Strategies of modeling and feature combination for the study of COVID-19 and its comorbidities

Author(s)

Juan Carlo Santos e Silva

• Document type: Doctoral Thesis
• Press: São Paulo.
• Institution: Universidade de São Paulo (USP). Conjunto das Químicas (IQ e FCF) (CQ/DBDCQ)
• Defense date: 2024-08-27

Examining board members

Helder Takashi Imoto Nakaya; André Fujita

• Advisor: Helder Takashi Imoto Nakaya

Abstract

The comorbidities presented by patients hospitalized with pulmonary complications can worsen the clinical picture and health conditions. This study employs different integrative biology strategies to analyze multiple variables in the context of COVID-19, highlighting comorbidities in the outcome of the disease. To this end, we conducted: stratification strategies for the group of patients infected with SARS-CoV-2; transcriptome analyses; selection of variables by relative importance; and construction of machine learning models using feature combination. The analysis of the lung transcriptome of deceased COVID-19 patients revealed significant changes in gene expression. The genes LCK and EGR2 were exclusively upregulated in obese patients and positively correlated with body mass index, indicating an interaction between metabolism and immunity with an emphasis on diabetes and obesity. The approach of variable combinations accompanied by patient stratification increased model performance by at least 20%, indicating that evaluating combinations can describe biological phenomena more accurately than individual variables. Additionally, the application of such techniques in pediatric epidemiological data elucidated determining factors in disease severity, considering symptoms and comorbidities. This work highlights the importance of considering multiple associations, even in case-control studies, as they can generate new perspectives on the problem presented. Consequently, these results reinforce the need for integrative approaches for effective understanding and intervention in infectious diseases, contributing to more precise and personalized prevention and treatment strategies. (AU)

• FAPESP's process: 19/27139-5 - Application of genomic tools for the study of epidemiological databases
• Grantee: Juan Carlo Santos e Silva
• Support Opportunities: Scholarships in Brazil - Doctorate (Direct)