The mutation significance cutoff: gene-level thresholds for variant predictions

Next-generation sequencing (NGS) has made it possible to identify about 20,000 variants in the protein-coding exome of each individual, of which only a few are likely to underlie a genetic disease. Variant-level methods such as PolyPhen-2, SIFT and CADD are useful for obtaining a prediction as to whether a given variant is benign/damaging1–3 or tolerant/intolerant1–3 (we hereafter use the terms benign/deleterious). These methods are commonly interpreted in a binary manner for filtering out benign variants from NGS data, with a single significance cutoff value across all protein-coding genes. PolyPhen-2 and SIFT integrate the fixed cutoff in the software. CADD proposed (but did not recommend for categorical usage) the fixed value of 15 (or another value between 10 and 20). Gene-level methods, such as RVIS, de novo excess and GDI are also useful4–6. Combining fixed gene-level and variant-level cutoffs is also applied in the RVIS hot zone approach4. However, owing to the diversity of medical and population genetic features between human genes and across populations, a uniform cutoff is unlikely to be accurate genome-wide.