engleski

Evolution of Inhibition Models for Fluorinated Cinchona Alkaloids by Machine Learning

Potential energy surfaces (PES) for 25 fluorinated Cinchona alkaloids derivatives were sampled by ab initio molecular dynamics [1] and used as independent variables in establishing activity/PES multivariate linear regression models (MLR) [2, 3]. Principal components of previously measured inhibitory activities towards human acetyl- and butyrylcholinesterase were used as dependent variables. An extensive machine learning protocol was applied for generating all possible MLR models with linear combinations of original variables as well as their higher-order polynomial terms. Evolution of regression model was monitored by calculation of R2, adjusted and predicted R2. Each regression model was fully validated by leave-one- out cross- validation (LOO-CV) and the best possible activity/PES models for different dimensionalities were selected based on R2 values and the LOO-CV mean squared errors (Fig. 1).

inhibition activity, Cinchona alkaloids, machine learning, principal component analysis, potential energy surfaces, ab initio molecular dynamics, multivariate linear regression

nije evidentirano