Determining a novel feature-space for SARS-CoV-2 sequence data

Abstract

The pandemicity & the ability of the SARS-COV-2 to reinfect a cured subject, among other damaging characteristics of it, took everybody by surprise. A global collaborative scientific effort was direly required to bring learned people from different niches of medicine & data science together. Such a platform was provided by COVID19 Virtual BioHackathon, organized from the 5th to the 11th of April, 2020, to ponder on the related pressing issues varying in their diversity from text mining to genomics. Under the "Machine learning" track, we determined optimal k-mer length for feature extraction, constructed continuous distributed representations for protein sequences to create phylogenetic trees in an alignment-free manner, and clustered predicted MHC class I and II binding affinity to aid in vaccine design. All the related work in available in a Github repository under an MIT license for future research.

<strong>Data Science in Unveiling COVID-19 Pathogenesis and Diagnosis: Evolutionary Origin to Drug Repurposing</strong>

“…Due to the low mutation rate and a high degree of similarity among SARS-COV-2 genome, very few studies have been performed using alignment-free methods. Correlation and partial information correlation (PIC) [57], optimal word (k-mer) to construct continuous distributed representations for protein sequences, to predict MHC class I and II binding affinity [58], combined k-mer and n-gram techniques [59], chemical properties (charge, hydropathy, side chain) of region-specific amino acid, mutations are few important features used during analysis [60,61]. These features are utilized in several studies attempting to identify the cluster of SARS-COV-2 in various ways, and its results demonstrated that this virus has multiple origins with different degrees.…”