Computational frameworks to aid pharmacological studies : Tools, Databases and Prediction models

Show full item record

Title: Computational frameworks to aid pharmacological studies : Tools, Databases and Prediction models
Author: Ravikumar, Balaguru
Contributor organization: University of Helsinki, Faculty of Biological and Environmental Sciences
Doctoral Programme in Integrative Life Science
Helsingin yliopisto, bio- ja ympäristötieteellinen tiedekunta
Integroivien biotieteiden tohtoriohjelma
Helsingfors universitet, bio- och miljövetenskapliga fakulteten
Doktorandprogrammet i integrerande biovetenskap
Publisher: Helsingin yliopisto
Date: 2020-02-28
Language: eng
Thesis level: Doctoral dissertation (article-based)
Abstract: Rational approaches to the traditional drug discovery process rely on high-throughput and lowthroughput bioactivity or phenotypic screening studies. Such profiling strategies have been the foremost step utilized when identifying or curating lead molecules, and more recently, when evaluating compound repositioning/repurposing opportunities. Although promising, such systematic approaches are inherently limited by their cost and labor constraints, and the necessity of specific screening equipment’s renders them outside the scope of many academic laboratories. Further, elucidation of any novel associations through such experimental techniques require exhaustive searches of compound libraries. Hence, many positive drug indications revealed, though rational, are often by the virtue of serendipity. The accumulation and standardization of existing compound profiling datasets has paved the way to the field of chemoinformatics. Wherein, data-driven computational approaches are employed as a pragmatic solution to challenge the inherent notion of serendipity in screening studies. Such in silico models serve as an efficient and cost-effective augmentation to the experimental screening approaches, circumventing the intrinsic limitations of current drug discovery methods. This study therefore was motivated towards identifying niches and limitations prevalent in the current pharmacological paradigm, where an effective computational framework could be utilized to complement and expedite the conventional drug discovery process. The various computational approaches proposed in this article-based thesis include exploratory web-tools, new data resources, and prediction models, that are introduced as supplements to extend the current computer-aided drug discovery process (CADD). Firstly, I developed a web-application, termed C-SPADE, a novel compound-centric chemoinformatic tool that facilitates interactive analysis and visualization of compound screening experiments using Compound-SPecific bioActivity DEndrograms. The tool employs compound-compound similarity metrics to estimate the diversity amidst compounds profiled in the screening panel and intuitively represents the chemical similarity space and the observed bioactivity values. The web-tool provides users an exploratory framework to perform pharmacological analysis and investigate novel compound associations employing diverse compound similarity clusters. Secondly, to address the heterogeneous and non-standardized bioactivity data in existing data resources, a comprehensive open-data platform, called Drug Target Commons (DTC), was developed. DTC feature tools for data annotation, standardization, curation to address intra-resource heterogeneity and provide users a one-stop resource for drug discovery and in silico model development endeavours. User-specific applications of both the above-mentioned resources have been demonstrated through several case studies and experimental validations. In addition to the tools and databases, I have also designed and implemented diverse machine learning models primarily to predict potent compound-kinase interactions and to fill the current experimental gaps in large-scale activity profiling studies. Firstly, through collaborative efforts, we employed the Kronecker kernel-based regularized least square regression (KronRLS) algorithm under different crossvalidation settings to predict both the uncharacterised binding measures in large-scale profiling studies and novel compound-target associations. As a case study, the off-target profile of an investigational VEGF receptor inhibitor tivozanib was predicted and experimentally validated. Secondly, I designed and implemented an efficient statistical model utilizing an ensemble SVM classifier to prioritize potent compound-kinase association for biochemical testing. The developed computational framework was termed Virtual Kinome Profiler (VKP) and was efficiently used in compound repositioning and lead identification studies, wherein 19 novel kinase-compound interactions spanning across different kinases were predicted and experimentally validated. Apart from elucidating the chemogenomic similarities prevalent among distinct kinase proteins, VKP with a positive prediction value (PPV) of 84% was shown to reduce the time and cost constraints related to traditional experimental screening process. Most of the computational frameworks proposed in this thesis are designed and deployed with an accompanying web-based graphic user interface (GUI). This in turn aids in the translatability of the platforms, overcoming the current prerequisites required when utilizing CADD models, including prior expertise in data analysis and scripting languages. These studies together exemplify new applications of computational models in diverse areas of the drug discovery process, subsequently making invaluable augmentations to a chemical biologist’s toolbox.
Subject: Biotechnology
Health Sciences
Rights: Julkaisu on tekijänoikeussäännösten alainen. Teosta voi lukea ja tulostaa henkilökohtaista käyttöä varten. Käyttö kaupallisiin tarkoituksiin on kielletty.

Files in this item

Total number of downloads: Loading...

Files Size Format View
COMPUTAT.pdf 20.17Mb PDF View/Open

This item appears in the following Collection(s)

Show full item record