A Primer on SAnDReS 2.0 for Scoring Function Design

Docking screens rely on several computational techniques to analyze protein-ligand interactions. In this work, we focus on supervised machine learning. We discuss the application of linear regression using the program SAnDReS 2.0 to build a model to predict the inhibition of enzymes. Linear regression is a method to construct a supervised machine learning model based on a training dataset. This model considers parameters that minimize a cost function based on experimental information. The cost function captures the adequacy of the model and indicates how close the predicted values are to the experimental values. Linear regression belongs to a class of methods named parametric models. This simple approach is of general application to docking screens. We discuss its mathematical aspects and implementation using the Scikit-Learn library. We present a simple implementation of linear regression and its application to a toy dataset based on randomly generated data. Also, we discuss an application of this regression algorithm to study the inhibition of cyclin-dependent kinase 2. We developed a linear regression model using the program SAnDReS 2.0 to predict the inhibition of this enzyme. Additionally, we propose end-of-chapter exercises to improve understanding of the concepts discussed here. We made available all the codes discussed here at GitHub: https://github.com/azevedolab/docking#readme .