Normalization for sampled count data

Genomics count normalization should stabilize technical variance, account for depth, and preserve within-cell abundance order. Common proportional-fitting and log workflows can retain depth dependence and conflate scaling with pseudocount choice. PFlog separates pseudocount choice from centered-log-ratio geometry, uniquely satisfies the three normalization desiderata, and outperformed alternatives across hundreds of single-cell RNA-seq datasets.