Quantitative Structure-activity Relationship Data Sets Research Articles

ProteoDockNet: Novel GNN-based ligand binding affinities prediction architecture via SMILES to key liver, kidney and brain proteins using QSAR data

Assessing the affinity between specific human proteins and drugs through docking scores is a crucial aspect of structure-based drug design. Traditionally, Quantitative Structure-Activity Relationship (QSAR) models have been utilized, but they fail to capture the complex interactions within protein-ligand complexes. This research explores the burgeoning potential of Graph Neural Networks (GNNs) to revolutionize this field, by employing QSAR models as original datasets for GNN architecture development to predict the molecular properties using simple SMILES notation of small molecules. Having selected significant proteins from three major organs – liver, kidney and brain, these were docked against MolPort database, and designed QSAR models using multiple linear regression and partial least squares via Schrodinger. The core dataset employed in this research comprises molecular information denoted by MolPort IDs alongside corresponding docking scores, essential for evaluating molecular interactions, from which the initial QSAR dataset was employed. A key focus of this study lies in data preprocessing, where each molecule, initially represented as a SMILES string, undergoes conversion into a graph structure. The GNN architecture was built with graphical convolutional network (GCN) and hybrid GCN (with attention mechanisms) to improve accuracy and to allow users to choose their algorithm to be run during protein-ligand binding prediction. All 60 generated models (10 proteins- GCN + GAT models for each), were assessed via RMSE and MAE metrics, before being deployed on Streamlit (https://dockingscoreappapp-vypudgobchyjwk4n3w2fy5.streamlit.app/). The outcomes show an immediate prediction of the docking score of any given SMILES, against the set of proteins trained in ProteoDockNet.

Development and Evaluation of Conformal Prediction Methods for Quantitative Structure-Activity Relationship.

The quantitative structure-activity relationship (QSAR) regression model is a commonly used technique for predicting the biological activities of compounds using their molecular descriptors. Besides accurate activity estimation, obtaining a prediction uncertainty metric like a prediction interval is highly desirable. Quantifying prediction uncertainty is an active research area in statistical and machine learning (ML), but the implementation for QSAR remains challenging. However, most ML algorithms with high predictive performance require add-on companions for estimating the uncertainty of their prediction. Conformal prediction (CP) is a promising approach as its main components are agnostic to the prediction modes, and it produces valid prediction intervals under weak assumptions on the data distribution. We proposed computationally efficient CP algorithms tailored to the most widely used ML models, including random forests, deep neural networks, and gradient boosting. The algorithms use a novel approach to the derivation of nonconformity scores from the estimates of prediction uncertainty generated by the ensembles of point predictions. The validity and efficiency of proposed algorithms are demonstrated on a diverse collection of QSAR data sets as well as simulation studies. The provided software implementing our algorithms can be used as stand-alone or easily incorporated into other ML software packages for QSAR modeling.

Collaborative analysis for drug discovery by federated learning on non-IID data.

With the increasing availability of large-scale QSAR (Quantitative Structure-Activity Relationship) datasets, collaborative analysis has become a promising approach for drug discovery. Traditional centralized analysis which typically concentrates data on a central server for training faces challenges such as data privacy and security. Distributed analysis such as federated learning offers a solution by enabling collaborative model training without sharing raw data. However, it may fail when the training data in the local devices are non-independent and identically distributed (non-IID). In this paper, we propose a novel framework for collaborative drug discovery using federated learning on non-IID datasets. We address the difficulty of training on non-IID data by globally sharing a small subset of data among all institutions. Our framework allows multiple institutions to jointly train a robust predictive model while preserving the privacy of their individual data. We leverage the federated learning paradigm to distribute the model training process across local devices, eliminating the need for data exchange. The experimental results on 15 benchmark datasets demonstrate that the proposed method achieves competitive predictive accuracy to centralized analysis while respecting data privacy. Moreover, our framework offers benefits such as reduced data transmission and enhanced scalability, making it suitable for large-scale collaborative drug discovery efforts.

Parsimonious Optimization of Multitask Neural Network Hyperparameters.

Neural networks are rapidly gaining popularity in chemical modeling and Quantitative Structure–Activity Relationship (QSAR) thanks to their ability to handle multitask problems. However, outcomes of neural networks depend on the tuning of several hyperparameters, whose small variations can often strongly affect their performance. Hence, optimization is a fundamental step in training neural networks although, in many cases, it can be very expensive from a computational point of view. In this study, we compared four of the most widely used approaches for tuning hyperparameters, namely, grid search, random search, tree-structured Parzen estimator, and genetic algorithms on three multitask QSAR datasets. We mainly focused on parsimonious optimization and thus not only on the performance of neural networks, but also the computational time that was taken into account. Furthermore, since the optimization approaches do not directly provide information about the influence of hyperparameters, we applied experimental design strategies to determine their effects on the neural network performance. We found that genetic algorithms, tree-structured Parzen estimator, and random search require on average 0.08% of the hours required by grid search; in addition, tree-structured Parzen estimator and genetic algorithms provide better results than random search.

Reliable Model Selection without Reference Values by Utilizing Model Diversity with Prediction Similarity.

Predictive modeling (calibration or training) with various data formats, such as near-infrared (NIR) spectra and quantitative structure-activity relationship (QSAR) data, provides essential information if a proper model is selected. Similarly, with a general model selection approach, spectral model maintenance (updating) from original modeling conditions to new conditions can be performed for dynamic modeling. Fundamental modeling (partial least-squares (PLS) and others) and maintenance processes (domain adaptation or transfer learning and others) require selection of tuning parameter(s) values to isolate models that can accurately predict new samples or molecules, e.g., number of PLS latent variables to predict analyte concentration. Regardless of the modeling task, model selection is complex and without a reliable protocol. Tuning parameter selection typically depends on only one model quality measure assessing model bias using prediction accuracy. Developed in this paper is a generic model selection process using concepts from consensus modeling and QSAR activity landscapes. It is a consensus filtering approach that prioritizes model diversity (MD) while conserving prediction similarity (PS) fused with a common bias-variance trade-off measure. A significant feature of MDPS is that a cross-validation scheme is not needed because models are selected relative to predicting new samples or molecules, i.e., model selection uses unlabeled samples (without reference values) for active predictions. The versatility and reliability of MDPS model selection is shown using four NIR data sets and a QSAR data set. The study also substantiates the Rashomon effect where there is not one best model tuning parameter value that provides accurate predictions.

An Improved Artificial Bee Colony for Feature Selection in QSAR

Quantitative Structure–Activity Relationship (QSAR) aims to correlate molecular structure properties with corresponding bioactivity. Chance correlations and multicollinearity are two major problems often encountered when generating QSAR models. Feature selection can significantly improve the accuracy and interpretability of QSAR by removing redundant or irrelevant molecular descriptors. An artificial bee colony algorithm (ABC) that mimics the foraging behaviors of honey bee colony was originally proposed for continuous optimization problems. It has been applied to feature selection for classification but seldom for regression analysis and prediction. In this paper, a binary ABC algorithm is used to select features (molecular descriptors) in QSAR. Furthermore, we propose an improved ABC-based algorithm for feature selection in QSAR, namely ABC-PLS-1. Crossover and mutation operators are introduced to employed bee and onlooker bee phase to modify several dimensions of each solution, which not only saves the process of converting continuous values into discrete values, but also reduces the computational resources. In addition, a novel greedy selection strategy which selects the feature subsets with higher accuracy and fewer features helps the algorithm to converge fast. Three QSAR datasets are used for the evaluation of the proposed algorithm. Experimental results show that ABC-PLS-1 outperforms PSO-PLS, WS-PSO-PLS, and BFDE-PLS in accuracy, root mean square error, and the number of selected features. Moreover, we also study whether to implement scout bee phase when tracking regression problems and drawing such an interesting conclusion that the scout bee phase is redundant when dealing with the feature selection in low-dimensional and medium-dimensional regression problems.

Auto-KPCA: A Two-Step Hybrid Feature Extraction Technique for Quantitative Structure–Activity Relationship Modeling

Quantitative structure-activity relationship (QSAR) modeling is an established approach for drug discovery, but many QSAR datasets suffer from the curse of dimensionality, a challenge that is usually addressed by using dimensionality reduction techniques such as principal component analysis (PCA). However, although linear feature extraction techniques have low computational cost and can handle linear relationships between descriptors, they cannot handle the complex structures found in QSAR data. Hybridization of feature extraction techniques is an effective approach to address the challenges of high-dimensional datasets, and combining the benefits of at least two dimensionality reduction techniques has been successful in many fields. This paper proposes Auto-KPCA, a two-step hybrid feature extraction technique that leverages (i) the fast computational capability of kernel PCA (KPCA) and (ii) the performance of a deep generalized autoencoder in handling complex data structures. Based on classification accuracy, the proposed approach is compared to other feature extraction techniques on the same benchmark dataset. The capability of Auto-KPCA is then investigated further by testing four deep-learning classification models, namely a convolutional neural network, a recurrent neural network, a feedforward deep neural network, and long short-term memory. To the best of the authors' knowledge, this study is the first to investigate hybridization of KPCA and a deep generalized autoencoder in the context of QSAR. The reported results (i) provide invaluable insights regarding the behavior of different techniques in predicting class labels and (ii) demonstrate increased classification accuracy and noticeably decreased mean square error when compared with KPCA and autoencoders.

Do we need different machine learning algorithms for QSAR modeling? A comprehensive assessment of 16 machine learning algorithms on 14 QSAR data sets.

Although a wide variety of machine learning (ML) algorithms have been utilized to learn quantitative structure-activity relationships (QSARs), there is no agreed single best algorithm for QSAR learning. Therefore, a comprehensive understanding of the performance characteristics of popular ML algorithms used in QSAR learning is highly desirable. In this study, five linear algorithms [linear function Gaussian process regression (linear-GPR), linear function support vector machine (linear-SVM), partial least squares regression (PLSR), multiple linear regression (MLR) and principal component regression (PCR)], three analogizers [radial basis function support vector machine (rbf-SVM), K-nearest neighbor (KNN) and radial basis function Gaussian process regression (rbf-GPR)], six symbolists [extreme gradient boosting (XGBoost), Cubist, random forest (RF), multiple adaptive regression splines (MARS), gradient boosting machine (GBM), and classification and regression tree (CART)] and two connectionists [principal component analysis artificial neural network (pca-ANN) and deep neural network (DNN)] were employed to learn the regression-based QSAR models for 14 public data sets comprising nine physicochemical properties and five toxicity endpoints. The results show that rbf-SVM, rbf-GPR, XGBoost and DNN generally illustrate better performances than the other algorithms. The overall performances of different algorithms can be ranked from the best to the worst as follows: rbf-SVM > XGBoost > rbf-GPR > Cubist > GBM > DNN > RF > pca-ANN > MARS > linear-GPR ≈ KNN > linear-SVM ≈ PLSR > CART ≈ PCR ≈ MLR. In terms of prediction accuracy and computational efficiency, SVM and XGBoost are recommended to the regression learning for small data sets, and XGBoost is an excellent choice for large data sets. We then investigated the performances of the ensemble models by integrating the predictions of multiple ML algorithms. The results illustrate that the ensembles of two or three algorithms in different categories can indeed improve the predictions of the best individual ML algorithms.

Predicting molecular activity on nuclear receptors by multitask neural networks

AbstractThe interest in multitask and deep learning strategies has been increasing in the last few years, in application to large and complex dataset for quantitative structure‐activity relationship (QSAR) analysis. Multitask approaches allow the simultaneous prediction of molecular properties that are related, through information sharing, whereas deep learning strategies increase the potential of capturing nonlinear relationships. In this work, we compare the binary classification capability of multitask deep and shallow neural networks to single‐task strategies used as benchmark (i.e., as k‐nearest neighbours, N‐nearest neighbours, random forest and Naïve Bayes), as well as multitask supervised self‐organizing maps.Comparison was carried out with an extended QSAR dataset containing annotations of molecular binding, agonism and antagonism activity on 11 nuclear receptors, for a total of 14,963 molecules, divided into training and test sets and labelled for their bioactivity on at least one of 30 binary tasks. Additional 304 chemicals were used as external evaluation set to further validate models.Although no approach systematically overperformed the others, task‐specific differences were found, suggesting the benefit of multitask learning for tasks that are less represented. On average, some of the single‐task approaches and multitask deep learning strategies had similar performances. However, the latter can have advantages, such as a simpler management of predictions and applicability domain assessment for future samples. On the other hand, the parameter tuning required by neural networks are generally time expensive suggesting that the modelling strategy should be evaluated case by case.

Feature Extraction Methods in Quantitative Structure–Activity Relationship Modeling: A Comparative Study

Computational approaches for synthesizing new chemical compounds have resulted in a major explosion of chemical data in the field of drug discovery. The quantitative structure–activity relationship (QSAR) is a widely used classification and regression method used to represent the relationship between a chemical structure and its activities. This research focuses on the effect of dimensionality-reduction techniques on a high-dimensional QSAR dataset. Because of the multi-dimensional nature of QSAR, dimensionality-reduction techniques have become an integral part of its modeling process. Principal component analysis (PCA) is a feature extraction technique with several applications in exploratory data analysis, visualization and dimensionality reduction. However, linear PCA is inadequate to handle the complex structure of QSAR data. In light of the wide array of current feature-extraction techniques, we perform a comparative empirical study to investigate five feature-extraction techniques: PCA, kernel PCA, deep generalized autoencoder (dGAE), Gaussian random projection (GRP), and sparse random projection (SRP). The experiments are performed on a high-dimensional QSAR dataset, which comprises 6394 features. The transformed low-dimensional dataset is inputted into a deep learning classification model to predict a QSAR biological activity. Three approaches are adopted to validate and measure the proposed techniques: (i) comparing the performance of the classification models, (ii) visualizing the relationship (correlation) between features in the low-dimension Euclidean space, and (iii) validating the proposed techniques using an external dataset. To the best of our knowledge, this study is the first to investigate and compare the aforementioned feature-extraction techniques in QSAR modeling context. The results obtained provide invaluable insights regarding the behavior of different techniques with both negative and positive classes. With linear PCA as a baseline, we prove that the investigated techniques substantially outperform the baseline in multiple accuracy measures and demonstrate useful ways of extracting significant features.

Descriptor Selection Improvements for Quantitative Structure-Activity Relationships.

Molecular descriptor selection is an essential procedure to improve a predictive quantitative structure-activity relationship (QSAR) model. However, within the QSAR model, there are a number of redundant, noisy and irrelevant descriptors. In this study, we propose a novel descriptor selection framework using self-paced learning (SPL) via sparse logistic regression (LR) with Logsum penalty (SPL-Logsum), which can simultaneously adaptively identify the simple and complex samples and avoid over-fitting. SPL is inspired by the learning process of humans or animals gradually learned from simple and complex samples to train models, and the Logsum penalized LR helps to select a small subset of significant molecular descriptors for improving the QSAR models. Experimental results on some simulations and three public QSAR datasets show that our proposed SPL-Logsum framework outperforms other existing sparse methods regarding the area under the curve, sensitivity, specificity, accuracy, and -values.

Analysis and Comparison of Vector Space and Metric Space Representations in QSAR Modeling.

The performance of quantitative structure–activity relationship (QSAR) models largely depends on the relevance of the selected molecular representation used as input data matrices. This work presents a thorough comparative analysis of two main categories of molecular representations (vector space and metric space) for fitting robust machine learning models in QSAR problems. For the assessment of these methods, seven different molecular representations that included RDKit descriptors, five different fingerprints types (MACCS, PubChem, FP2-based, Atom Pair, and ECFP4), and a graph matching approach (non-contiguous atom matching structure similarity; NAMS) in both vector space and metric space, were subjected to state-of-art machine learning methods that included different dimensionality reduction methods (feature selection and linear dimensionality reduction). Five distinct QSAR data sets were used for direct assessment and analysis. Results show that, in general, metric-space and vector-space representations are able to produce equivalent models, but there are significant differences between individual approaches. The NAMS-based similarity approach consistently outperformed most fingerprint representations in model quality, closely followed by Atom Pair fingerprints. To further verify these findings, the metric space-based models were fitted to the same data sets with the closest neighbors removed. These latter results further strengthened the above conclusions. The metric space graph-based approach appeared significantly superior to the other representations, albeit at a significant computational cost.

A new hybrid firefly algorithm and particle swarm optimization for tuning parameter estimation in penalized support vector machine with application in chemometrics

In quantitative structure–activity relationship (QSAR) classification, descriptor selection is one of the most important topics in the chemometrics. The selection of descriptors can be considered to be a variable selection problem that aims to find a small subset of descriptors that has the most discriminative information for the classification target. Penalized support vector machine (PSVM) proved its effectiveness by creating a strong classifier that combines the advantages of the support vector machine and the penalization. PSVM with L1-norm and smoothly clipped absolute deviation (SCAD) penalty is the most widely used methods. However, the efficiency of PSVM with these penalties depends on appropriately choosing the tuning parameter which is involved in both penalties. Hybrid metaheuristics algorithms are of the most interesting recent trends in optimization to escape from the trapped in the local optimal. In this paper, a new hybrid firefly algorithm and particle swarm optimization is proposed to determine the tuning parameter in PSVM. Our proposed algorithm can efficiently exploit the strong points of both firefly and particle swarm algorithms in finding the most relevant descriptors with high classification performance. The experimental results on four benchmark QSAR datasets show the superior performance of the proposed algorithm in terms of classification accuracy and the number of selected descriptors compared with other competitor methods.

Sparse Generative Topographic Mapping for Both Data Visualization and Clustering.

To achieve simultaneous data visualization and clustering, the method of sparse generative topographic mapping (SGTM) is developed by modifying the conventional GTM algorithm. While the weight of each grid point is constant in the original GTM, it becomes a variable in the proposed SGTM, enabling data points to be clustered on two-dimensional maps. The appropriate number of clusters is determined by optimization based on the Bayesian information criterion. Analysis of numerical simulation data sets along with quantitative structure-property relationship and quantitative structure-activity relationship data sets confirmed that the proposed SGTM provides the same degree of visualization performance as the original GTM and clusters data points appropriately. Python and MATLAB codes for the proposed algorithm are available at https://github.com/hkaneko1985/gtm-generativetopographicmapping .

Optimal Piecewise Linear Regression Algorithm for QSAR Modelling.

Quantitative Structure-Activity Relationship (QSAR) models have been successfully applied to lead optimisation, virtual screening and other areas of drug discovery over the years. Recent studies, however, have focused on the development of models that are predictive but often not interpretable. In this article, we propose the application of a piecewise linear regression algorithm, OPLRAreg, to develop both predictive and interpretable QSAR models. The algorithm determines a feature to best separate the data into regions and identifies linear equations to predict the outcome variable in each region. A regularisation term is introduced to prevent overfitting problems and implicitly selects the most informative features. As OPLRAreg is based on mathematical programming, a flexible and transparent representation for optimisation problems, the algorithm also permits customised constraints to be easily added to the model. The proposed algorithm is presented as a more interpretable alternative to other commonly used machine learning algorithms and has shown comparable predictive accuracy to Random Forest, Support Vector Machine and Random Generalised Linear Model on tests with five QSAR data sets compiled from the ChEMBL database.

Automatic outlier sample detection based on regression analysis and repeated ensemble learning

The fields of chemoinformatics and chemometrics require regression models with high prediction performance. To construct predictive regression models by appropriately detecting outlier samples, a new outlier detection and regression method based on ensemble learning is proposed. Multiple regression models are constructed and y-values are estimated based on ensemble learning. Outlier samples are then detected by comprehensively considering all regression models. Furthermore, it is possible to detect outlier samples robustly and independently by repeated calculations. By analyzing a numerical simulation dataset, two quantitative structure-activity relationship datasets and two quantitative structure-property relationship datasets, it is confirmed that automatic outlier sample detection can be achieved, informative compounds can be selected, and the estimation performance of regression models is improved.

Discussion on Regression Methods Based on Ensemble Learning and Applicability Domains of Linear Submodels.

To develop a new ensemble learning method and construct highly predictive regression models in chemoinformatics and chemometrics, applicability domains (ADs) are introduced into the ensemble learning process of prediction. When estimating values of an objective variable using subregression models, only the submodels with ADs that cover a query sample, i.e., the sample is inside the model's AD, are used. By constructing submodels and changing a list of selected explanatory variables, the union of the submodels' ADs, which defines the overall AD, becomes large, and the prediction performance is enhanced for diverse compounds. By analyzing a quantitative structure-activity relationship data set and a quantitative structure-property relationship data set, it is confirmed that the ADs can be enlarged and the estimation performance of regression models is improved compared with traditional methods.

A combined Fisher and Laplacian score for feature selection in QSAR based drug design using compounds with known and unknown activities.

Quantitative structure-activity relationship (QSAR) is an effective computational technique for drug design that relates the chemical structures of compounds to their biological activities. Feature selection is an important step in QSAR based drug design to select the most relevant descriptors. One of the most popular feature selection methods for classification problems is Fisher score which aim is to minimize the within-class distance and maximize the between-class distance. In this study, the properties of Fisher criterion were extended for QSAR models to define the new distance metrics based on the continuous activity values of compounds with known activities. Then, a semi-supervised feature selection method was proposed based on the combination of Fisher and Laplacian criteria which exploits both compounds with known and unknown activities to select the relevant descriptors. To demonstrate the efficiency of the proposed semi-supervised feature selection method in selecting the relevant descriptors, we applied the method and other feature selection methods on three QSAR data sets such as serine/threonine-protein kinase PLK3 inhibitors, ROCK inhibitors and phenol compounds. The results demonstrated that the QSAR models built on the selected descriptors by the proposed semi-supervised method have better performance than other models. This indicates the efficiency of the proposed method in selecting the relevant descriptors using the compounds with known and unknown activities. The results of this study showed that the compounds with known and unknown activities can be helpful to improve the performance of the combined Fisher and Laplacian based feature selection methods.

Descriptor Selection via Log-Sum Regularization for the Biological Activities of Chemical Structure.

The quantitative structure-activity relationship (QSAR) model searches for a reliable relationship between the chemical structure and biological activities in the field of drug design and discovery. (1) Background: In the study of QSAR, the chemical structures of compounds are encoded by a substantial number of descriptors. Some redundant, noisy and irrelevant descriptors result in a side-effect for the QSAR model. Meanwhile, too many descriptors can result in overfitting or low correlation between chemical structure and biological bioactivity. (2) Methods: We use novel log-sum regularization to select quite a few descriptors that are relevant to biological activities. In addition, a coordinate descent algorithm, which uses novel univariate log-sum thresholding for updating the estimated coefficients, has been developed for the QSAR model. (3) Results: Experimental results on artificial and four QSAR datasets demonstrate that our proposed log-sum method has good performance among state-of-the-art methods. (4) Conclusions: Our proposed multiple linear regression with log-sum penalty is an effective technique for both descriptor selection and prediction of biological activity.

Development of MLR and SVM Aided QSAR Models to Identify Common SAR of GABA Uptake Herbal Inhibitors used in the Treatment of Schizophrenia.

Background: Alterations in GABAnergic system are implicated in the pathophysiology of schizophrenia. Available antipsychotics that target GABA receptor form a desirable therapeutic strategy in the treatment regimen of schizophrenia, unfortunately, suffer serious setback due to their prolonged side effects. The present investigation focuses on developing QSAR models from the biological activity of herbal compounds and their derivatives that promise to be alternative candidates to GABA uptake inhibitors.Methods: Three sets of compounds were undertaken in the study to develop QSAR models. The first set consisted of nine compounds which included Magnolol, Honokiol and other GABA acting established compounds. The second set consisted of 16 derivatives of N-diarylalkenyl-piperidinecarboxylic acid. The third QSAR dataset was made up of thirty two compounds which were Magnolol and Honokiol derivatives. Multiple linear regressions (MLR) and support vector machine (SVM) supervised quantitative structure-activity relationship (QSAR) models were developed to predict the biological activity of these three sets. The purpose of taking three QSAR sets of diverse chemical structures but identical in their GABA targeting and pharmacological action was to identify common chemical structure features responsible for structure-activity relationship (SAR).Results: Linear and non-linear QSAR models confirmed that the three sets shared common structural descriptors derived from WHIM (Weighted Holistic Invariant Molecular descriptors), 3D-MoRSE and Eigenvalue classes.Conclusion: It was concluded that properties like electro negativity and polarizability play a crucial role in controlling the activity of herbal compounds against GABA receptor.