Sequence feature selection to solve biological questions related to variant analysis and Anti-SARS-CoV-2 siRNAs development.
data integration; variant analysis; pathogenicity; decision tree; siRNA; therapies; database; COVID-19.
Analysis of variants in clinical context and the support for the development of therapies against viral diseases are two areas which several research have used processes of integration and analysis of omics data. Assessing whether a given variant has a pathogenic impact is a challenge in the analysis of variants, especially when different tools for predicting pathogenicity point to divergent results. Regarding the development of RNA interference-based therapies, it is observed that there is a continuing need to design and evaluate the efficiency of new small-interfering RNAs (siRNAs) for each new virus that arises, like SARS-CoV-2, responsible for the COVID-19 pandemic. In this sense, it is argued in this thesis, based on the discussion of two works, that data integration and feature selection processes can contribute to the resolution of issues related to the identification of pathogenicity of variants and, in a second moment, to the availability of information and characteristics of sequences that may serve as the basis for therapies for COVID-19. In general terms, the study aimed (a) to develop data integration methods and selection of variant characteristics to measure pathogenicity and (b) to develop data integration methods for the construction of a database of siRNAs for SARS-CoV-2. To achieve the first objective, a decision tree-based classification model was proposed to estimate the pathogenicity of variants, built through an integration process of public data of already cataloged variants with pathogenicity predictions provided by machine learning-based tools. The model was able to present a higher accuracy than the state of the art regarding the prediction of pathogenicity of variants, constituting an important tool to support health professionals, such as in the diagnosis of genetic diseases. In the second objective, data on available properties, thermodynamics, toxicity, similarity, and efficiency were combined to assemble a global catalog of siRNAs for SARS-CoV-2. The integration of diverse properties related to siRNAs in a single consolidated database is an information reference that allows the realization of simple and targeted filtering in siRNA, saving the execution of many wet-lab tests on candidate molecules for COVID-19 antiviral therapies. These studies have common features with other data integration works in aspects involving data diversity, reproducibility, and knowledge discovery. Finally, it was found that these studies have potential for clinical application, either to increase the understanding of variants related to different genetic comorbidities, in the case of the first work, or to support the development of therapies against COVID-19, in the case of second job.