A consideration of publication-derived immune-related associations in Coronavirus and related lung damaging diseases

Nophar Geifman; Anthony D Whetton

Our approach to capturing immune-related associations from MeSH descriptors is not without limitations. We first assume that co-occurrences of MeSH descriptors within a PubMed record represent a true relationship or dependency; further, some of the associations and co-occurrences were low (i.e. present in only few abstracts). However, our previous investigation into the extent to which MeSH term co-occurrence captures real association has found that at least 70% of co-occurrences of different types of entities (disease, cell type, or cytokine) represent true direct or indirect dependencies, but it is likely to be higher than that [11] . Additionally, patterns of MeSH co-occurrence have shown to capture known medical associations, as well as identify potentially novel ones, thus providing further confidence in the approach. A second limitation is a lack of directionality and type for the associations captured by approach; nevertheless, we show that these mere co-occurrences may still hold valuable information. Finally, since at the time of writing, very few publications directly related to COVID-19 are available, our data mining has had to focus on Coronavirus-and other related lung-damaging diseases as a proxy for COVID-19. Nevertheless, we have created a paradigm for such research which is easy to use and apply, and demonstrated its utility.