15.1 Creating Low-Overlapping Aggregates of Cells
ArchR facilitates many integrative analyses that involve correlation of features. Performing these calculations with sparse single-cell data can lead to substantial noise in these correlative analyses. To circumvent this challenge, we adopted an approach introduced by Cicero to create low-overlapping aggregates of single cells prior to these analyses. We filter aggregates with greater than 80% overlap with any other aggregate in order to reduce bias. To improve the speed of this approach, we developed an implementation of an optimized iterative overlap checking routine and a implementation of fast feature correlations in C++ using the “Rcpp” package. These optimized methods are used in ArchR for calculating peak co-accessibility, peak-to-gene linkage, and for other linkage analyses. The use of these low-overlapping aggregates happens under the hood but we mention it here for clarity.