Fragment-based prediction of skin sensitization using recursive partitioning

Skin Doctor: Machine Learning Models for Skin Sensitization Prediction that Provide Estimates and Indicators of Prediction Reliability

The ability to predict the skin sensitization potential of small organic molecules is of high importance to the development and safe application of cosmetics, drugs and pesticides. One of the most widely accepted methods for predicting this hazard is the local lymph node assay (LLNA). The goal of this work was to develop in silico models for the prediction of the skin sensitization potential of small molecules that go beyond the state of the art, with larger LLNA data sets and, most importantly, a robust and intuitive definition of the applicability domain, paired with additional indicators of the reliability of predictions. We explored a large variety of molecular descriptors and fingerprints in combination with random forest and support vector machine classifiers. The most suitable models were tested on holdout data, on which they yielded competitive performance (Matthews correlation coefficients up to 0.52; accuracies up to 0.76; areas under the receiver operating characteristic curves up to 0.83). The most favorable models are available via a public web service that, in addition to predictions, provides assessments of the applicability domain and indicators of the reliability of the individual predictions.

Cheminformatics-driven discovery of polymeric micelle formulations for poorly soluble drugs

Many drug candidates fail therapeutic development because of poor aqueous solubility. We have conceived a computer-aided strategy to enable polymeric micelle-based delivery of poorly soluble drugs. We built models predicting both drug loading efficiency (LE) and loading capacity (LC) using novel descriptors of drug-polymer complexes. These models were employed for virtual screening of drug libraries, and eight drugs predicted to have either high LE and high LC or low LE and low LC were selected. Three putative positives, as well as three putative negative hits, were confirmed experimentally (implying 75% prediction accuracy). Fortuitously, simvastatin, a putative negative hit, was found to have the desired micelle solubility. Podophyllotoxin and simvastatin (LE of 95% and 87% and LC of 43% and 41%, respectively) were among the top five polymeric micelle-soluble compounds ever studied experimentally. The success of the strategy described herein suggests its broad utility for designing drug delivery systems.

QSAR models of human data can enrich or replace LLNA testing for human skin sensitization

Skin sensitization is a major environmental and occupational health hazard. Although many chemicals have been evaluated in humans, there have been no efforts to model these data to date. We have compiled, curated, analyzed, and compared the available human and LLNA data. Using these data, we have developed reliable computational models and applied them for virtual screening of chemical libraries to identify putative skin sensitizers. The overall concordance between murine LLNA and human skin sensitization responses for a set of 135 unique chemicals was low (R = 28-43%), although several chemical classes had high concordance. We have succeeded to develop predictive QSAR models of all available human data with the external correct classification rate of 71%. A consensus model integrating concordant QSAR predictions and LLNA results afforded a higher CCR of 82% but at the expense of the reduced external dataset coverage (52%). We used the developed QSAR models for virtual screening of CosIng database and identified 1061 putative skin sensitizers; for seventeen of these compounds, we found published evidence of their skin sensitization effects. Models reported herein provide more accurate alternative to LLNA testing for human skin sensitization assessment across diverse chemical data. In addition, they can also be used to guide the structural optimization of toxic compounds to reduce their skin sensitization potential.

Finding candidate drugs for hepatitis C based on chemical-chemical and chemical-protein interactions

Hepatitis C virus (HCV) is an infectious virus that can cause serious illnesses. Only a few drugs have been reported to effectively treat hepatitis C. To have greater diversity in drug choice and better treatment options, it is necessary to develop more drugs to treat the infection. However, it is time-consuming and expensive to discover candidate drugs using experimental methods, and computational methods may complement experimental approaches as a preliminary filtering process. This type of approach was proposed by using known chemical-chemical interactions to extract interactive compounds with three known drug compounds of HCV, and the probabilities of these drug compounds being able to treat hepatitis C were calculated using chemical-protein interactions between the interactive compounds and HCV target genes. Moreover, the randomization test and expectation-maximization (EM) algorithm were both employed to exclude false discoveries. Analysis of the selected compounds, including acyclovir and ganciclovir, indicated that some of these compounds had potential to treat the HCV. Hopefully, this proposed method could provide new insights into the discovery of candidate drugs for the treatment of HCV and other diseases.

Exploration of structural and physicochemical requirements and search of virtual hits for aminopeptidase N inhibitors

Aminopeptidase N (APN) inhibitors have been reported to be effective in treating of life threatening diseases including cancer. Validated ligand- and structure-based pharmacophore mapping approaches were combined with Bayesian modeling and recursive partitioning to identify structural and physicochemical requirements for highly active APN inhibitors. Based on the assumption that ligand- and structure-based pharmacophore models are complementary, the efficacy of 'multiple pharmacophore screening' for filtering true positive virtual hits was investigated. These multiple pharmacophore screening methods were utilized to search novel virtual hits for APN inhibition. The number of hits was refined and reduced by recursive partitioning, drug-likeliness, pharmacokinetic property prediction, and comparative molecular-docking studies. Four compounds were proposed as the potential virtual hits for APN enzyme inhibition.