vaRHC: an R package for semi-automation of variant classification in hereditary cancer genes according to ACMG/AMP and gene-specific ClinGen guidelines

Abstract

Motivation: Germline variant classification allows accurate genetic diagnosis and risk assessment. However, it is a tedious iterative process integrating information from several sources and types of evidence. It should follow gene-specific (if available) or general updated international guidelines. Thus, it is the main burden of the incorporation of next-generation sequencing into the clinical setting.

Results: We created the vaRiants in HC (vaRHC) R package to assist the process of variant classification in hereditary cancer by: (i) collecting information from diverse databases; (ii) assigning or denying different types of evidence according to updated American College of Molecular Genetics and Genomics/Association of Molecular Pathologist gene-specific criteria for ATM, CDH1, CHEK2, MLH1, MSH2, MSH6, PMS2, PTEN, and TP53 and general criteria for other genes; (iii) providing an automated classification of variants using a Bayesian metastructure and considering CanVIG-UK recommendations; and (iv) optionally printing the output to an .xlsx file. A validation using 659 classified variants demonstrated the robustness of vaRHC, presenting a better criteria assignment than Cancer SIGVAR, an available similar tool.

Availability and implementation: The source code can be consulted in the GitHub repository (https://github.com/emunte/vaRHC) Additionally, it will be submitted to CRAN soon.

Figure 1

varRHC package: main functions and workflow. vaR() contains three functions as follows: (A) vaRinfo: retrieves variant information from distinct databases; (B) vaRclass: combines information to assign or deny ACMG criteria returning a final classification of the variant; and (C) vaRreport: prints the results in a spreadsheet (.xlsx) file. vaRbatch() allows to do the process sequentially.

Figure 2

Tool performance assessed by gene and criteria. Stacked bar diagrams show the proportion of variants falling within each scenario (Table 1), labelled with a different colour (see legend in the figure). Each gene and criterion is evaluated separately (only fully or partially automated ones). Maximum number of variants analysed per gene is in parenthesis and corresponds to variants evaluated for BA1 and BS1. Since BA1 is a standalone criterion (incompatible with BS1), when BA1 was assigned by manual classification, the remaining criteria were not assessed in most variants and omitted from the comparison.

Figure 3

Flowchart showing the algorithm implemented in the vaRHC package to assign different PVS1 criterion strengths to LoF variants, based on ClinGen recommendations (Abou Tayoun et al. 2018).

Figure 4

Performance of Cancer-SIGVAR and vaRHC in comparison with the modified benchmark dataset. Stacked bar diagrams show the proportion of variants falling within each scenario (see Sections 3, 3.3, and Table 1), labelled with a different colour (see legend in the figure), for CDH1 and PTEN genes. The maximum number of variants analysed per gene is in parenthesis. This corresponds to variants evaluated for BA1 and BS1. Since BA1 is a standalone criterion (incompatible with BS1), when BA1 was assigned by manual classification, the remaining criteria were not assessed in most variants and omitted from the comparison.