Journal: Nature communications
This study addresses the challenges of using whole genome sequencing (WGS) on formalin-fixed paraffin-embedded (FFPE) samples in clinical oncology. FFPE samples typically show lower data quality and increased artifacts compared to fresh frozen (FF) tissue.
Analysis of 56 matched FF and FFPE pairs revealed a 20-fold increase in artifactual mutations in FFPE samples, which compromises detection of key biomarkers like homologous recombination deficiency (HRD).
Key findings include:
- Consensus calling significantly reduces false structural variant calls but is insufficient for filtering single nucleotide variants (SNVs) and indels.
- The authors developed FFPErase, a machine learning tool that effectively filters these artifacts, achieving clinical-grade variant accuracy.
- FFPErase demonstrated 99% sensitivity compared to FDA-approved panel tests and identified 24% more clinically relevant findings.
Overall, FFPErase improves the reliability of WGS from FFPE samples in oncology practice.