MP4: a machine learning based classification tool for prediction and functional annotation of pathogenic proteins from metagenomic and genomic datasets.

Bacteria can exceptionally evolve and develop pathogenic features making it crucial to determine novel pathogenic proteins for specific therapeutic interventions. Therefore, we have developed a machine-learning tool that predicts and functionally classifies pathogenic proteins into their respective pathogenic classes. Through construction of pathogenic proteins database and optimization of ML algorithms, Support Vector Machine was selected for the model construction. The developed SVM classifier yielded an accuracy of 81.72% on the blind-dataset and classified the proteins into three classes: Non-pathogenic proteins (Class-1), Antibiotic Resistance Proteins and Toxins (Class-2), and Secretory System Associated and capsular proteins (Class-3). The classifier provided an accuracy of 79% on real dataset-1, and 72% on real dataset-2. Based on the probability of prediction, users can estimate the pathogenicity and annotation of proteins under scrutiny. Tool will provide accurate prediction of pathogenic proteins in genomic and metagenomic datasets providing leads for experimental validations. Tool is available at: http://metagenomics.iiserb.ac.in/mp4 .

Molib: A machine learning based classification tool for the prediction of biofilm inhibitory molecules.

Identification of biofilm inhibitory small molecules appears promising for therapeutic intervention against biofilm-forming bacteria. However, the experimental identification of such molecules is a time-consuming task, and thus, the computational approaches emerge as promising alternatives. We developed the 'Molib' tool to predict the biofilm inhibitory activity of small molecules. We curated a training dataset of biofilm inhibitory molecules, and the structural and chemical features were used for feature selection, followed by algorithms optimization and building of machine learning-based classification models. On five-fold cross validation, Random Forest-based descriptor, fingerprint and hybrid classification models showed accuracies of 0.93, 0.88 and 0.90, respectively. The performances of all models were evaluated on two different validation datasets including biofilm inhibitory and non-inhibitory molecules, attesting to its accuracy (≥ 0.90). The Molib web server would serve as a highly useful and reliable tool for the prediction of biofilm inhibitory activity of small molecules.

Mechanistic and structural insight into promiscuity based metabolism of cardiac drug digoxin by gut microbial enzyme.

The recent advances in microbiome studies have revealed the role of gut microbiota in altering the pharmacological properties of oral drugs, which contributes to patient-response variation and undesired effect of the drug molecule. These studies are essential to guide us for achieving the desired efficacy and pharmacological activity of the existing drug molecule or for discovering novel and more effective therapeutics. However, one of the main limitations is the lack of atomistic details on the binding and metabolism of these drug molecules by gut-microbial enzymes. Therefore, in this study, for a well-known and important FDA-approved cardiac glycoside drug, digoxin, we report the atomistic details and energy economics for its binding and metabolism by the Cgr2 protein of Eggerthella lenta DSM 2243. It was observed that the binding pocket of digoxin to Cgr2 primarily involved the negatively charged polar amino acids and a few non-polar hydrophobic residues. The drug digoxin was found to bind Cgr2 at the same binding site as that of fumarate, which is the proposed natural substrate. However, digoxin showed a much lower binding energy (17.75 ± 2 Kcal mol-1 ) than the binding energy (42.17 ± 2 Kcal mol-1 ) of fumarate. This study provides mechanistic insights into the structural and promiscuity-based metabolism of widely used cardiac drug digoxin and presents a methodology, which could be useful to confirm the promiscuity-based metabolism of other orally administrated drugs by gut microbial enzymes and also help in designing strategies for improving the efficacy of the drugs.

GutBug: A Tool for Prediction of Human Gut Bacteria Mediated Biotransformation of Biotic and Xenobiotic Molecules Using Machine Learning.

Dietary components and bioactive molecules present in functional foods and nutraceuticals provide various beneficial effects including modulation of host gut microbiome. These metabolites along with orally administered drugs can be potentially bio-transformed by gut microbiome, which can alter their bioavailability and intended biological or pharmacological activity resulting in individual or population-specific variation in drug and dietary responses. Experimental determination of microbiome-mediated metabolism of orally ingested molecules is difficult due to the enormous diversity and complexity of the gut microbiome. To address this problem, we developed "GutBug", a web-based resource that predicts all possible bacterial metabolic enzymes that can potentially biotransform xenobiotics and biotic molecules using a combination of machine learning, neural networks and chemoinformatic methods. Using 3,457 enzyme substrates for training and a curated database of 363,872 enzymes from ∼700 gut bacterial strains, GutBug can predict complete EC number of the bacterial enzymes involved in a biotransformation reaction of the given molecule along with the reaction centres with accuracies between 0.78 and 0.97 across different reaction classes. Validation of GutBug's performance using 27 molecules known to be biotransformed by human gut bacteria, including complex polysaccharides, flavonoids, and oral drugs further attests to GutBug's accuracy and utility. Thus, GutBug enhances our understanding of various metabolite-gut bacterial interactions and their resultant effects on the human host health across populations, which will find enormous applications in diet design and intervention, identification and administration of new prebiotics, development of nutraceutical products, and improvements in drug designing. GutBug is available at https://metabiosys.iiserb.ac.in/gutbug.

Discovery of a tetracyclic indole alkaloid that postpones fibrillation of hen egg white lysozyme protein.

Protein aggregation, such as amyloid fibril formation, is molecular hallmark of many neurodegenerative disorders including Alzheimer's, Parkinson's, and Prion disease. Indole alkaloids are well-known as the compounds having the ability to inhibit protein fibrillation. In this study, we experimentally and computationally have investigated the anti-amyloid property of a derivative of a synthesized tetracyclic indole alkaloid (TCIA), possessing capable functional groups. The fibrillation reaction of Hen White Egg Lysozyme (HEWL) was performed in absence and presence of the indole alkaloid. For quantitative analysis, we used Thioflovin T binding assay which showed ~50% reduction in fibril formation in the presence of 20 µM TCIA. Using TEM imaging, we observed a significant morphological change in our model protein in the presence of TCIA. In addition, we exploited FT-IR assay by which Amide I peak's shifting toward lower wavenumber was clearly observed. Using Molecular Docking, the interaction of the inhibitor (TCIA) with the protein's amyloidogenic region was modeled. Also, different biophysical parameters were calculated by Molecular Dynamics (MD) simulation. Various biochemical assays, conformational change, and hydrophobicity exposure of the protein during amyloid formation indicated that the compound assists HEWL to keep its native structure via destabilizing ß-sheet structure.

ToxiM: A Toxicity Prediction Tool for Small Molecules Developed Using Machine Learning and Chemoinformatics Approaches.

The experimental methods for the prediction of molecular toxicity are tedious and time-consuming tasks. Thus, the computational approaches could be used to develop alternative methods for toxicity prediction. We have developed a tool for the prediction of molecular toxicity along with the aqueous solubility and permeability of any molecule/metabolite. Using a comprehensive and curated set of toxin molecules as a training set, the different chemical and structural based features such as descriptors and fingerprints were exploited for feature selection, optimization and development of machine learning based classification and regression models. The compositional differences in the distribution of atoms were apparent between toxins and non-toxins, and hence, the molecular features were used for the classification and regression. On 10-fold cross-validation, the descriptor-based, fingerprint-based and hybrid-based classification models showed similar accuracy (93%) and Matthews's correlation coefficient (0.84). The performances of all the three models were comparable (Matthews's correlation coefficient = 0.84-0.87) on the blind dataset. In addition, the regression-based models using descriptors as input features were also compared and evaluated on the blind dataset. Random forest based regression model for the prediction of solubility performed better (R2 = 0.84) than the multi-linear regression (MLR) and partial least square regression (PLSR) models, whereas, the partial least squares based regression model for the prediction of permeability (caco-2) performed better (R2 = 0.68) in comparison to the random forest and MLR based regression models. The performance of final classification and regression models was evaluated using the two validation datasets including the known toxins and commonly used constituents of health products, which attests to its accuracy. The ToxiM web server would be a highly useful and reliable tool for the prediction of toxicity, solubility, and permeability of small molecules.