LoLoPicker An Improved Method Of Detecting Low-Fraction Variants From Cancer

Jian CARROT-ZHANG, McGill University, Canada

1 Department of Human Genetics, McGill University
2 Genome Quebec Innovation Centre

The detection of tumor-only mutations remains challenging. One of the major complexities is that low-fraction variants (LFVs) are commonly observed in tumor samples, owing to normal-tissue contamination and cancer heterogeneity. Failing to remove unknown errors can significantly affect the specificity of variant calling, especially in calling variants at low-fraction, since most of the random or systematic errors are not frequent. The situation is even worse in calling LFVs from formalin fixed paraffin embedded (FFPE) samples, since error rates in these samples are much higher than high-quality samples.
WES has emerged as a promising tool to discover disease-causing genes. For many basic research or clinical laboratories, the number of samples being sequenced has increased dramatically. Some laboratories build their in-house database to enable them filtering out false-positive calls that are specific to similar protocols, instruments and environmental factors. Such database provides an opportunity to precisely estimate site-specific error rates from control samples, and gives the advantage to increase the sensitivity of calling LFVs on sites with lower error rates, and reduce false positives on sites with high error rates.
No existing software is sufficient to call variants on the whole-exome level, with high sensitivity and low false-positive rate. Here we present LoLoPicker, a tool dedicated to call LFVs from WES data using tumor and its matched normal tissue, plus a user-defined panel of control samples. We observed superior performance of LoLoPicker in comparison with MuTect and VarScan2. While LoLoPicker maintains highest sensitivity among other programs, the specificity of LoLoPicker is significantly improved. Our approach is particularly suited for FFPE samples, since FFPE-specific errors can be identified from a panel of FFPE controls. This tool will open doors for studying targeted therapy and drug resistance.
This algorithm is implemented in Python language and the package is released at