FFixR: A Machine Learning Framework for Accurate Somatic Mutation Calling from FFPE RNA-Seq Data in Cancer
Livne, O.; Yizhak, K.
Show abstract
Formalin-fixed paraffin-embedded (FFPE) tissues are widely used in clinical and research settings, yet their use for detecting somatic mutations from RNA sequencing (RNA-seq) is hindered by artefactual mutations introduced by cytosine deamination and strand-specific damage. Existing FFPE noise-filtering tools are tailored to DNA-seq and rely on strand bias, rendering them unsuitable for RNA-seq. Here, we present FFixR, a machine learning-based framework that filters FFPE-induced artefacts from RNA-seq data without requiring matched-normal samples. Trained on FFPE melanoma samples with matched DNA, FFixR leverages allele-specific read counts, variant features, and mutational signature probabilities. FFixR removed up to 98% of artefactual mutations while maintaining [~]92% recall of true variants. SHAP analysis revealed key feature interactions guiding model decisions. When applied to an independent cohort, FFixR restored the correlation between RNA- and DNA-derived tumor mutational burden (R{superscript 2} = 0.881) and recovered biologically meaningful mutational signatures. FFixR enables accurate somatic variant calling from FFPE RNA-seq data, expanding the utility of archival samples for research and clinical applications.
Matching journals
The top 9 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
Similar papers in this journal
- Interpretable deep learning for chromatin-informed inference of transcriptional programs driven by somatic alterations across cancers 95%
- Methyl-CODEC enables simultaneous methylation and duplex sequencing 94%
- OmicsFootPrint: a framework to integrate and interpret multi-omics data using circular images and deep neural networks 94%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.