Towards better understanding of feature-selection or reduction techniques for Quantitative Structure–Activity Relationship models

Application of spectroscopic technology with machine learning in Chinese herbs from seeds to medicinal materials: The case of genus Paris

To ensure the safety and efficacy of Chinese herbs, it is of great significance to conduct rapid quality detection of Chinese herbs at every link of their supply chain. Spectroscopic technology can reflect the overall chemical composition and structural characteristics of Chinese herbs, with the multi-component and multitarget characteristics of Chinese herbs. This review took the genus Paris as an example, and applications of spectroscopic technology with machine learning (ML) in supply chain of the genus Paris from seeds to medicinal materials were introduced. The specific contents included the confirmation of germplasm resources, identification of growth years, cultivar, geographical origin, and original processing and processing methods. The potential application of spectroscopic technology in genus Paris was pointed out, and the prospects of combining spectroscopic technology with blockchain were proposed. The summary and prospects presented in this paper will be beneficial to the quality control of the genus Paris in all links of its supply chain, so as to rationally use the genus Paris resources and ensure the safety and efficacy of medication.

Predicting ruminal degradability and chemical composition of corn silage using near-infrared spectroscopy and multivariate regression

The aim of this study was to develop and validate regression models to predict the chemical composition and ruminal degradation parameters of corn silage by near-infrared spectroscopy (NIR). Ninety-four samples were used to develop and validate the models to predict corn silage composition. A subset of 23 samples was used to develop and validate models to predict ruminal degradation parameters of corn silage. Wet chemistry methods were used to determine the composition values and ruminal degradation parameters of the corn silage samples. The dried and ground samples had their NIR spectra scanned using a poliSPECNIR 900-1700 model NIR sprectrophotometer (ITPhotonics S.r.l, Breganze, IT.). The models were developed using regression by partial least squares (PLS), and the ordered predictor selection (OPS) method was used. In general, the regression models obtained to predict the corn silage composition (P>0.05), except the model for organic matter (OM), adequately estimated the studied properties. It was not possible to develop prediction models for the potentially degradable fraction in the rumen of OM and crude protein and the degradation rate of OM. The regression models that could be obtained to predict the ruminal degradation parameters showed correlation coefficient of calibration between 0.530 and 0.985. The regression models developed to predict CS composition accurately estimated the CS composition, except the model for OM. The NIR has potential to be used by nutritionists as a rapid prediction tool for ruminal degradation parameters in the field.

Utilization of multiple-dilution fluorescence fingerprint facilitates prediction of chemical attributes in spice extracts

Fluorescence Fingerprint (FF) is a powerful tool for rapid quality assessment of various foods and plant-derived products. However, the conventional utilization of FFs measured at a single dilution level (DL) to substitute chemical analyses is extremely challenging, especially for multicomponent materials like spice extracts because fluorescence intensity and concentration widely differ between components, with complex phenomena like inner filter effects. Here, we proposed a new strategy to use the meta-data comprised of FFs measured at multiple DLs with machine learning to estimate common chemical attributes including total polyphenol and flavonoid contents, and antioxidant abilities. This strategy achieved more consistently satisfactory performance in estimation of all chemical attributes of spice extracts compared to using a single DL. Hence, the workflow employed in this study is expected to serve as an alternative method to quickly evaluate the chemical quality of spice extracts, as well as other plant products and food materials.

A Primer on 2D Descriptors in Selectivity Modeling for Asymmetric Catalysis

Machine learning has permeated all fields of research, including chemistry, and is now an integral part of the design of novel compounds with desired properties. In the field of asymmetric catalysis, the preference still lies with models based on a physical understanding of the catalysis phenomenon and the electronic and steric properties of catalysts. However, such models require quantum chemical calculations and are thus limited by their computational cost. Here, we highlight the recent advances in modeling catalyst selectivity by using the 2D structures of catalysts and substrates. While these have a less explicit mechanistic connection to the modeled property, 2D descriptors, such as topological indices, molecular fingerprints, and fragments, offer the tremendous advantages of low cost and high speed of calculations. This makes them optimal for the in-silico screening of large amounts of data. We provide an overview of common quantitative structure-property relationship workflow, model building and validation techniques, applications of these methodologies in asymmetric catalysis design, and an outlook on improving the understanding of 2D-based models.

A partial least squares and artificial neural network study for a series of arylpiperazines as antidepressant agents

Depression affects more than 300 million people around the world and can lead to suicide. About 30% of patients on treatment for depression drop out of therapy due to side effects or to latency time associated to therapeutic effects. 5-HT receptor, known as serotonin, is considered the key in depression treatment. Arylpiperazine compounds are responsible for several pharmacological effects and are considered as ligands in serotonin receptors, such as the subtype 5-HT_2a. Here, in silico studies were developed using partial least squares (PLSs) and artificial neural networks (ANNs) to design new arylpiperazine compounds that could interact with the 5-HT_2a receptor. First, molecular and electronic descriptors were calculated and posteriorly selected from correlation matrixes and genetic algorithm (GA). Then, the selected descriptors were used to construct PLS and ANN models that showed to be robust and predictive. Lastly, new arylpiperazine compounds were designed and their biological activity values were predicted by both PLS and ANN models. It is worth to highlight compounds G5 and G7 (predicted by the PLS model) and G3 and G15 (predicted by the ANN model), whose predicted pIC₅₀ values were as high as the three highest values from the arylpiperazine original set studied here. Therefore, it can be asserted that the two models (PLS and ANN) proposed in this work are promising for the prediction of the biological activity of new arylpiperazine compounds and may significantly contribute to the design of new drugs for the treatment of depression.

A modified binary particle swarm optimization with a machine learning algorithm and molecular docking for QSAR modelling of cholinesterase inhibitors

The acetylcholinesterase (AChE) and butyrylcholinesterase (BuChE) inhibitors play a key role in treating Alzheimer's disease. This study proposes an approach that integrates a modified binary particle swarm optimization (PSO) with a machine learning algorithm for building QSAR models to predict the activity of inhibitors for AChE and BuChE enzymes. More precisely, it uses a transfer function to convert the continuous search space of PSO to binary. Furthermore, it utilizes the concepts of catfish effect and chaotic map to improve exploration ability in searching for an optimum subset of descriptors for QSAR model constructions. Then, through a statistical method, it employs a machine learning algorithm to evaluate the fitness value of each candidate subset of features. Different combinations of four transfer functions with four machine learning algorithms, including K-nearest neighbour, multiple linear regression, support vector machine, and regression tree, were used to build several variants of the proposed algorithm. QSAR models constructed by each version were verified by internal and external validations. The best variants were selected based on a method called sum of ranking differences.

A deep learning approach to predict blood-brain barrier permeability

The blood-brain barrier plays a crucial role in regulating the passage of 98% of the compounds that enter the central nervous system (CNS). Compounds with high permeability must be identified to enable the synthesis of brain medications for the treatment of various brain diseases, such as Parkinson's, Alzheimer's, and brain tumors. Throughout the years, several models have been developed to solve this problem and have achieved acceptable accuracy scores in predicting compounds that penetrate the blood-brain barrier. However, predicting compounds with "low" permeability has been a challenging task. In this study, we present a deep learning (DL) classification model to predict blood-brain barrier permeability. The proposed model addresses the fundamental issues presented in former models: high dimensionality, class imbalances, and low specificity scores. We address these issues to enhance the high-dimensional, imbalanced dataset before developing the classification model: the imbalanced dataset is addressed using oversampling techniques and the high dimensionality using a non-linear dimensionality reduction technique known as kernel principal component analysis (KPCA). This technique transforms the high-dimensional dataset into a low-dimensional Euclidean space while retaining invaluable information. For the classification task, we developed an enhanced feed-forward deep learning model and a convolutional neural network model. In terms of specificity scores (i.e., predicting compounds with low permeability), the results obtained by the enhanced feed-forward deep learning model outperformed those obtained by other models in the literature that were developed using the same technique. In addition, the proposed convolutional neural network model surpassed models used in other studies in multiple accuracy measures, including overall accuracy and specificity. The proposed approach solves the problem inevitably faced with obtaining low specificity resulting in high false positive rate.

Improvement of quantitative structure-retention relationship models for chromatographic retention prediction of peptides applying individual local partial least squares models

In Reversed-Phase Liquid Chromatography, Quantitative Structure-Retention Relationship (QSRR) models for retention prediction of peptides can be built, starting from large sets of theoretical molecular descriptors. Good predictive QSRR models can be obtained after selecting the most informative descriptors. Reliable retention prediction may be an aid in the correct identification of proteins/peptides in proteomics and in chromatographic method development. Traditionally, global QSRR models are built, using a calibration set containing a representative range of analytes. In this study, a strategy is presented to build individual local Partial Least Squares (PLS) models for peptides, based on selected local calibration samples, most similar to the specific query peptide to be predicted. Similar local calibration peptides are selected from a possible calibration set. The calibration samples with the lowest Euclidian distances to the query peptide are considered as most similar. Two Euclidian distances are investigated as similarity parameter, (i) in the autoscaled descriptor space and, (ii) in the PLS factor space of the global calibration samples, both after variable selection by the Final Complexity Adapted Models (FCAM) method. The predictive abilities of individual local QSRR PLS models for peptides, developed with both Euclidian distances, are found significantly better than those of two global models, i.e. before and after FCAM variable selection. The predictive abilities of the local models, developed with distances calculated in the PLS factor space, were best.

Identification of novel monoamine oxidase selective inhibitors employing a hierarchical ligand-based virtual screening strategy

Aim: Due to the pivotal role in the oxidative deamination of monoamine neurotransmitters, two distinct monoamine oxidase (MAO) subtypes, MAO-A and MAO-B, present a significant pharmacological interest. Here, we reported a hierarchical and time-efficient ligand-based virtual screening strategy to identify potent selective and reversible MAO inhibitors. Result: A total of 130 compounds were assessed in dose–response biochemical assay against MAOs. Among them, 70 compounds were active with inhibition higher than 70%, involving 25 compounds with IC50 values less than 1 μM. Conclusion: Our research demonstrated the validity of Biologically Relevant Spectrum (BRS-3D) in predicting subtype-selective ligands and afforded a novel highly efficient way to develop selective inhibitors in the early stage of drug discovery.

Natural products as inhibitors of Leishmania major dihydroorotate dehydrogenase

The flavoenzyme dihydroorotate dehydrogenase (DHODH) catalyzes the fourth reaction of the de novo pyrimidine biosynthetic pathway, which exerts vital functions in the cells, especially within DNA and RNA biosynthesis. Thus, this enzyme stands out as a new key molecular target for parasites causing Neglected Diseases (NDs). Focused on contributing to the development of new therapeutic alternatives for NDs, in this study, for the first time, a screening of 57 natural products for in vitro inhibition of Leishmania major DHODH (LmDHODH) was carried out, including cross validation against the human DHODH (HsDHODH). A subset of natural products consisting of 21 sesquiterpene lactones (STLs) was submitted to QSAR studies. Additionally, thermostability studies by differential scanning fluorimetry (DSF) were performed to determine whether the STLs are effectively or not binding to the enzyme. The IC₅₀ values against LmDHODH varied from 27 to 1200 μM; only irrelevant inhibition was obtained on HsDHODH. DSF assays confirmed binding of STLs to LmDHODH; moreover, it is suggested that such inhibitors might act in a different site other than the active site. A reliable QSAR model based on molecular descriptors was obtained (R²: 0.83; Q²_CV: 0.69 and Q²_EXT/F2: 0.66) indicating that stronger inhibition requires a balanced distribution of the hydrophobic regions across the molecular surface, as well as higher width and lower hydrophobicity of the molecules. A pharmacophore-based 3D-QSAR approach also afforded a useful model (R²: 0.72; Q²_CV: 0.50 and Q²_EXT/F2: 0.62), which confirmed the importance of proper orientation of the ligands, molecular surface features and shape for stronger inhibition, reflecting properties of a putative common binding site. These data indicated for the first time that natural products can actually inhibit LmDHODH and highlighted some metabolites as potentially interesting starting points for the discovery of more potent LmDHODH inhibitors, ultimately aiming at new effective therapeutic alternatives for leishmaniasis and, possibly, other NDs caused by trypanosomatids.

Computational modeling approaches to quantitative structure-binding kinetics relationships in drug discovery

Simple comparative correlation analyses and quantitative structure-kinetics relationship (QSKR) models highlight the interplay of kinetic rates and binding affinity as an essential feature in drug design and discovery. The choice of the molecular series, and their structural variations, used in QSKR modeling is fundamental to understanding the mechanistic implications of ligand and/or drug-target binding and/or unbinding processes. Here, we discuss the implications of linear correlations between kinetic rates and binding affinity constants and the relevance of the computational approaches to QSKR modeling.

QSAR modelling using combined simple competitive learning networks and RBF neural networks

The aim of this study was to propose a QSAR modelling approach based on the combination of simple competitive learning (SCL) networks with radial basis function (RBF) neural networks for predicting the biological activity of chemical compounds. The proposed QSAR method consisted of two phases. In the first phase, an SCL network was applied to determine the centres of an RBF neural network. In the second phase, the RBF neural network was used to predict the biological activity of various phenols and Rho kinase (ROCK) inhibitors. The predictive ability of the proposed QSAR models was evaluated and compared with other QSAR models using external validation. The results of this study showed that the proposed QSAR modelling approach leads to better performances than other models in predicting the biological activity of chemical compounds. This indicated the efficiency of simple competitive learning networks in determining the centres of RBF neural networks.

A novel applicability domain technique for mapping predictive reliability across the chemical space of a QSAR: reliability-density neighbourhood

The ability to define the regions of chemical space where a predictive model can be safely used is a necessary condition to assure the reliability of new predictions. This implies that reliability must be determined across chemical space in the attempt to localize “safe” and “unsafe” regions for prediction. As a result we devised an applicability domain technique that addresses the data locally instead of handling it as a whole—the reliability-density neighbourhood (RDN). The main novelty aspect of this method is that it characterizes each single training instance according to the density of its neighbourhood in the training set, as well as its individual bias and precision. By scanning through the chemical space (by iteratively increasing the applicability domain area), it was observed that new test compounds are successively included into the applicability domain region in such a manner that strongly correlates to their predictive performance. This allows the mapping of local reliability across different locations in the training set space, and thus allows identifying regions where the model has low reliability. This method also showed matching profiles between two external sets, which is an indication that it performs robustly with new data. Another novel aspect in this technique is that it is paired with a specific feature selection algorithm. As a result, the impact of the feature set used was studied from which the top 20 features selected by ReliefF yielded the best results, as opposed to using the model’s features or the entire feature set as commonly done. As the third novel aspect, in this work we propose a new scoring function to help evaluate the quality of an applicability domain profile (i.e., the curve of accuracy vs the applicability domain measure in question). Overall, the RDN showed to be a promising method that can correctly sort new instances according to predictive performance. As a result, this technique can be received by an end-user as proof of concept for the performance of a QSAR model in new data, thus promoting the user’s trust on the QSAR output.Graphical abstract

Joint L2,1 Norm and Fisher Discrimination Constrained Feature Selection for Rational Synthesis of Microporous Aluminophosphates

Feature selection has been regarded as an effective tool to help researchers understand the generating process of data. For mining the synthesis mechanism of microporous AlPOs, this paper proposes a novel feature selection method by joint l_2,1 norm and Fisher discrimination constraints (JNFDC). In order to obtain more effective feature subset, the proposed method can be achieved in two steps. The first step is to rank the features according to sparse and discriminative constraints. The second step is to establish predictive model with the ranked features, and select the most significant features in the light of the contribution of improving the predictive accuracy. To the best of our knowledge, JNFDC is the first work which employs the sparse representation theory to explore the synthesis mechanism of six kinds of pore rings. Numerical simulations demonstrate that our proposed method can select significant features affecting the specified structural property and improve the predictive accuracy. Moreover, comparison results show that JNFDC can obtain better predictive performances than some other state-of-the-art feature selection methods.

Support vector regression-guided unravelling: antioxidant capacity and quantitative structure-activity relationship predict reduction and promotion effects of flavonoids on acrylamide formation

We used the support vector regression (SVR) approach to predict and unravel reduction/promotion effect of characteristic flavonoids on the acrylamide formation under a low-moisture Maillard reaction system. Results demonstrated the reduction/promotion effects by flavonoids at addition levels of 1–10000 μmol/L. The maximal inhibition rates (51.7%, 68.8% and 26.1%) and promote rates (57.7%, 178.8% and 27.5%) caused by flavones, flavonols and isoflavones were observed at addition levels of 100 μmol/L and 10000 μmol/L, respectively. The reduction/promotion effects were closely related to the change of trolox equivalent antioxidant capacity (ΔTEAC) and well predicted by triple ΔTEAC measurements via SVR models (R: 0.633–0.900). Flavonols exhibit stronger effects on the acrylamide formation than flavones and isoflavones as well as their O-glycosides derivatives, which may be attributed to the number and position of phenolic and 3-enolic hydroxyls. The reduction/promotion effects were well predicted by using optimized quantitative structure-activity relationship (QSAR) descriptors and SVR models (R: 0.926–0.994). Compared to artificial neural network and multi-linear regression models, SVR models exhibited better fitting performance for both TEAC-dependent and QSAR descriptor-dependent predicting work. These observations demonstrated that the SVR models are competent for predicting our understanding on the future use of natural antioxidants for decreasing the acrylamide formation.

Development of quantitative structure activity relationship (QSAR) model for disinfection byproduct (DBP) research: A review of methods and resources

Quantitative structure-activity relationship (QSAR) models are tools for linking chemical activities with molecular structures and compositions. Due to the concern about the proliferating number of disinfection byproducts (DBPs) in water and the associated financial and technical burden, researchers have recently begun to develop QSAR models to investigate the toxicity, formation, property, and removal of DBPs. However, there are no standard procedures or best practices regarding how to develop QSAR models, which potentially limit their wide acceptance. In order to facilitate more frequent use of QSAR models in future DBP research, this article reviews the processes required for QSAR model development, summarizes recent trends in QSAR-DBP studies, and shares some important resources for QSAR development (e.g., free databases and QSAR programs). The paper follows the four steps of QSAR model development, i.e., data collection, descriptor filtration, algorithm selection, and model validation; and finishes by highlighting several research needs. Because QSAR models may have an important role in progressing our understanding of DBP issues, it is hoped that this paper will encourage their future use for this application.

A structure–activity relationship study of the toxicity of ionic liquids using an adapted Ferreira–Kiralj hydrophobicity parameter

An alternative molecular descriptor for characterizing the hydrophobicity of ionic liquids, W_cAdap, is presented in this study.