I’m seeing persistent monoallelic expression at about 20–30% of heterozygous sites in a 10x Genomics PBMC dataset (about 6,200 cells on NovaSeq S4, processed with STARsolo), and I’m trying to parse true imprinting/ASE from alignment bias and UMI dropout. Has anyone validated a statistical filter or simulation framework that separates biological allelic imbalance from technical artifacts without locus-specific PCR, and if so which metrics (e.g., per-gene allelic coverage, read position bias) best predict replicability across runs?