Fig. 1

Overview of the FIREVAT workflow. The top panel represents the conventional workflow for processing sequencing data of the tumor and matched normal genomes. Sequencing artifacts are introduced in various steps of the experimental process. After calling variants with software such as MuTect2 and Muse, manual filtering is performed to mitigate false positives. FIREVAT automates this variant refinement task by leveraging the COSMIC mutational signatures (version 3). The primary input parameters for FIREVAT include a VCF file and a configuration JSON file, which specifies instructions on how to compute the desired filter parameters. In the preprocessing step, FIREVAT derives these parameters for each point mutation. A set of candidate solutions is generated to help the optimization process converge faster. Next, FIREVAT searches for a set of parameters that maximizes the objective function (“Evaluate”) using GA, a gradient-free approach. At each iteration, the original set of mutations is divided into refined and artifactual groups, followed by a mutational signature analysis of each group. FIREVAT further analyzes the refinement outcomes by assessing strand bias and significant differences between the refined and artifactual variant groups. Each variant is also annotated using clinical databases. FIREVAT generates output files for the following items: HTML report, optimization logs, and VCFs for refined, and artifactual variants