Modeling and predicting chiral stationary phase enantioselectivity: An efficient random forest classifier using an optimally balanced training dataset and an aggregation strategy

Abstract

Predicting whether a chiral column will be effective is a daily task for many analysts. Moreover, finding the best chiral column for separating a particular racemic compound is mostly a matter of trial and error that may take up to a week in some cases. In this study we have developed a novel prediction approach based on combining a random forest classifier and an optimized discretization method for dealing with enantioselectivity as a continuous variable. Using the optimization results, models were trained on data sets divided into four enantioselectivity classes. The best model performances were achieved by over-sampling the minority classes (α ≤ 1.10 and α ≥ 2.00), down-sampling the majority class (1.2 ≤ α < 2.0), and aggregating multicategory predictions into binary classifications. We tested our method on 41 chiral stationary phases using layered fingerprints as descriptors. Experimental results show that this learning methodology was successful in terms of average area under the Receiver Operating Characteristic curve, Kappa indices and F-measure for structure-based prediction of the enantioselective behavior of 34 chiral columns.

Modelling approaches for chiral chromatography on polysaccharide-based and macrocyclic antibiotic chiral selectors: A review

“…More recently, the random forest (RF) method is also used [63,77]. If an RF approach is applied, several decision trees are created.…”

Role of simple descriptors and applicability domain in predicting change in protein thermostability

“…It has previously been shown to be useful in predicting dTm values. [ 30 ] Models were built using a previous algorithm [ 44 ] with the default parameters of generating 100 trees without limiting the number of levels, choosing 1/3 of the descriptors at each branch point, and needing at least five molecules in a node for a branch point to be created. The Gini index was used as the splitting criterion.…”

New mathematical measures for apprehending complexity of chiral molecules using information entropy

“…25 In a recent work, we also found that 2D-fingerprint descriptors were more powerful than 3D-descriptors to build models to find the most promising chiral selector to achieve the separation of a chiral compound. 26 It looks like meaningful 2D information is lost during the construction process of 3Ddescriptors. We thus checked if this effect could be observed between H Ã 2D S chiral ð Þ and H Ã 3D S chiral ð Þ entropy measures.…”