Health Data Science / 2021 / Article / Fig 3

Research Article

Machine Learning Highlights Downtrending of COVID-19 Patients with a Distinct Laboratory Profile

Figure 3

Unified Manifold Approximation and Projection (UMAP) analysis of the laboratory profiles associated with the RT-PCR SARS-CoV-2-positive and SARS-CoV-2-negative testing results during March, April, May, and June combined (a), as well as separately in March (b), April (c), May (d), and June (e). Blue and red dots represent positive and negative RT-PCR results, respectively. The black circle depicts the high-density positive RT-PCR region. The singleton cluster on the right of the UMAP embedding includes 105 patients with 90% feature values missing in their profile vectors. Those missing values are imputed as the overall mean of each feature, which makes those profiles almost identical to each other. Since UMAP preserves the pairwise similarity during the mapping process, these vectors are mapped to a tiny crowd, which was excluded from our next analysis. Percentage of positive RT-PCR within and outside the circle and percentage of negative RT-PCR within and outside the circle are shown in the table, respectively.