engleski

Multi-target antimicrobial activity model of Cinchona alkaloids established by machine learning

Antimicrobial activity of Cinchona alkaloids derivatives [1] was previously evaluated by using disc diffusion assay against a panel of various Gram-positive and Gram-negative bacteria. Principal components of the activity data were extracted by 2nd-order tensor decomposition and used as dependent variables for multivariate linear regression, whereas theoretically computed energy fingerprints of all compounds were used as independent variables. Potential energy surfaces (PES) of compounds were sampled by performing molecular dynamics simulations and then decomposed by principal component analysis. Regression models were generated by extensive machine learning multivariate linear regression – linear combinations of original variables were used as well as their higher-order polynomial terms. Obtained models were thoroughly validated by leave-one-out cross- validation technique (LOO- CV) [2]. The optimal activity/PES model based on the adjusted and the predicted R2 values as well as LOO-CV mean squared error will be presented.

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

nije evidentirano