibiotics.

Main Results:

• The SOLE system was successfully used to model transgene expression efficacy.
• Achieved high predictive accuracy with test set r2 values ranging from 0.96 to 0.98.
• Demonstrated model stability and robustness against overfitting through Y-scrambling tests, with test set R2 values between 0.79 and 0.84.

Conclusions:

• SOLE offers a user-friendly interface for conducting Quantitative Structure-Activity Relationship (QSAR) and Quantitative Structure-Property Relationship (QSPR) studies.
• The tool incorporates advanced feature selection, model selection, and model evaluation processes.
• SOLE is applicable across diverse research areas in cheminformatics and materials science.