Computational models for the classification of mPGES-1 inhibitors with fingerprint descriptors

Abstract

Human microsomal prostaglandin [Formula: see text] synthase (mPGES)-1 is a promising drug target for inflammation and other diseases with inflammatory symptoms. In this work, we built classification models which were able to classify mPGES-1 inhibitors into two groups: highly active inhibitors and weakly active inhibitors. A dataset of 1910 mPGES-1 inhibitors was separated into a training set and a test set by two methods, by a Kohonen's self-organizing map or by random selection. The molecules were represented by different types of fingerprint descriptors including MACCS keys (MACCS), CDK fingerprints, Estate fingerprints, PubChem fingerprints, substructure fingerprints and 2D atom pairs fingerprint. First, we used a support vector machine (SVM) to build twelve models with six types of fingerprints and found that MACCS had some advantage over the other fingerprints in modeling. Next, we used naïve Bayes (NB), random forest (RF) and multilayer perceptron (MLP) methods to build six models with MACCS only and found that models using RF and MLP methods were better than NB. Finally, all the models with MACCS keys were used to make predictions on an external test set of 41 compounds. In summary, the models built with MACCS keys and using SVM, RF and MLP methods show good prediction performance on the test sets and the external test set. Furthermore, we made a structure-activity relationship analysis between mPGES-1 and its inhibitors based on the information gain of fingerprints and could pinpoint some key functional groups for mPGES-1 activity. It was found that highly active inhibitors usually contained an amide group, an aromatic ring or a nitrogen heterocyclic ring, and several heteroatoms substituents such as fluorine and chlorine. The carboxyl group and sulfur atom groups mainly appeared in weakly active inhibitors.

Molecular fingerprint and machine learning to accelerate design of high‐performance homochiral metal–organic frameworks

“…Moreover, after the comparison of averages of four evaluation indicators (accuracy, precision, sensitivity, and specificity), it is found that the NCA–MACCS pair is the most suitable to predict the performance of the FHMOFs compared with the other pairs, as shown in Table S9. In previous work, 68 Xia et al have shown that MACCS as the input variables of ML was suitable to predict the activity inhibition properties of materials. Therefore, the combination of NCA and MACCS is selected to evaluate quantitatively the influencing degree (PI, represented by feature weight) of each substructure for the design of top‐10 high‐performance FHMOFs.…”

SAR study on inhibitors of GIIA secreted phospholipase A2 using machine learning methods

“…There were nine common descriptors used in Models 1A and 1B simultaneously, as shown in Table . The Pearson correlation coefficient (PCC) between the descriptors and the activity levels, and the ranking orders of the descriptors in recursive feature elimination (RFE) method (Isabelle Guyon et al., ; Lei, Chen, et al., ; Lei, Sun, et al., ) and the information gain (IG) method (Sokolova & Szpakowicz, ; Xia & Yan, ) are listed in Table (Bauerschmidt & Gasteiger, ; Gasteiger & Hutchings, ; Gasteiger & Marsili, ; Lipinski, Lombardo, Dominy, & Feeney, ; R. Wang, Gao, & Lai, ) as well. These descriptors are all very significant to our classification models, especially 3DACorr:Polariz:Cor3D:ori1_3, whose ranking order was the topest either in RFE or in IG.…”

“…We built models with the input CORINA descriptors from 1 to 30, which were added one by one gradually, according to the ranked orders obtained from the recursive feature elimination (RFE) method (Guyon et al., ; Lei, Chen et al., ; Lei, Sun et al., ) or the information gain (IG) method. (Sokolova & Szpakowicz, ; Xia & Yan, ) We also built models with 88 MACCS fingerprints and with 394 ECFP4 fingerprints. The models were generated based on training set.…”

“…The second method for the ranking was information gain (IG) derived from information entropy theory. (Sokolova & Szpakowicz, ; Xia & Yan, ) The IG values were calculated by the program Weka v3.8.1 . The ranking orders of the 30 CORINA descriptors in both RFE and IG are presented in Table S1 in Supporting information.…”

“…CORINA descriptors that we used are single values (global molecular descriptors) which represent structural, chemical, or physicochemical feature or property of the molecule, and vectorial molecular descriptors (3D property-weighted autocorrelation descriptors) are derived from 3D structure of a molecule and atom pair properties. (Wagener et al, 1995) In order to investigate the 30 CORINA descriptors fully and find the best descriptors sets for modeling, the 30 descriptors were ranked with two methods: recursive feature elimination (RFE) (Isabelle Guyon et al, 2002;Lei, Chen et al, 2017; and information gain (IG; Sokolova & Szpakowicz, 2010;Xia & Yan, 2017) Based on either list of ranked descriptors, we built 30 models based on training set 1 when the descriptors were added one by one from 1 to 30. Following the process above, we applied three machine learning algorithms including support vector machine (SVM), (Cortes & Vapnik, 1995;Guyon et al, 1993;Sun et al, 2016;Wang et al, 2016) decision tree (DT), (Joshi et al, 2016) and random forest (RF), (Breiman, 2001;Lei et al, 2016) respectively.…”