engleski

Machine learning-based prediction of multi-target antimicrobial activity

Reduced space of multi-target antimicrobial activities was used as a dependent variable for estimation of Cinchona alkaloids derivatives [1] activities. A panel of various Gram- positive and Gram-negative bacteria provided activity data whose principal components were extracted by 2nd-order tensor decomposition. These principal components were regressed on the theoretically computed energy fingerprints of all compounds. Extensive machine learning procedure was applied for generation of multivariate linear regression models with linear combination of original variables as well as their higher- order polynomial terms. Regression models of antimicrobial activity in dependence on molecular dynamics data were builded and thoroughly validated by leave-one- out cross- validation technique (LOO-CV) [2]. The most optimal representation was selected on the basis of R2 values and LOO-CV mean squared error and is presented on Fig. 1.

machine learning ; multivariate linear regression ; principal component analysis ; potential energy surface ; ab initio molecular dynamics ; antimicrobial activity ; Cinchona alkaloids derivatives

nije evidentirano