Predicting Formalin-fixed Paraffin-embedded (FFPE) Sequencing Artefacts from Breast Cancer Exome Sequencing Data using Machine Learning

Cancer is a lethal disease caused by the accumulation of genomic alterations. In order to investigate these diseases, cancer genome studies have largely relied on fresh frozen tissues. However, fresh frozen material is often unavailable for large-scale studies. In contrast, formalin-fixed paraffin-embedded (FFPE) tissues are abundantly available but difficult to use due to severe DNA degradation resulting in artefactual mutations. Accurate removal of these sequencing artefacts can help enable the widespread use of FFPE material for molecular analysis of cancers and other diseases. In this project, I used commercial FFPE DNA repair and library preparation kits and trained a random forest classifier to reduce the number of sequencing artefacts in FFPE breast cancer exome data. These experiments show that the classifier is able to retain the majority of true somatic variants while filtering many FFPE-induced sequencing artefacts and is suitable for application to low-quality FFPE breast cancer exomes.M.A.S