ECOLE: Learning to call copy number variants on whole exome sequencing data

Abstract

Copy number variants (CNV) are shown to contribute to the etiology of several genetic disorders. Accurate detection of CNVs on whole exome sequencing (WES) data has been a long sought after goal for use in clinic. This was not possible despite recent improvements in performance because algorithms mostly suffer from low precision and even lower recall on expert-curated gold standard call sets. Here, we present a deep learning-based somatic and germline CNV caller for WES data, named ECOLE. Based on a variant of the transformer architecture, the model learns to call CNVs per exon, using high confidence calls made on matched WGS samples. We further train and fine-tune the model with a small set of expert calls via transfer learning. We show that ECOLE is able mimic the expert labeling for the first time with 68.7% precision and 49.6% recall. This corresponds to precision and recall improvements of 18.7% and 30.8% over the next best performing methods, respectively. ECOLE is the first method to achieve high precision and recall in mimicking human expert CNV calling. We also show that same fine tuning strategy using tumor samples enables ECOLE to detect RT-qPCR validated variations in bladder cancer samples without the need for a control sample. We think these features of ECOLE make CNV calling on WES data feasible for clinical use. ECOLE is available at https://github.com/ciceklab/ECOLE.

A method to comprehensively identify germline SNVs, INDELs and CNVs from whole exome sequencing data of BRCA1/2 negative breast cancer patients

“…At the same time, enhancing the pipeline’s ability to detect CNVs (recall 60.53%, precision 72.72%, F1 score 66.07%), particularly by improving recall, is a pivotal area for future development. Strategies to achieve this may include refining existing detection algorithms, integrating additional CNV-specific quality metrics, and leveraging advanced computational techniques to better interpret complex genomic regions ( 44 ). Of note, CNV callers tends to be more challenging both because: (i) these variants are more difficult to accurately detect using short read sequencing data, which makes structural variants calling more error-prone than small variants calling, (ii) the precise breakpoints for CNVs are not always well defined, which makes comparison between call-sets more complex ( 45 ).…”