Computational approaches for multi-OMICs data analysis towards rational drug design

Abstract

In recent years, a huge amount of OMICs data has been collected from different biological systems using advanced high-throughput technology such as next-generation sequencing. This comprehensive information which includes genomics, proteomics, transcriptomics, and metagenomics encourage scientists to use this priceless information to develop novel drugs. Processing and analyzing of this huge amount of information is a daunting task and using an optimized computational approach is necessary to overcome this complexity. In this dissertation, different computational approaches have been used to process OMICs data to find a solution for rational drug design. Firstly, a hierarchical virtual screening pipeline has been developed to discover novel compounds to combat bacteria. Here, a pharmacophore analysis has been performed to select those compounds which have more chance to bind to the target protein. Finally, the compounds have been ranked based on their docking score and the top 11 compounds have been selected for further experiments. The results show two of these compounds can be used as a poetical antibiotic against Helicobacter pylori. Also, a new machine learning approach has been developed to predict drug-target interaction. Here, human single nucleotide polymorphisms (SNPs) data have been incorporated into the drug side effects. This data has been used as the input of a novel neural network model. 10-fold cross-validation results show this model can predict the interaction between drug and protein with AUC of 0.93. The proposed model can be used in polypharmacology and drug repositioning projects to discover novel protein-drugs interactions. Moreover, a multi-channel convolutional deep learning approach has been developed to predict disease-associated mutations in metal binding sites of proteins. The 3D structures of the metal binding sites have been converted to energy based affinity grid maps. These grid maps have been used to train the model besides sequential features. The results show this new model can predict those mutations which are associated with diseases with AUC of 0.9. This model can be widely used in metallodrug discovery project. Furthermore, bacteria which live in our body play important role in human health. A de novo pipeline has been developed to process whole metagenome samples to discover novel metagenomics biomarkers. This pipeline has been used to discover colorectal cancer (CRC) specific biomarkers. The results show CRC samples can be distinguished from healthy ones with AUC of 0.86. This pipeline can be used to discover bacterial conserved domains which play important role in human health. Finally, a pipeline has been developed to quantify the abundance of bacterial secondary metabolites from whole metagenome samples. Bacterial secondary metabolites are small compounds which are secreted by bacteria and potentially can interact with host biological pathways. The pipeline has been applied to whole metagenome samples that are suffered from CRC. The results show lipopolysaccharide and siderophores are more abundant in individuals with CRC than healthy ones. This pipeline can be used in host-pathogen interaction study. Also, it can be helpful in drug design projects by finding those bacterial secondary metabolites which promote human health.