FFixR: a machine learning framework for accurate somatic mutation calling from FFPE RNA-seq data in cancer.
Summary
A novel machine learning framework, FFixR, has been developed to address the challenges of somatic mutation detection from RNA sequencing (RNA-seq) data derived from formalin-fixed paraffin-embedded (FFPE) tissues. These tissues are prone to introducing artifacts that hinder accurate variant identification. FFixR effectively filters these artefactual mutations without requiring matched-normal samples. The tool demonstrated the ability to remove up to 98% of artifacts while maintaining good recall for true variants. This advancement enables more reliable analysis of archived FFPE samples, expanding their potential for research and clinical applications.
Analysis
🟢 BIOMOL: This study introduces FFixR, a significant advancement for the molecular analysis of FFPE samples, which are abundant but challenging for RNA-seq due to DNA/RNA damage. The key discovery is FFixR's ability to distinguish true somatic mutations from FFPE-induced artifacts without requiring a matched-normal sample, which is crucial for retrospective cohorts. The technology employed is RNA-seq combined with machine learning, enabling the valorization of archived samples and expanding their use for biomarker detection or tumor characterization. 🔵 BIOINFO: FFixR is a novel machine learning framework specifically designed for artifact filtering in FFPE RNA-seq data. It was trained on melanoma samples with matched DNA and benchmarked on independent cohorts, demonstrating 98% artifact removal and 92% recall of true variants. The tool is open-source and available on GitHub, ensuring its reproducibility. Its clinical deployment could transform the workflow for FFPE sample analysis by enabling reliable somatic mutation detection, which is essential for tumor characterization and precision medicine.