Abstract 2722: Evaluation of in silico tools for variant classification in missense variants of solid cancer with actionable genetic targets
Abstract Increased use of next-generation sequencing has led to the discovery of many variants of uncertain significance, which are not clearly categorized as pathogenic or benign. In silico tools were developed to help classify these variants. However, the outcomes are different due to variations b...
Saved in:
| Published in: | Cancer research (Chicago, Ill.) Ill.), 2022-06, Vol.82 (12_Supplement), p.2722-2722 |
|---|---|
| Main Authors: | , , , , , , , |
| Format: | Article |
| Language: | eng |
| Online Access: | Get full text |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract
Increased use of next-generation sequencing has led to the discovery of many variants of uncertain significance, which are not clearly categorized as pathogenic or benign. In silico tools were developed to help classify these variants. However, the outcomes are different due to variations between the prediction algorithms. The purpose of this study is to evaluate the performance of 6 widely-used in silico tools in identifying drug-actionable gene variants as pathogenic or benign in 9 solid cancers.
We selected drug-actionable genes according to NCCN guidelines in 9 common solid cancers: breast, ovarian, colorectal, melanoma of skin, thyroid, bladder, pancreatic, prostate, and biliary. Although lung cancer is also a common cancer, given that it has many genetic markers that are already widely studied, we focused on other solid cancer types for our analysis. From these genes, we collected information on 642 total missense variants (pathogenic = 435, benign = 217). Each variant was determined to be benign or pathogenic based on assertions from three databases: ClinVar, OncoKB, and My Cancer Genome. We selected variants with at least 2 or more concordant databases and excluded variants with conflicting classifications. We evaluated the performance of the in silico tools (Polyphen-2, Align-GVGD, MutationTaster2021 (MT2021), CADD, CONDEL, and REVEL) using overall accuracy, sensitivity, specificity, Matthews correlation coefficient (MCC), and likelihood ratio (LR+, LR-).
All of the in silico tools demonstrated high sensitivity (0.78 - 1.00). However, excluding MT2021, all tools demonstrated low specificity (MT2021: 0.85, Other: 0.26 - 0.59). Overall accuracy ranged from moderate to high values across the tools (0.62 - 0.94). Besides MT2021, MCCs were found to be especially low (MT2021: 0.87, Other: 0.08 - 0.48) as well as LR+ (MT2021: 6.65, Other: 1.09 - 1.88). Most of the tools had high values of LR- (0.14 - 0.74) but MT2021 and CADD showed low LR- (0.06, 0.00). MT2021 showed the highest level of performance with the highest specificity (0.85), overall accuracy (0.94), MCC (0.87), and LR+ (6.65), a high sensitivity (0.99), and a low LR- (0.06). Conversely, Align-GVGD had the lowest level of performance with the lowest specificity (0.26), overall accuracy (0.62), MCC (0.08), and LR+ (1.09), and the highest LR- (0.74).
The results show that even widely-used tools have very different performance and limitations as diagnostic tools. The high sensitivity gro |
|---|---|
| ISSN: | 1538-7445 1538-7445 |