Search

Recent advancements in deep learning have led to widespread applications of its algorithms to synthetic planning and reaction predictions in the field of chemistry. One major area, known as supervised learning, is being explored for predicting certain properties such as reaction yields and types. Many chemical descriptors known as fingerprints are being explored as potential candidates for reaction properties prediction. However, there are few studies that describe the permutational invariance of chemical fingerprints, which are concatenated at some stage before being fed to deep learning architecture. In this work, we show that by utilizing permutational invariance, we consistently see improved results in terms of accuracy relative to previously published studies. Furthermore, we are able to accurately predict hydrogen peroxide loss with our own dataset, which consists of more than 20 ingredients in each chemical formulation.

Graphical abstract, The world is facing the COVID-19 pandemic caused by Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2). Likewise, other viruses of the Coronaviridae family were responsible for causing epidemics earlier. To tackle these viruses, there is a lack of approved antiviral drugs. Therefore, we have developed robust computational methods to predict the repurposed drugs using machine learning techniques namely Support Vector Machine, Random Forest, k-Nearest Neighbour, Artificial Neural Network, and Deep Learning. We used the experimentally validated drugs/chemicals with anticorona activity (IC50/EC50) from ‘DrugRepV’ repository. The unique entries of SARS-CoV-2 (142), SARS (221), MERS (123), and overall Coronaviruses (414) were subdivided into the training/testing and independent validation datasets, followed by the extraction of chemical/structural descriptors and fingerprints (17968). The highly relevant features were filtered using the recursive feature selection algorithm. The selected chemical descriptors were used to develop prediction models with Pearson’s correlation coefficients ranging from 0.60 to 0.90 on training/testing. The robustness of the predictive models was further ensured using external independent validation datasets, decoy datasets, applicability domain, and chemical analyses. The developed models were used to predict promising repurposed drug candidates against coronaviruses after scanning the DrugBank. Top predicted molecules for SARS-CoV-2 were further validated by molecular docking against the spike protein complex with ACE receptor. We found potential repurposed drugs namely Verteporfin, Alatrofloxacin, Metergoline, Rescinnamine, Leuprolide, and Telotristat ethyl with high binding affinity. These ‘anticorona' computational models would assist in antiviral drug discovery against SARS-CoV-2 and other Coronaviruses.

There has recently been a rapid progress in computational methods for determining protein targets of small molecule drugs, which will be termed as compound protein interaction (CPI). In this review, we comprehensively review topics related to computational prediction of CPI. Data for CPI has been accumulated and curated significantly both in quantity and quality. Computational methods have become powerful ever to analyze such complex the data. Thus, recent successes in the improved quality of CPI prediction are due to use of both sophisticated computational techniques and higher quality information in the databases. The goal of this article is to provide reviews of topics related to CPI, such as data, format, representation, to computational models, so that researchers can take full advantages of these resources to develop novel prediction methods. Chemical compounds and protein data from various resources were discussed in terms of data formats and encoding schemes. For the CPI methods, we grouped prediction methods into five categories from traditional machine learning techniques to state-of-the-art deep learning techniques. In closing, we discussed emerging machine learning topics to help both experimental and computational scientists leverage the current knowledge and strategies to develop more powerful and accurate CPI prediction methods.

Drug discovery is a costly process which usually takes more than 10 years and billions of dollars for one successful drug to enter the market. Despite all the safety tests, drugs may still cause adverse reactions and be restricted in use or even withdrawn from the market. Drug-induced liver injury (DILI) is one of the major adverse drug reactions, and computational models may be used to predict and reduce it. To assess the computational prediction performance of DILI, we curated DILI endpoints from three databases and prepared drug features including chemical descriptors, therapeutic classifications, gene expressions, and binding proteins. We trained machine-learning models to predict the various DILI endpoints using different drug features. Using the optimal feature sets, the top-performing models obtained areas under the receiver operating characteristic curve (AUC) around 0.8 for some DILI endpoints. We found that some features, including therapeutic classifications and proteins, have good prediction performance towards DILI. We also discovered that the severity of DILI endpoints as well as the selection of negative samples may significantly affect the prediction results. Overall, our study provided a comprehensive collection, curation, and prediction of DILI endpoints using various drug features, which may help the drug researchers to better understand and prevent DILI during the drug discovery process.

The current standard of inflammatory bowel disease (IBD), especially, Crohn’s disease (CD), diagnosis is set through an invasive endoscopy procedure. However, serum metabolites hold potential as useful biomarkers for non-invasive diagnosis and treatment of IBDs. The goal of this research was to elucidate the biomarkers including metabolites and genes related to IBDs, to show their distinguishing and common features, and to create a machine-learning (ML) model for recognition of each disease. We explored metabolic pathways and gene–metabolite networks related to unspecified-IBD (uIBD), Crohn’s disease (CD), and ulcerative colitis (UC). P38 MAPK, ERK1/2, AMPK, and proinsulin were found to be closely related to the pathology of IBDs. The best performing ML model, trained on filtered disease-specific metabolite datasets, was able to predict metabolite class with 92.17% accuracy. Through examination of IBD-related serum, significant relationships between the inputted metabolites and certain metabolic and signaling pathways were found, which can be pinpointed and used to increase accuracy of disease diagnoses. Development of a ML model including metabolites and their chemical descriptors made it possible to achieve considerable accuracy of prediction of the IBDs. Our results elucidate a large variety of metabolites, genes, and pathways that could be used for better understanding of IBDs’ molecular mechanisms.

International audience; The rising atmospheric CO2 level is partly responsible for global warming. Despite numerous warnings from scientists during the past years, nations are reacting too slowly, and thus, we will probably reach a situation needing rapid and effective techniques to reduce atmospheric CO2. Therefore, advanced engineering methods are particularly important to decrease the greenhouse effect, for instance, by capturing CO2 using solvents. Experimental testing of many solvents under different conditions is necessary but time-consuming. Alternatively, modeling CO2 capture by solvents using a nonlinear fitting machine learning is a rapid way to select potential solvents, prior to experimentation. Previous predictive machine learning models were mainly designed for blended solutions in water using the solution concentration as the main input of the model, which was not able to predict CO2 solubility in different types of physical solvents. To address this issue, here, we developed a new descriptor-based chemoinformatics model for predicting CO2 solubility in physical solvents in the form of mole fraction. The input factors include organic structural and bond information, thermodynamic properties, and experimental conditions. We studied the solvents from 823 data, including methanol (165 data), ethanol (138), n-propanol (98), n-butanol (64), n-pentanol (59), ethylene glycol (52), propylene glycol (54), acetone (51), 2-butanone (49), ethylene glycol monomethyl ether (46 data), and ethylene glycol monoethyl ether (47), using artificial neural networks as the machine learning model. Results show that our descriptor-based model predicts the CO2 absorption in physical solvents with generally higher accuracy and low root-mean-squared errors. Our findings show that using a set of simple but effective chemoinformatics-based descriptors, intrinsic relationships between the general properties of physical solvents and their CO2 solubility can be precisely fitted with machine learning.

Pseudomonas aeruginosa is a Gram-negative bacillus included among the six "ESKAPE" microbial species with an outstanding ability to "escape" currently used antibiotics and developing new antibiotics against it is of the highest priority. Whereas minimum inhibitory concentration (MIC) values against Pseudomonas aeruginosa have been used previously for QSAR model development, disk diffusion results (inhibition zones) have not been apparently used for this purpose in the literature, and we decided to explore their use in this sense. We developed multiple QSAR methods using several machine learning algorithms (Support vector classifier, K Nearest Neighbors, Random Forest Classifier, Decision Tree Classifier, AdaBoost Classifier, Logistic Regression, and Naive Bayes Classifier). The main descriptors used in building the models belonged to the families of adjacency matrix, constitutional descriptors, first highest eigenvalue of Burden matrix, centered Moreau-Broto autocorrelation, and averaged and centered Moreau-Broto autocorrelation descriptors. A total of 32 models were built, of which 28 were selected and stacked to create a meta-model. In terms of balanced accuracy, the best performance was provided by KNN, SVM and AdaBoost algorithms, but the ensemble method had slightly superior results in nested cross-validation.

Non-targeted strategies based on high-performance liquid chromatography with ultraviolet detection (HPLC-UV) and fluorescence detection (HPLC-FLD) fingerprints were evaluated to accomplish the classification and authentication of instant coffee (40 samples), instant decaf coffee (26 samples), and chicory (22 samples, including both ground and instant), as well as to detect and quantify frauds based on chicory adulteration by multivariate chemometric methods. HPLC-UV and HPLC-FLD fingerprints were simultaneously obtained with a HPLC-UV-FLD instrument, and they proved to be excellent chemical descriptors for the classification of coffee and decaf coffee against chicory samples by partial least squares regression-discriminant analysis (PLS-DA). In contrast, HPLC-UV fingerprints improved the classification results when addressing coffee against decaf coffee samples (94.4% classification rate in comparison to 83.3% for HPLC-FLD fingerprints). Besides, the proposed methodologies resulted to be excellent to detect and quantify fraud levels in coffee and decaf coffee samples adulterated with chicory by using partial least squares (PLS) regression, exhibiting good calibration linearities, calibration errors, and prediction errors. In this case, improved capabilities were observed with HPLC-FLD fingerprints, providing better PLS calibration linearities (R2 > 0.999), lower calibration errors (≤0.8%), and similar to better prediction errors (2.9–3.2%) in comparison to HPLC-UV fingerprints.

The principle of computational toxicity prediction is that chemicals with similar molecular structures may possess similar toxicological pathways and effects. There have been many methods that represented each chemical by a set of descriptors, which are identified by experts as promising properties for predicting biological activity or toxicity. These chemical descriptors play a critical role in computational methods, that task correlated descriptors are favorable to achieve high prediction performance. However, there are few work compare the effectiveness of chemical descriptors and evaluate their performance in toxicity prediction. In this paper, we propose a novel ensemble feature selection method based on random under-sampling to analysis the effectiveness of chemical descriptors adopted in toxicity prediction application. The proposed method is efficient and can relief the imbalanced data problem of toxicity. Experiment results on the tox21 toxicity prediction dataset show that "molecular property", "connectivity" and "topological" descriptor are the three most important descriptors for toxicity prediction tasks among the 12 popular descriptors adopted in toxicity prediction applications. The results of this study can be used as a guide to propose new descriptors for chemical toxicity prediction.

We describe a novel algorithm for generating representational embeddings of chemical matter based on the biomedical literature/semantic contexts in which they occur. We then demonstrate that these chemical descriptors have utility in nearest neighbor retrieval for early drug discovery tasks such as mechanism of action and target activity predictions.

Density functional theory (DFT) was performed in order to predict the structural, chemical descriptors and optoelectronic properties of the drugs Hydroxychloroquine and Azithromycin using the wB97XD, O3LYP and B3LYP functional with 6-31+G(d,p) basis set. It is observed from our studies that most of the descriptors presented show association with some processes, including absorption, blood-brain barrier transport, binding and even toxicity. Hence, the treatment of COVID-19 using Hydroxychloroquine and Azithromycin in some patients as single dose and their combination in patients with Corona virus resistance can be more effective. Our results show that these therapeutic molecules may also have good nonlinear optical applications, may have semiconductor character with wide band gap and can also be promising materials in the production of optoelectronic devices. The density of states and thermodynamic properties were equally determined., Pharmaceutical chemistry, Azithromycin; Hydroxychloroquine; COVID-19; Chemical descriptors; Optoelectronic devices; Nonlinear optical application

There are various types of hepatic steatosis of which non-alcoholic fatty liver disease, which may be caused by exposure to chemicals and environmental pollutants is the most prevalent, representing a potential major health risk. QSAR modelling has the potential to provide a rapid and cost-effective method to identify compounds which may trigger steatosis. Although models exist to predict key molecular initiating events of steatosis such as nuclear receptor binding, we are aware of no models to predict the apical effect steatosis. In this study, we describe the development of a QSAR model to predict steatosis using freely available machine learning tools. It was built using a dataset of 207 pharmaceuticals and pesticides which were identified as steatotic or non-steatotic from existing data from in vivo human and animal studies. The best performing model developed using the linear discriminant analysis module in TANAGRA, based on four chemical descriptors, had an accuracy of 70 %, a sensitivity of 66% and a specificity of 74 %. The expansion of the steatosis dataset to other chemical types, to enable the development of further models, would be of benefit in the identification of compounds with a range of mechanisms of action contributing to steatosis.

In the present study, a systematic effort has been made to predict the hemolytic potency of chemically modified peptides. All models have been trained, tested, and evaluated on a dataset that contains 583 modified hemolytic peptides and a balanced number of non-hemolytic peptides. Machine learning techniques have been used to build the classification models using an immense range of peptide features that include 2D, 3D descriptors, fingerprints, atom, and diatom compositions. Random Forest based model developed using fingerprints as an input feature achieved maximum accuracy of 78.33% with AUC of 0.86 on the main dataset and accuracy of 78.29% with AUC of 0.85 on the validation dataset. Models developed in this study have been incorporated in a web server “HemoPImod” to facilitate the scientific community (http://webs.iiitd.edu.in/raghava/hemopimod/).

Read-across is an important data gap filling technique used within category and analog approaches for regulatory hazard identification and risk assessment. Although much technical guidance is available that describes how to develop category/analog approaches, practical principles to evaluate and substantiate analog validity (suitability) are still lacking. This case study uses hindered phenols as an example chemical class to determine: (1) the capability of three structure fingerprint/descriptor methods (PubChem, ToxPrints and MoSS MCSS) to identify analogs for read-across to predict Estrogen Receptor (ER) binding activity and, (2) the utility of data confidence measures, physicochemical properties, and chemical R-group properties as filters to improve ER binding predictions. The training dataset comprised 462 hindered phenols and 257 non- hindered phenols. For each chemical of interest (target), source analogs were identified from two datasets (hindered and non-hindered phenols) that had been characterized by a fingerprint/descriptor method and by two cut-offs: (1) minimum similarity distance (range: 0.1 – 0.9) and, (2) N closest analogs (range: 1 – 10). Analogs were then filtered using: (1) physicochemical properties of the phenol (termed global filtering) and, (2) physicochemical properties of the R-groups neighboring the active hydroxyl group (termed local filtering). A read-across prediction was made for each target chemical on the basis of a majority vote of the N closest analogs. The results demonstrate that: (1) concordance in ER activity increases with structural similarity, regardless of the structure fingerprint/descriptor method, (2) increased data confidence significantly improves read-across predictions, and (3) filtering analogs using global and local properties can help identify more suitable analogs. This case study illustrates that the quality of the underlying experimental data and use of endpoint relevant chemical descriptors to evaluate source analogs are critical to achieving robust read-across predictions.

The computational prediction of ligand-biopolymer affinities is a crucial endeavor in modern drug discovery and one that still poses major challenges. The choice of the appropriate computational method often reveals itself as a trade-off between accuracy and speed, with mathematical devices referred to as scoring functions being the fastest. Among the many shortcomings of scoring functions there is the lack of universal applicability to every molecular system. This is so largely due to their reliance on atom type perception and/or parametrization. This article proposes the use of nonparametric Model of Effective Radii of Atoms descriptors that can be readily computed for the entire Periodic Table and demonstrate that, in combination with machine learning algorithms, they can yield competitive performances and chemically meaningful insights.

We created earlier a large machine‐readable database of 10,000 chemicals and 800,000 associated studies by natural language processing of the public parts of Registration, Evaluation, Authorisation and Restriction of Chemicals (REACH) registrations until December 2014. This database was used to assess the reproducibility of the six most frequently used Organisation for Economic Co‐operation and Development (OECD) guideline tests. These tests consume 55% of all animals in safety testing in Europe, i.e. about 600,000 animals. With 350–750 chemicals with multiple results per test, reproducibility (balanced accuracy) was 81% and 69% of toxic substances were found again in a repeat experiment (sensitivity 69%). Inspired by the increasingly used read‐across approach, we created a new type of QSAR, which is based on similarity of chemicals and not on chemical descriptors. A landscape of the chemical universe using 10 million structures was calculated, when based on Tanimoto indices similar chemicals are close and dissimilar chemicals far from each other. This allows placing any chemical of interest into the map and evaluating the information available for surrounding chemicals. In a data fusion approach, in which 74 different properties were taken into consideration, machine learning (random forest) allowed a fivefold cross‐validation for 190,000 (non‐) hazard labels of chemicals for which nine hazards were predicted. The balanced accuracy of this approach was 87% with a sensitivity of 89%. Each prediction comes with a certainty measure based on the homogeneity of data and distance of neighbours. Ongoing developments and future opportunities are discussed. published

Multiple approaches to quantitative structure-activity relationship (QSAR) modeling using various statistical or machine learning techniques and different types of chemical descriptors have been developed over the years. Oftentimes models are used in consensus to make more accurate predictions at the expense of model interpretation. We propose a simple, fast, and reliable method termed Multi-Descriptor Read Across (MuDRA) for developing both accurate and interpretable models. The method is conceptually related to the well-known kNN approach but uses different types of chemical descriptors simultaneously for similarity assessment. To benchmark the new method, we have built MuDRA models for six different end points (Ames mutagenicity, aquatic toxicity, hepatotoxicity, hERG liability, skin sensitization, and endocrine disruption) and compared the results with those generated with conventional consensus QSAR modeling. We find that models built with MuDRA show consistently high external accuracy similar to that of conventional QSAR models. However, MuDRA models excel in terms of transparency, interpretability, and computational efficiency. We posit that due to its methodological simplicity and reliable predictive accuracy, MuDRA provides a powerful alternative to a much more complex consensus QSAR modeling. MuDRA is implemented and freely available at the Chembench web portal ( https://chembench.mml.unc.edu/mudra ).

In QSAR, a statistical model is generated from a training set of molecules (represented by chemical descriptors) and their biological activities (an "activity model"). The aim of the field of domain applicability (DA) is to estimate the uncertainty of prediction of a specific molecule on a specific activity model. A number of DA metrics have been proposed in the literature for this purpose. A quantitative model of the prediction uncertainty (an "error model") can be built using one or more of these metrics. A previous publication from our laboratory ( Sheridan , R. P. J. Chem. Inf.2013 , 53 , 2837 - 2850 ) suggested that QSAR methods such as random forest could be used to build error models by fitting unsigned prediction errors against DA metrics. The QSAR paradigm contains two useful techniques: descriptor importance can determine which DA metrics are most useful, and cross-validation can be used to tell which subset of DA metrics is sufficient to estimate the unsigned errors. Previously we studied 10 large, diverse data sets and seven DA metrics. For those data sets for which it is possible to build a significant error model from those seven metrics, only two metrics were sufficient to account for almost all of the information in the error model. These were TREE_SD (the variation of prediction among random forest trees) and PREDICTED (the predicted activity itself). In this paper we show that when data sets are less diverse, as for example in QSAR models of molecules in a single chemical series, these two DA metrics become less important in explaining prediction error, and the DA metric SIMILARITYNEAREST1 (the similarity of the molecule being predicted to the closest training set compound) becomes more important. Our recommendation is that when the mean pairwise similarity (measured with the Carhart AP descriptor and the Dice similarity index) within a QSAR training set is less than 0.5, one can use only TREE_SD, PREDICTED to form the error model, but otherwise one should use TREE_SD, PREDICTED, SIMILARITYNEAREST1.

Using molecular simulation for adsorbent screening is computationally expensive and thus prohibitive to materials discovery. Machine learning (ML) algorithms trained on fundamental material properties can potentially provide quick and accurate methods for screening purposes. Prior efforts have focused on structural descriptors for use with ML. In this work, the use of chemical descriptors, in addition to structural descriptors, was introduced for adsorption analysis. Evaluation of structural and chemical descriptors coupled with various ML algorithms, including decision tree, Poisson regression, support vector machine and random forest, were carried out to predict methane uptake on hypothetical metal organic frameworks. To highlight their predictive capabilities, ML models were trained on 8% of a data set consisting of 130,398 MOFs and then tested on the remaining 92% to predict methane adsorption capacities. When structural and chemical descriptors were jointly used as ML input, the random forest model with 10-fold cross validation proved to be superior to the other ML approaches, with an R

A method to discover and correct errors in mass spectral libraries is described. Comparing across a set of highly curated reference libraries compounds that have the same chemical structure quickly identifies entries that are outliers. In cases where three or more entries for the same compound are compared, the outlier as determined by visual inspection was almost always found to contain the error. These errors were either in the spectrum itself or in the chemical descriptors that accompanied it. The method is demonstrated on finding errors in compounds of forensic interest in the NIST/EPA/NIH Mass Spectral Library. The target list of compounds checked was the Scientific Working Group for the Analysis of Seized Drugs (SWGDRUG) mass spectral library. Some examples of errors found are described. A checklist of errors that curators should look for when performing inter-library comparisons is provided. Graphical Abstract ᅟ.

The potential of visible fingerprints and physical-chemical parameters in combination with multivariate data analysis was examined to classify extra virgin olive oils (EVOOs) from different Spanish regions according to their geographical origin. Firstly, spectral and quality parameters matrices were processed separately and subsequently were joined to evaluate the effect of synergy on the information obtained from the different methodologies. Linear discriminant analysis (LDA) and partial least squares discriminant analysis (PLS-DA) were performed as classification methods. The results revealed a perfect discrimination between the defined categories after performing PLS-DA on the Fused matrix, reaching 100% of correct classifications and showed a clear improvement in the overall prediction rates (92.5%), so that the effect of synergy was confirmed. These accurate results emphasise the feasibility of the proposed strategy and encourage the development of similar approaches based on visible spectroscopy in olive oil quality and traceability determination.

Uncertainty was introduced to chemical descriptors of 16 publicly available data sets to various degrees and in various ways in order to investigate the effect on the predictive performance of the state-of-the-art method decision tree ensembles. A number of strategies to handle uncertainty in decision tree ensembles were evaluated. The main conclusion of the study is that uncertainty to a large extent may be introduced in chemical descriptors without impairing the predictive performance of ensembles and without the predictive performance being significantly reduced from a practical point of view. The investigation further showed that even when distributions of uncertain values were provided, the ensembles method could generate equally effective models from single-point samples from these distributions. Hence, there seems to be no advantage in using more elaborate methods for handling uncertainty in chemical descriptors when using decision tree ensembles as a modeling method for the considered types of introduced uncertainty.

Bacterial lipoproteins play critical roles in various physiological processes including the maintenance of pathogenicity and numbers of them are being considered as potential candidates for generating novel vaccines. In this work, we put forth an algorithm to identify and predict ligand-binding sites in bacterial lipoproteins. The method uses three types of pocket descriptors, namely fpocket descriptors, 3D Zernike descriptors and shell descriptors, and combines them with Support Vector Machine (SVM) method for the classification. The three types of descriptors represent shape-based properties of the pocket as well as its local physio-chemical features. All three types of descriptors, along with their hybrid combinations are evaluated with SVM and to improve classification performance, WEKA-InfoGain feature selection is applied. Results obtained in the study show that the classifier successfully differentiates between ligand-binding and non-binding pockets. For the combination of three types of descriptors, 10 fold cross-validation accuracy of 86.83% is obtained for training while the selected model achieved test Matthews Correlation Coefficient (MCC) of 0.534. Individually or in combination with new and existing methods, our model can be a very useful tool for the prediction of potential ligand-binding sites in bacterial lipoproteins.

A study on the differentiation of three types of blonde beer, such as lager, pilsner and low-alcohol, has been carried out. Several chemical descriptors, such as chloride, phosphate, sulphate, total amino acids, total polyphenols, pH, dry extract and absorbance at 430 nm, have been selected according to their significance on the brewing process as well as their ability as quality indicators. Principal component analysis has been used to visualise data trends. Lager and low-alcohol beers appeared separated in the plane of the first two principal components, but pilsner are mixed with these two types. Linear discriminant analysis was applied in order to construct a suitable classification model. The model was validated by applying a stratified delete-a-group jackknife resampling procedure obtaining an overall prediction ability of 92.7% and a specificity of 96.3%. This work solves the classification problem with results that are comparable to those obtained in previous studies. The main advantage of the proposed model is that it is built with chemical descriptors that are considered as quality indicators in brewery, being easily determined in routine analyses by using inexpensive equipment and common procedures.

Diversity-oriented organic synthesis (DOS) is a strategy to make compound collections to probe biological systems1-7. Designing better DOS libraries requires having methods to assess the consequences of different synthesis decisions on the biological performance of resulting library members8. Since we are particularly interested in how stereochemistry affects performance in biological assays, we prepared a disaccharide library containing systematic stereochemical variations, assayed the library for different biological effects, and developed methods to assess the similarity of performance between members across multiple assays. These methods allow us to ask which subsets of stereochemical features best predict similarity in patterns of biological performance between individual members and which features produce the greatest variation of outcomes. We anticipate that the data-analysis approach presented here can be generalized to other sets of biological assays and other chemical descriptors. Methods to assess which structural features of library members produce the greatest similarity in performance for a given set of biological assays should help prioritize synthesis decisions in second-generation library development targeting the underlying cell-biological processes. Methods to assess which structural features of library members produce the greatest variation in performance should help guide decisions about what synthetic methods need to be developed to make optimal small-molecule screening collections.

Antibiotics used in human and veterinary medicine have been found in samples from diverse environments in many parts of the world. To assess the environmental risks associated with them, data regarding their toxicity, occurrence, and fate are needed, but gathering such data is time-consuming and expensive. An efficient approach to address these difficulties would be to select a small subset of antibiotics with a wide variation in chemical characteristics, perform experimental tests on this subset, and then extrapolate the results to larger numbers of antibiotics, including the most potentially hazardous compounds. To assess the potential utility of such an approach, a set of 92 antibiotics for human use was studied and their structural properties were described with 24 chemical descriptors that included information on their steric, lipophilic, and electronic properties. Principal component analysis in combination with statistical experimental design was used to map the chemical diversity of the antibiotics and to select a small subset, a "training set", of 20 antibiotics. The chemical representativity of the training set was assessed in a quantitative structure-activity model established to predict ultimate biodegradation. The selected antibiotics showed to cover the chemical variation of the studied antibiotics and are suggested for use in future testing programs to assess antibiotics' fate and effects in the environment.

Millions of compounds are available as potential drug candidates. High throughput screening (HTS) is widely used in drug discovery to assay compounds for a particular biological activity. A common approach is to build a classification model using a smaller sample of assay data to predict the activity of unscreened compounds and hence select further compounds for assay. This improves the efficiency of the search by increasing the proportion of hits found among the assayed compounds. In many assays, the biological activity is dichotomized into a binary indicator variable; the explanatory variables are chemical descriptors capturing compound structure. A tree model is interpretable, which is key, since it is of interest to identify diverse chemical classes among the active compounds to serve as leads for drug optimization. Interpretability of a tree is often reduced, however, by the sheer size of the tree model and the number of variables and rules of the terminal nodes. We develop a "tree harvesting" algorithm to filter out redundant "junk" rules from the tree while retaining its predictive accuracy. This simplification can facilitate the process of uncovering key relations between molecular structure and activity and may clarify rules defining multiple activity mechanisms. Using data from the National Cancer Institute, we illustrate that many of the rules used to build a classification tree may be redundant. Unlike tree pruning, tree harvesting allows variables with junk rules to be removed near the top of the tree. The reduction in complexity of the terminal nodes improves the interpretability of the model. The algorithm also aims to reorganize the tree nodes associated with the interesting "active" class into larger, more coherent groups, thus facilitating identification of the mechanisms for activity.

Natural products play a significant role in cancer chemotherapy. They are likely to provide many lead structures, which can be used as templates for the construction of novel drugs with enhanced antitumor activity. Traditional research approaches studied structure-activity relationship of natural products and obtained key structural properties, such as chemical bond or group, with the purpose of ascertaining their effect on a single cell line or a single tissue type. Here, for the first time, we develop a machine learning method to comprehensively predict natural products responses against a panel of cancer cell lines based on both the gene expression and the chemical properties of natural products. The results on two datasets, training set and independent test set, show that this proposed method yields significantly better prediction accuracy. In addition, we also demonstrate the predictive power of our proposed method by modeling the cancer cell sensitivity to two natural products, Curcumin and Resveratrol, which indicate that our method can effectively predict the response of cancer cell lines to these two natural products. Taken together, the method will facilitate the identification of natural products as cancer therapies and the development of precision medicine by linking the features of patient genomes to natural product sensitivity.

QSAR is defined by its embrace of quantitative modelling to build relationships between a molecule's structure and its properties. It was brought into being by Corwin Hansch's application of linear regression to biological activity data. Restricted to substituent constants extracted from experiments and hand-crafted into tables, graphs and simple regression models, early QSAR was very limited in scope but had the advantage that the practitioner became intimately connected to the data and the patterns that emerged from an analysis could be fitted into a coherent chemical explanation. There has been a profound shift in the capabilities available to us as QSAR modellers, particularly in the expansion of readily calculated chemical descriptors and model building methods, but at the heart of it remains the question of whether we are seeking models that the human expert can understand and interpret, or models that are useful, validated and predictive? Thischapter describes a framework for understanding the model building options available in the context of drug discovery decision making and points to more recent work that seeks to identify more rigorous quantifiable metrics for assessing performance of data mining techniques in QSAR modelling. The chapter highlights common modelling methods used in QSAR, with examples, and also covers more recent ensemble and meta-modelling developments.

Indian Plant Anticancer Compounds Database (InPACdb) is a web-based open access database of phytochemicals. The objective of this initiative is to project the potential of anticancer phytochemicals from Indian pharmacopoeia in an integrated environment. This database is unique in providing comprehensive information covering cancer type, molecular target, 3D Stereochemical structures (tautomers, stereoisomers, conformers and resonance structures) and Chemical descriptors etc. for each entry, enabling effective cheminformatics analysis. The complete dataset of InPACdb encompasses 32 descriptive fields for each entry, and is freely available for download at http://www.inpacdb.org.

From the perspective of 2D chemical descriptors, error in docking activity predictions is separated into noise and systematic components. This error framework explains how fitting docking scores to a 2D-QSAR equation often improves accuracy as well as its logical limits. Intriguingly, in examined cases where multiple docking models (e.g., multiple crystal structures or multiple scoring functions) are available for an enzyme, the noise component of error dominates the difference between the more accurate and less accurate docking models. When this is true, the QSAR equation fit statistics can rank each docking-score set's accuracy in the absence of experimental activity data.

Quantitative structure–activity relationship (QSAR) problems do not have, in general, linear solutions, and the problem is how to model those situations. Another consideration is that the nonlinear model should not be assumed but should emerge from data analysis. This chapter integrates the best models individually developed for each endpoint into a hybrid system for that endpoint. This has to be flexible to accept further inputs or modules, if available. Whereas inputs to the basic models are the chemical descriptors, input to the hybrid model are the n values predicted for each molecule by the n integrated models; the output is always the toxicity for that molecule. The basic theory behind the combinations, as well as the models obtained is illustrated.

Using the computer system PASS (prediction of activity spectra for substances), which predicts simultaneously several hundreds of biological activities, a training set for discriminating between drugs and nondrugs is created. For the training set, two subsets of databases of drugs and nondrugs (a subset of the World Drug Index, WDI, vs the Available Chemicals Directory, ACD) are used. The high value of prediction accuracy shows that the chemical descriptors and algorithms used in PASS provide highly robust structure-activity relationships and reliable predictions. Compared to other methods applied in this field, the direct benchmark undertaken with this paper showed that the results obtained with PASS are in good accordance with these approaches. In addition, it has been shown that the more specific drug information used in the training set of PASS, the more specific discrimination between drug and nondrug can be obtained.

Eight metals Zn, Mn, Mg, Cu, Al, Ca, Ba and K selected as chemical descriptors were analysed by inductively coupled plasma atomic emission spectrometry. The metal profile was studied in 48 samples of commercial tea - 26 black, 18 green, and four Oolong teas. Pattern recognition procedures such as display methods (Principal Component analysis and Cluster analysis) and supervised techniques (Linear Discriminant analysis and Artificial Neural Networks trained by back propagation) were applied to attain the differentiation of the three tea classes.

In this paper a new continuous variable called core-ratio is deflned to describe the probability for a residue to be in a binding site, thereby replacing the previous binary description of the interface residue using 0 and 1. So we can use the support vector machine regression method to flt the core-ratio value and predict the protein binding sites. We also design a new group of physical and chemical descriptors to characterize the binding sites. The new descriptors are more efiective, with an averaging procedure used. Our test shows that much better prediction results can be obtained by the support vector regression (SVR) method than by the support vector classiflcation method.

The development of non-testing approaches for the identification of chemicals with the potential to cause environmental hazards, have gained increased attention. In this study the hypothesis that experts’ judgement of chemical reactivity could be used for this purpose has been tested. The judgments of chemical reactivity for 200 organic chemicals, based on their molecular structure were performed by four senior organic and environmental chemists. Potentials for reactivity (namely: oxidative/reductive reactivity, hydrolysis, direct photolysis, reactivity towards radicals) were scored on the 1– scale. Both the experts’ individual judgment, and a compromised judgment after joint discussion for clarification of cases with divergent opinions, was analysed. Furthermore we searched for relations between the expert judgement data and: (a) the chemical characteristics representedby 40 chemical descriptors using partial least squares regression (PLS), and (b) experimental and in silico data for different toxicological and physico-chemical end-points. The results of on-going analyses indicate that for the majority of the chemicals, the individual judgments differed significantly between experts but after discussion among the experts for clarification, the judgments becamemuch more concurrent. The relation between averaged experts’ predicted photolytic degradation potential and chemical characteristic based on a PLS model (with three significant components explaining 76% of the variation) was found. We have also found that in the group of the chemicals with the higher score of the experts judged oxidative reactivity there was higher percentage of biodegradable chemicals when compar

Increasing age is a risk factor for many diseases; therefore developing pharmacological interventions that slow down ageing and consequently postpone the onset of many age‐related diseases is highly desirable. In this work we analyse data from the DrugAge database, which contains chemical compounds and their effect on the lifespan of model organisms. Predictive models were built using the machine learning method random forests to predict whether or not a chemical compound will increase Caenorhabditis elegans’ lifespan, using as features Gene Ontology (GO) terms annotated for proteins targeted by the compounds and chemical descriptors calculated from each compound’s chemical structure. The model with the best predictive accuracy used both biological and chemical features, achieving a prediction accuracy of 80%. The top 20 most important GO terms include those related to mitochondrial processes, to enzymatic and immunological processes, and terms related to metabolic and transport processes. We applied our best model to predict compounds which are more likely to increase C. elegans’ lifespan in the DGIdb database, where the effect of the compounds on an organism’s lifespan is unknown. The top hit compounds can be broadly divided into four groups: compounds affecting mitochondria, compounds for cancer treatment, anti‐inflammatories, and compounds for gonadotropin‐ releasing hormone therapies.

There is an urgent need to establish a fundamental understanding of the mechanisms of nanomaterial (NM) interaction with living systems and the environment, in order for regulation of NMs to keep pace with their increasing industrial application. Identification of critical properties (physicochemical descriptors) that confer the ability to induce harm in biological systems is crucial, enabling both prediction of impacts from related NMs (via quantitative nanostructure–activity relationships (QNARs) and read-across approaches) and development of strategies to ensure these features are avoided or minimised in NM production in the future (“safety by design”). A number of challenges to successful implementation of such a strategy exist, including: (i) the lack of widely available systematically varied libraries of NMs to enable generation of sufficiently robust datasets for development and validation of QNARs; (ii) the fact that many physicochemical properties of pristine NMs are inter-related and thus cannot be varied systematically in isolation from others (e.g. increasing surface charge may impact on hydrophobicity, or changing the shape of a NM may introduce defects or alter the atomic configuration of the surface); and (iii) the effect of ageing, transformation and biomolecule coating of NMs under environmental or biological conditions. A novel approach to identify interlinked physicochemical properties, and on this basis identify overarching descriptors (axes or principle components) which can be used to correlate with toxicity is proposed. An example of the approach is provided, using three principle components which we suggest can be utilised to fully describe each NM, these being the intrinsic (inherent) properties of the NM, composition (which we propose as a separate parameter) and extrinsic properties (interaction with media, molecular coronas etc.).