Computational tools for designing therapeutic molecules against virulent factors of pathogens

Abstract

Microbes are minute, unicellular, multicellular organisms, such as bacteria, algae, fungi, viruses and protozoans, that can be only visible through the microscope. These can be infectious as well as non-infectious in nature. The ability of the infectious agent to cause diseases in the host cell is known as pathogenicity, while the ability of the pathogen to infect or cause damage to the host tissues is determined by the virulence factors. According to Centers for Diseases Control and Prevention reports, various infectious diseases such as COVID-19, tuberculosis, measles, and influenza are responsible for causing morbidity in the human population. Microbial pathogens pose an alarming threat to the healthcare sector worldwide. These micro-organisms cause severe diseases that lead to high mortality and morbidity rate. The rise in re-emergence of life-threatening infectious diseases and increasing incidence of antibiotic-resistant strains of the pathogens poses a danger to the healthcare sector. In the past, several studies reported that many distinct pathogenic micro-organisms share the common mechanisms of causing the infections to the host cell. Virulence factors of these pathogens play an important role in host-pathogen interactions and disease-mechanism. These factors include invasion, colonization and damage to the host cell, which contribute to pathogenicity. Thus, virulence factors are major drug/vaccine targets for designing therapeutic molecules against these pathogens. Some pathogenic micro-organisms release several molecules which cause damage to the host cell, induce the infection and evoke the diseases. These molecules include toxins released by certain pathogens to cause toxicity and induce allergic reactions in the host cell. Advances in various technologies led to the explosive growth of experimentally verified proteomic data related to virulence factors, which is available in the form of repositories. Thus, the present thesis focuses on utilising the publicly available experimentally verified data to develop computational tools to identify the potential virulence factors and pathogenicity associated with pathogens, such as toxicity and allergenicity, and to design the therapeutic molecules against them. Taking this into consideration, we aimed to develop in-silico models to explore, predict and identify the potential virulence factors of pathogens. We build a machine learning-based method named ‘VirFacPred’ to identify novel virulence factors to aid the clinicians and scientific community. The best performing model achieved the maximum area under the receiver operating characteristic curve 0.97 with Matthews correlation coefficient 0.77 on the dataset. The best machine learning models have been implemented in the web server, which allows the prediction, designing, mapping and motif search for the virulent proteins of the pathogens. To address the pathogenicity caused by the pathogenic organisms to the host cell, such as toxicity and allergy, we have developed “ToxinPred2” and “AlgPred 2.0” that will facilitate the identification of toxic and allergic proteins. Toxins are one of the major virulence factors that play a crucial role in damaging the host cell. We have developed a highly accurate method, ToxinPred2, for predicting toxins with better precision. We have integrated a hybrid method that combines three approaches, i.e., similarity-, motif-, and composition-based machine learning model, which achieved a maximum area under the receiver operating characteristic curve around 0.99 with Matthews correlation coefficient 0.91 on the dataset. We have provided the standalone version of the method, which can be freely accessed at GitHub. This is a general method developed for predicting the toxicity of proteins regardless of their source of origin. On the other hand, a method called “AlgPred 2.0” has been developed for identifying allergenic proteins with high accuracy that allows the prediction of allergens, designing of non-allergenic proteins, mapping of IgE epitope, motif search and BLAST search. The ensemble approach, i.e., similarity-, motif-, and composition-based machine learning model, has been used for predicting allergenic protein by combining prediction scores. The best model achieved maximum performance in terms of area under the receiver operating characteristic curve 0.98 with Matthews correlation coefficient 0.85 on the dataset. Besides proteins and peptides, some chemical compounds are known to induce allergic reactions to the host cell, known as chemical allergy. A therapeutic molecule may cause side effects due to its allergic potential. A first attempt has been made to develop the method using machine learning techniques that can predict the allergenic potential of chemicals. To aid the scientific community, we developed a novel method named “ChAlPred” that allows to predict and design the chemicals with allergenic properties. Our fingerprint-based analysis suggests that certain chemical fingerprints such as PubChemFP129 and GraphFP1014 are abundant in allergic compounds. We have also identified the FDA-approved drugs causing allergic symptoms (e.g., Cefuroxime, Spironolactone, Tioconazole) using our best model incorporated in the web server. In summary, attempts have been made to develop in-silico models that can be used to design directly/indirectly therapeutic molecules against disease-causing agents. To facilitate the scientific community, web-based services and standalone software have been developed.