Ideafix: a decision tree-based method for the refinement of variants in FFPE DNA sequencing data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Next-generation sequencing (NGS) of formalin-fixed and paraffin-embedded (FFPE) biopsies is crucial for cancer treatment decisions.
• FFPE samples are susceptible to sequence artefacts that can complicate accurate variant interpretation.
• Identifying and mitigating these artefacts is essential for reliable molecular testing.

Purpose of the Study:

• To develop and validate a machine learning-based algorithm for identifying sequence artefacts in FFPE samples.
• To improve the accuracy of variant calling from FFPE-derived NGS data.
• To provide a reliable tool for molecular testing in oncology.

Main Methods:

• A machine learning algorithm was designed using over 1.6 million variants from 27 paired FFPE and fresh-frozen breast cancer samples.
• Variant features were assembled and evaluated using five machine learning algorithms, including XGBoost and random forest.
• Leave-one-sample-out cross-validation and independent datasets were used for performance testing.

Main Results:

• XGBoost and random forest algorithms achieved an area under the receiver operating characteristic curve (AUC) >0.86 in cross-validation.
• Independent dataset validation yielded AUC values of 0.96, outperforming previously published tools (max AUC 0.92).
• Key discriminating features included read pair orientation bias, genomic context, and variant allele frequency.

Conclusions:

• The developed machine learning algorithm effectively identifies sequence artefacts in FFPE samples.
• This approach shows significant promise for enhancing the reliability of molecular testing using FFPE biopsies.
• The algorithm is available as an R package named Ideafix (DEAmination FIXing).