A comprehensive comparative assessment of 3D molecular similarity tools in ligand-based virtual screening.

Three-dimensional (3D) molecular similarity, one major ligand-based virtual screening (VS) method, has been widely used in the drug discovery process. A variety of 3D molecular similarity tools have been developed in recent decades. In this study, we assessed a panel of 15 3D molecular similarity programs against the DUD-E and LIT-PCBA datasets, including commercial ROCS and Phase, in terms of screening power and scaffold-hopping power. The results revealed that (1) SHAFTS, LS-align, Phase Shape_Pharm and LIGSIFT showed the best VS capability in terms of screening power. Some 3D similarity tools available to academia can yield relatively better VS performance than commercial ROCS and Phase software. (2) Current 3D similarity VS tools exhibit a considerable ability to capture actives with new chemotypes in terms of scaffold hopping. (3) Multiple conformers relative to single conformations will generally improve VS performance for most 3D similarity tools, with marginal improvement observed in area under the receiving operator characteristic curve values, enrichment factor in the top 1% and hit rate in the top 1% values showed larger improvement. Moreover, redundancy and complementarity analyses of hit lists from different query seeds and different 3D similarity VS tools showed that the combination of different query seeds and/or different 3D similarity tools in VS campaigns retrieved more (and more diverse) active molecules. These findings provide useful information for guiding choices of the optimal 3D molecular similarity tools for VS practices and designing possible combination strategies to discover more diverse active compounds.

Prediction and Structure-Activity Relationship Analysis on Ready Biodegradability of Chemical Using Machine Learning Method.

Persistent contaminants from different industries have already caused significant risks to the environment and public health. In this study, a data set containing 1306 not readily biodegradable (NRB) and 622 readily biodegradable (RB) chemicals was collected and characterized by CORINA descriptors, MACCS fingerprints, and ECFP_4 fingerprints. We utilized decision tree (DT), support vector machine (SVM), random forest (RF), and deep neural network (DNN) to construct 34 classification models that could predict the biodegradability of compounds. The best model (model 5F) built using a Transformer-CNN algorithm had a balanced accuracy of 86.29% and a Matthews correlation coefficient of 0.71 on the test set. By analyzing the top 10 CORINA descriptors used for modeling, the properties containing solubility, π/σ atom charges, rotatable bonds number, lone pair/π/σ atom electronegativities, molecular weight, and number of nitrogen atom based hydrogen bonding acceptors were determined to be critical for biodegradability. The substructure investigations confirmed earlier studies that the presence of aromatic rings and nitrogen or halogen substitutions in a molecule will hinder the biodegradation of the compound, while the ester groups and carboxyl groups promote biodegradability. We also identified the representative fragments affecting biodegradability by analyzing the frequency differences of substructural fragments between the NRB and RB compounds. The results of the study can provide excellent guidance for the discovery and design of compounds with good chemical biodegradability.

Classification models and SAR analysis on HDAC1 inhibitors using machine learning methods.

Histone deacetylase (HDAC) 1, a member of the histone deacetylases family, plays a pivotal role in various tumors. In this study, we collected 7313 human HDAC1 inhibitors with bioactivities to form a dataset. Then, the dataset was divided into a training set and a test set using two splitting methods: (1) Kohonen's self-organizing map and (2) random splitting. The molecular structures were represented by MACCS fingerprints, RDKit fingerprints, topological torsions fingerprints and ECFP4 fingerprints. A total of 80 classification models were built by using five machine learning methods, including decision tree (DT), random forest, support vector machine, eXtreme Gradient Boosting and deep neural network. Model 15A_2 built by the XGBoost algorithm based on ECFP4 fingerprints showed the best performance, with an accuracy of 88.08% and an MCC value of 0.76 on the test set. Finally, we clustered the 7313 HDAC1 inhibitors into 31 subsets, and the substructural features in each subset were investigated. Moreover, using DT algorithm we analyzed the structure-activity relationship of HDAC1 inhibitors. It may conclude that some substructures have a significant effect on high activity, such as N-(2-amino-phenyl)-benzamide, benzimidazole, AR-42 analogues, hydroxamic acid with a middle chain alkyl and 4-aryl imidazole with a midchain of alkyl whose α carbon is chiral.

Classification of FLT3 inhibitors and SAR analysis by machine learning methods.

FMS-like tyrosine kinase 3 (FLT3) is a type III receptor tyrosine kinase, which is an important target for anti-cancer therapy. In this work, we conducted a structure-activity relationship (SAR) study on 3867 FLT3 inhibitors we collected. MACCS fingerprints, ECFP4 fingerprints, and TT fingerprints were used to represent the inhibitors in the dataset. A total of 36 classification models were built based on support vector machine (SVM), random forest (RF), eXtreme Gradient Boosting (XGBoost), and deep neural networks (DNN) algorithms. Model 3D_3 built by deep neural networks (DNN) and TT fingerprints performed best on the test set with the highest prediction accuracy of 85.83% and Matthews correlation coefficient (MCC) of 0.72 and also performed well on the external test set. In addition, we clustered 3867 inhibitors into 11 subsets by the K-Means algorithm to figure out the structural characteristics of the reported FLT3 inhibitors. Finally, we analyzed the SAR of FLT3 inhibitors by RF algorithm based on ECFP4 fingerprints. The results showed that 2-aminopyrimidine, 1-ethylpiperidine,2,4-bis(methylamino)pyrimidine, amino-aromatic heterocycle, [(2E)-but-2-enyl]dimethylamine, but-2-enyl, and alkynyl were typical fragments among highly active inhibitors. Besides, three scaffolds in Subset_A (Subset 4), Subset_B, and Subset_C showed a significant relationship to inhibition activity targeting FLT3.

Machine learning-based classification models for non-covalent Bruton's tyrosine kinase inhibitors: predictive ability and interpretability.

In this study, we built classification models using machine learning techniques to predict the bioactivity of non-covalent inhibitors of Bruton's tyrosine kinase (BTK) and to provide interpretable and transparent explanations for these predictions. To achieve this, we gathered data on BTK inhibitors from the Reaxys and ChEMBL databases, removing compounds with covalent bonds and duplicates to obtain a dataset of 3895 inhibitors of non-covalent. These inhibitors were characterized using MACCS fingerprints and Morgan fingerprints, and four traditional machine learning algorithms (decision trees (DT), random forests (RF), support vector machines (SVM), and extreme gradient boosting (XGBoost)) were used to build 16 classification models. In addition, four deep learning models were developed using deep neural networks (DNN). The best model, Model D_4, which was built using XGBoost and MACCS fingerprints, achieved an accuracy of 94.1% and a Matthews correlation coefficient (MCC) of 0.75 on the test set. To provide interpretable explanations, we employed the SHAP method to decompose the predicted values into the contributions of each feature. We also used K-means dimensionality reduction and hierarchical clustering to visualize the clustering effects of molecular structures of the inhibitors. The results of this study were validated using crystal structures, and we found that the interaction between the BTK amino acid residue and the important features of clustered scaffold was consistent with the known properties of the complex crystal structures. Overall, our models demonstrated high predictive ability and a qualitative model can be converted to a quantitative model to some extent by SHAP, making them valuable for guiding the design of new BTK inhibitors with desired activity.

Discovering the Active Ingredients of Medicine and Food Homologous Substances for Inhibiting the Cyclooxygenase-2 Metabolic Pathway by Machine Learning Algorithms.

Cyclooxygenase-2 (COX-2) and microsomal prostaglandin E2 synthase (mPGES-1) are two key targets in anti-inflammatory therapy. Medicine and food homology (MFH) substances have both edible and medicinal properties, providing a valuable resource for the development of novel, safe, and efficient COX-2 and mPGES-1 inhibitors. In this study, we collected active ingredients from 503 MFH substances and constructed the first comprehensive MFH database containing 27,319 molecules. Subsequently, we performed Murcko scaffold analysis and K-means clustering to deeply analyze the composition of the constructed database and evaluate its structural diversity. Furthermore, we employed four supervised machine learning algorithms, including support vector machine (SVM), random forest (RF), deep neural networks (DNNs), and eXtreme Gradient Boosting (XGBoost), as well as ensemble learning, to establish 640 classification models and 160 regression models for COX-2 and mPGES-1 inhibitors. Among them, ModelA_ensemble_RF_1 emerged as the optimal classification model for COX-2 inhibitors, achieving predicted Matthews correlation coefficient (MCC) values of 0.802 and 0.603 on the test set and external validation set, respectively. ModelC_RDKIT_SVM_2 was identified as the best regression model based on COX-2 inhibitors, with root mean squared error (RMSE) values of 0.419 and 0.513 on the test set and external validation set, respectively. ModelD_ECFP_SVM_4 stood out as the top classification model for mPGES-1 inhibitors, attaining MCC values of 0.832 and 0.584 on the test set and external validation set, respectively. The optimal regression model for mPGES-1 inhibitors, ModelF_3D_SVM_1, exhibited predictive RMSE values of 0.253 and 0.35 on the test set and external validation set, respectively. Finally, we proposed a ligand-based cascade virtual screening strategy, which integrated the well-performing supervised machine learning models with unsupervised learning: the self-organized map (SOM) and molecular scaffold analysis. Using this virtual screening workflow, we discovered 10 potential COX-2 inhibitors and 15 potential mPGES-1 inhibitors from the MFH database. We further verified candidates by molecular docking, investigated the interaction of the candidate molecules upon binding to COX-2 or mPGES-1. The constructed comprehensive MFH database has laid a solid foundation for the further research and utilization of the MFH substances. The series of well-performing machine learning models can be employed to predict the COX-2 and mPGES-1 inhibitory capabilities of unknown compounds, thereby aiding in the discovery of anti-inflammatory medications. The COX-2 and mPGES-1 potential inhibitor molecules identified through the cascade virtual screening approach provide insights and references for the design of highly effective and safe novel anti-inflammatory drugs.

Discovery of Novel Epidermal Growth Factor Receptor (EGFR) Inhibitors Using Computational Approaches.

The epidermal growth factor receptor (EGFR) signaling pathway plays an important role in cell growth, proliferation, differentiation, and other physiological processes, which makes the EGFR a promising target for anticancer therapies. The discovery of novel EGFR inhibitors may provide a solution to the problem of drug resistance. In this work, we performed a ligand-based virtual screening (LBVS) protocol for finding novel EGFR inhibitors from a 5.3 million compound library. First, the 3D shape-based similarity was used to obtain structurally novel EGFR inhibitors. In this study, we tried three queries; two were crystal structures and one was generated from deep generative models of graphs (DGMG). Next, we have built four structure-activity relationship (SAR) models and three quantitative structure-activity relationship (QSAR) models based on an SVM method for further screening of highly active EGFR inhibitors. Experimental validations led to the identification of nine hits out of 18 tested compounds. Among them, hit 1, hit 5, and hit 6 had IC50 values around 80 nM against EGFR whose interactions with EGFR were further investigated by molecular dynamics simulations.

Building 2D classification models and 3D CoMSIA models on small-molecule inhibitors of both wild-type and T790M/L858R double-mutant EGFR.

Epidermal growth factor receptor (EGFR) has received widespread attention because it is an important target for anticancer drug design. Mutations in the EGFR, especially the T790M/L858R double mutation, have made cancer treatment more difficult. We herein built the structure-activity relationship models of small-molecule inhibitors on wild-type and T790M/L858R double-mutant EGFR with a whole dataset of 379 compounds. For 2D classification models, we used ECFP4 fingerprints to build support vector machine and random forest models and used SMILES to build self-attention recurrent neural network models. Each of all six models resulted in an accuracy of above 0.87 and the Matthews correlation coefficient value of above 0.76 on the test set, respectively. We concluded that inhibitors containing anilinoquinoline and methoxy or fluoro phenyl are highly active against wild EGFR. Substructures such as anilinopyrimidine, acrylamide, amino phenyl, methoxy phenyl, and thienopyrimidinyl amide appeared more in highly active inhibitors against double-mutant EGFR. We also used self-organizing map to cluster the inhibitors into six subsets based on ECFP4 fingerprints and analyzed the activity characteristics of different scaffolds in each subset. Among them, three datasets, which are based on pteridin, anilinopyrimidine, and anilinoquinoline scaffold, were selected to build 3D comparative molecular similarity analysis models individually. Models with the leave-one-out coefficient of determination (q2) above 0.65 were selected, and five descriptor types (steric, electrostatic, hydrophobic, donor, and acceptor) were used to study the effects of side chains of inhibitors on the activity against wild-type and mutant-type EGFR.

Classification models and SAR analysis on CysLT1 receptor antagonists using machine learning algorithms.

Cysteinyl leukotrienes 1 (CysLT1) receptor is a promising drug target for rhinitis or other allergic diseases. In our study, we built classification models to predict bioactivities of CysLT1 receptor antagonists. We built a dataset with 503 CysLT1 receptor antagonists which were divided into two groups: highly active molecules (IC50 < 1000 nM) and weakly active molecules (IC50 ≥ 1000 nM). The molecules were characterized by several descriptors including CORINA descriptors, MACCS fingerprints, Morgan fingerprint and molecular SMILES. For CORINA descriptors and two types of fingerprints, we used the random forests (RF) and deep neural networks (DNN) to build models. For molecular SMILES, we used recurrent neural networks (RNN) with the self-attention to build models. The accuracies of test sets for all models reached 85%, and the accuracy of the best model (Model 2C) was 93%. In addition, we made structure-activity relationship (SAR) analyses on CysLT1 receptor antagonists, which were based on the output from the random forest models and RNN model. It was found that highly active antagonists usually contained the common substructures such as tetrazoles, indoles and quinolines. These substructures may improve the bioactivity of the CysLT1 receptor antagonists.

Classification of Cyclooxygenase-2 Inhibitors Using Support Vector Machine and Random Forest Methods.

This work reports the classification study conducted on the biggest COX-2 inhibitor data set so far. Using 2925 diverse COX-2 inhibitors collected from 168 pieces of literature, we applied machine learning methods, support vector machine (SVM) and random forest (RF), to develop 12 classification models. The best SVM and RF models resulted in MCC values of 0.73 and 0.72, respectively. The 2925 COX-2 inhibitors were reduced to a data set of 1630 molecules by removing intermediately active inhibitors, and 12 new classification models were constructed, yielding MCC values above 0.72. The best MCC value of the external test set was predicted to be 0.68 by the RF model using ECFP_4 fingerprints. Moreover, the 2925 COX-2 inhibitors were clustered into eight subsets, and the structural features of each subset were investigated. We identified substructures important for activity including halogen, carboxyl, sulfonamide, and methanesulfonyl groups, as well as the aromatic nitrogen atoms. The models developed in this study could serve as useful tools for compound screening prior to lab tests.

Identification of Novel Aurora Kinase A (AURKA) Inhibitors via Hierarchical Ligand-Based Virtual Screening.

Aurora kinases are essential for cell mitosis, amplified, and overexpressed in various human malignancies. Therefore, Aurora kinases have been promising targets for anticancer therapies, which has prompted an intensive search for their small-molecule inhibitors. In this work, we performed a hierarchical and time-efficient virtual screening cascade for scaffold hopping, aiming to obtain structurally novel and highly potent hit compounds targeting Aurora kinases. The cascade consisted of a shape- and an electrostatic-based protocol, combined with a QSAR-based selection protocol. This virtual screening cascade was used to screen two databases, one commercial database named the J&K database containing about 5.2 million diverse molecules and the Drugbank database. Experimental validations led to the identification of one structurally novel and highly potent hit compound (hit 1, found to possess an IC50 of 8.1 and 19 nM for Aurora kinases A and B, respectively), which can be a promising starting point for further exploration. Additionally, Aurora kinases were identified as off-targets for hits 2-6 (Crizotinib, CI-1033, Dasatinib, Bosutinib, MLN-518), which are approved or investigational drugs as listed in Drugbank, plausibly suggesting targeting Aurora kinases may even contribute to their mechanism of action.

QSAR studies of the bioactivity of hepatitis C virus (HCV) NS3/4A protease inhibitors by multiple linear regression (MLR) and support vector machine (SVM).

In this study, quantitative structure-activity relationship (QSAR) models using various descriptor sets and training/test set selection methods were explored to predict the bioactivity of hepatitis C virus (HCV) NS3/4A protease inhibitors by using a multiple linear regression (MLR) and a support vector machine (SVM) method. 512 HCV NS3/4A protease inhibitors and their IC50 values which were determined by the same FRET assay were collected from the reported literature to build a dataset. All the inhibitors were represented with selected nine global and 12 2D property-weighted autocorrelation descriptors calculated from the program CORINA Symphony. The dataset was divided into a training set and a test set by a random and a Kohonen's self-organizing map (SOM) method. The correlation coefficients (r2) of training sets and test sets were 0.75 and 0.72 for the best MLR model, 0.87 and 0.85 for the best SVM model, respectively. In addition, a series of sub-dataset models were also developed. The performances of all the best sub-dataset models were better than those of the whole dataset models. We believe that the combination of the best sub- and whole dataset SVM models can be used as reliable lead designing tools for new NS3/4A protease inhibitors scaffolds in a drug discovery pipeline.

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

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.

Computer modeling in predicting the bioactivity of human 5-lipoxygenase inhibitors.

5-Lipoxygenase (5-LOX) is a key enzyme in the inflammatory path. Inhibitors of 5-LOX are useful for the treatment of diseases like arthritis, cancer, and asthma. We have collected a dataset including 220 human 5-LOX inhibitors for classification. A self-organizing map (SOM), a support vector machine (SVM), and a multilayer perceptron (MLP) algorithm were used to build models with selected descriptors for classifying 5-LOX inhibitors into active and weakly active ones. MACCS fingerprints were used in this model building process. The accuracy (Q) and Matthews correlation coefficient (MCC) of the best SOM model (Model 1A) were 86.49% and 0.73 on the test set, respectively. The Q and MCC of the best SVM model (Model 2A) were 82.67% and 0.64 on the test set, respectively. The Q and MCC of the best MLP model (Model 3B) were 84.00% and 0.67 on the test set, respectively. In addition, 180 inhibitors with bioactivities measured by fluorescence method were further used for a quantitative prediction. Multiple linear regression (MLR) and SVM algorithms were used to build models to predict the [Formula: see text] values. The correlation coefficients (R) of the MLR model (Model Q1) and the SVM model (Model Q2) were 0.72 and 0.74 on the test set, respectively.

Ligand and structure based hierarchical virtual screening cascade for finding novel epidermal growth factor receptor inhibitors.

The epidermal growth factor receptor (EGFR) tyrosine kinase plays an important role in tumor formation and growth by mediating cell growth and other physiological processes. Therefore, EGFR is a promising target for the treatment of cancer. In this work, we combined ligand-based and structure-based virtual screening methods to identify novel EGFR inhibitors from a library of more than 103 thousand compounds. We first obtained hundreds of compounds with similar physiochemical properties through 3D molecular shape and electrostatic similarity screening with potent inhibitors AEE788 and Afatinib as queries. Next, we identified compounds with strong binding affinities to the EGFR pocket through molecular docking, which makes good use of the structure information of the receptor. After molecular scaffold analysis, our bioassay confirmed 13 compounds with EGFR inhibitory activity and three compounds had IC50 values below 1000 nM. In addition, we collected 5371 EGFR inhibitors from online databases, and clustered them into 7 groups by K-means method using their ECFP4 fingerprints as input. Each cluster had typical molecular fragments and corresponding activity characteristics, which could guide the design of EGFR inhibitors, and we concluded that the fragments from some of the hits are indicated in the highly active scaffolds.

Carcinogenicity prediction of noncongeneric chemicals by a support vector machine.

The ability to identify carcinogenic compounds is of fundamental importance to the safe application of chemicals. In this study, we generated an array of in silico models allowing the classification of compounds into carcinogenic and noncarcinogenic agents based on a data set of 852 noncongeneric chemicals collected from the Carcinogenic Potency Database (CPDBAS). Twenty-four molecular descriptors were selected by Pearson correlation, F-score, and stepwise regression analysis. These descriptors cover a range of physicochemical properties, including electrophilicity, geometry, molecular weight, size, and solubility. The descriptor mutagenic showed the highest correlation coefficient with carcinogenicity. On the basis of these descriptors, a support vector machine-based (SVM) classification model was developed and fine-tuned by a 10-fold cross-validation approach. Both the SVM model (Model A1) and the best model from the 10-fold cross-validation (Model B3) runs gave good results on the test set with prediction accuracy over 80%, sensitivity over 76%, and specificity over 82%. In addition, extended connectivity fingerprints (ECFPs) and the Toxtree software were used to analyze the functional groups and substructures linked to carcinogenicity. It was found that the results of both methods are in good agreement.

Prediction of bioactivity of HIV-1 integrase ST inhibitors by multilinear regression analysis and support vector machine.

In this study, four computational quantitative structure-activity relationship models were built to predict the biological activity of HIV-1 integrase strand transfer (ST) inhibitors. 551 Inhibitors whose bioactivities were detected by radiolabeling method were collected. The molecules were represented with 20 selected MOE descriptors. All inhibitors were divided into a training set and a test set with two methods: (1) by a Kohonen's self-organizing map (SOM); (2) by a random selection. For every training set and test set, a multilinear regression (MLR) analysis and a support vector machine (SVM) were used to establish models, respectively. For the test set divided by SOM, the correlation coefficients (rs) were over 0.91, and for the test set split randomly, the rs were over 0.86.

Prediction of bioactivity of ACAT2 inhibitors by multilinear regression analysis and support vector machine.

Two quantitative structure-activity relationships (QSAR) models for predicting 95 compounds inhibiting Acyl-coenzyme A: cholesterol acyltransferase2 (ACAT2) were developed. The whole data set was randomly split into a training set including 72 compounds and a test set including 23 compounds. The molecules were represented by 11 descriptors calculated by software ADRIANA.Code. Then the inhibitory activity of ACAT2 inhibitors was predicted using multilinear regression (MLR) analysis and support vector machine (SVM) method, respectively. The correlation coefficients of the models for the test sets were 0.90 for MLR model, and 0.91 for SVM model. Y-randomization was employed to ensure the robustness of the SVM model. The atom charge and electronegativity related descriptors were important for the interaction between the inhibitors and ACAT2.

Classification of Plasmodium falciparum glucose-6-phosphate dehydrogenase inhibitors by support vector machine.

Plasmodium falciparum glucose-6-phosphate dehydrogenase (PfG6PD) has been considered as a potential target for severe forms of anti-malaria therapy. In this study, several classification models were built to distinguish active and weakly active PfG6PD inhibitors by support vector machine method. Each molecule was initially represented by 1,044 molecular descriptors calculated by ADRIANA.Code. Correlation analysis and attribute selection methods in Weka were used to get the best reduced set of molecular descriptors, respectively. The best model (Model 2w) gave a prediction accuracy (Q) of 93.88 % and a Matthew's correlation coefficient (MCC) of 0.88 on the test set. Some properties such as [Formula: see text] atom charge, [Formula: see text] atom charge, and lone pair electronegativity-related descriptors are important for the interaction between the PfG6PD and the inhibitor.

Discriminating of ATP competitive Src kinase inhibitors and decoys using self-organizing map and support vector machine.

A data set containing 686 Src kinase inhibitors and 1,941 Src kinase non-binding decoys was collected and used to build two classification models to distinguish inhibitors from decoys. The data set was randomly split into a training set (458 inhibitors and 972 decoys) and a test set (228 inhibitors and 969 decoys). Each molecule was represented by five global molecular descriptors and 18 2D property autocorrelation descriptors calculated using the program ADRIANA.Code. Two machine learning methods, a Kohonen's self-organizing map (SOM) and a support vector machine (SVM), were utilized for the training and classification. For the test set, classification accuracy (ACC) of 99.92% and Matthews correlation coefficient (MCC) of 0.98 were achieved for the SOM model; ACC of 99.33% and MCC of 0.98 were obtained for the SVM model. Some molecular properties, such as molecular weight, number of atoms in a molecule, hydrogen bond properties, polarizabilities, electronegativities, and hydrophobicities, were found to be important for the inhibition of Src kinase.