The use of S and AAF metrics improves SARS-CoV-2 de novo iSNVs’ PHATE structure by mitigating sequencing center batch effects and artifacts. (A) PHATE visualization of the unfiltered de novo matrix containing 8 000 668 iSNVs, labelled by the libraries’ sequencing centers. (B) PHATE visualization as in (A), but with labels showing the percentage of k = 100 nearest neighbours that share the same sequencing center annotation as the library itself and the total PNN_SC value is displayed at the top left. (C and D) Boxplots displaying PNN_SC values for each PHATE visualization, derived from a sub-sampling controlled experiment across ten replicates (see Method section 2.5). (C) Shows PNN_SC values across various S metric thresholds, and (D) presents PNN_SC values across different AAF metric thresholds. (E) PHATE visualization of the de novo matrix, filtered based on S and AAF thresholds, labelled by sequencing centers. The arrow points to a set of libraries that seem to diverge from the main cluster. (F) PHATE visualization as in (E), but with labels showing PNN_SC values, and the total PNN_SC value is displayed at the top left. In this representation sequencing centers with at least 1000 libraries in our dataset are explicitly labelled with its sequencing center as follows: Welcome Sanger Institute (WSI), National Institute of Health DR. Ricardo Jorge (NIHRJ), Doherty Institute (DI), CDC-OAMD (CDC), Comenius University in Bratislava (CUB), Ravi Kant (RK), University of Tartu in Estonia (UTE), Chan Zuckerberg Biohub (CZB), Kwazulu-Natal Sequencing Platform (KSP), INAB Insitute in Certh (IIC), BROAD GCID (BROAD), Wales Specialist Virology Center (WSVR). While the remaining libraries were grouped under the ‘Other’ label.