Chemoinformatic regression methods and their applicability domain.

Area of Science:

• Chemoinformatics
• Computational Chemistry
• Machine Learning

Background:

• Regression models map explanatory variables to continuous outputs, with performance limited by training data.
• Model uncertainty is a growing concern in chemoinformatics, necessitating reliable evaluation methods.
• Outlier detection and defining the applicability domain are crucial for robust regression models.

Purpose of the Study:

• To summarize widely used regression techniques in chemoinformatics.
• To explain methods for estimating the reliability and uncertainty of these models.
• To define the theoretical background and applicability domain for regression techniques.

Main Methods:

• Review of common regression techniques used in chemoinformatics.
• Explanation of built-in and generic procedures for quantifying model uncertainty.
• Discussion of outlier detection methods to enhance model performance.
• Elaboration on defining the applicability domain for regression models.

Main Results:

• Identification of key regression techniques and their uncertainty quantification methods.
• Understanding of how training data and outliers impact model reliability.
• Framework for defining specific and general applicability domains.
• Insights into the theoretical principles behind model uncertainty estimation.

Conclusions:

• Reliable estimation of chemoinformatic model uncertainty is essential for trustworthy predictions.
• Defining the applicability domain enhances the robustness and interpretability of regression models.
• A comprehensive understanding of regression techniques and their limitations is vital for advancing chemoinformatics.