Detecting Contamination in Cell-free DNA Sequencing Libraries using Germline Markers
Pillar developed a panel-agnostic Bayesian method to detect sample-to-sample contamination within cell-free DNA (cfDNA) sequencing batches, where variant allele fractions below 0.5% make false positive calls especially costly. The algorithm genotypes germline single nucleotide polymorphisms (SNPs) across a batch, flags sites with discordant allele fractions, and computes the posterior probability that observed read counts reflect contamination rather than stochastic noise. Across in silico mixes, contrived cell line mixes, and CNV-positive tumor samples it detected contamination down to 1% using 30 homozygous SNPs while holding 99–100% specificity.
