Impact of Clinical and Genomic Factors on SARS-CoV2 Disease Severity

Abstract

The SARS-CoV2 virus behind the COVID-19 pandemic is manifesting itself in different ways among infected people. While many are experiencing mild flue-like symptoms or are even remaining asymptomatic after infection, the virus has also led to serious complications, overloading ICUs while claiming more than 2.6 million lives world-wide. In this work, we apply AI methods to better understand factors that drive the severity of the disease. From the UK BioBank dataset we analyzed both clinical and genomic data of patients infected by this virus. Leveraging positiveunlabeled machine learning algorithms coupled with RubricOE, a state-of-the-art genomic analysis framework forgenomic feature extraction, we propose severity prediction algorithms with high F1 score. Furthermore, we extracted insights on clinical and genomic factors driving the severity prediction. We also report on how these factors have evolved during the pandemic w.r.t. significant events such as the emergence of the B.1.1.7 SARS-CoV2 virus strain.

Epidemiological topology data analysis links severe COVID-19 to RAAS and hyperlipidemia associated metabolic syndrome conditions

“…Our LR predicting severe C19 from the C19 patients subset in the C19 cohort while pooling ARB and ACE inhibitor RAAS drugs indicated an OR = 1.005 (95% CI: 0.97–1.04, P = .793) ( Fig. 2g ), consistent with other reports that HL is unimportant to severe C19 ( Atmosudigdo et al 2021 , Dey et al 2021 , Kristan et al 2021 , Ryan et al 2021 ). To understand the signals RTDA detected, we performed LRs stratified by HL ( Fig.…”

“…HL was found protective against mortality, with statins (STAT) and fibrates suggested as possible reducers of lipids-induced inflammation. An increased risk of severe C19 with HL in large cohorts and meta studies was also observed ( Atmosudigdo et al 2021 , Dey et al 2021 ).…”

Human-centered explainability for life sciences, healthcare, and medical informatics

“…As an example, we applied SP-LIME to extract insights from an algorithm that predicted the severity of coronavirus disease (COVID-19) based on both clinical and genomic data of patients in a dataset extracted from the UK Biobank (see Figure 3 A). 65

Figure 3 Applications of four popular XAI methods (A) LIME optimizes the local faithfulness and complexity of explanation. It has two versions to find both local and global explanations, which we applied on a COVID-19 longitudinal dataset to represent the clinico-genomic factors associated with COVID-19 severity.

…”

“…It has two versions to find both local and global explanations, which we applied on a COVID-19 longitudinal dataset to represent the clinico-genomic factors associated with COVID-19 severity. 65 Local/global importance of single nucleotide polymorphisms, indicated by their chromosomic location and clinical variables relative to the patient outcome, are shown in green or red for positive or negative association, respectively. (B) SHAP is a more generalized version of six linear-based explainable models using Shapely regression values.…”