The proportion of mutations predicted to have a deleterious effect differs between gain and loss of function genes in neurodegenerative disease

Abstract

As more studies are turning to bioinformatic prediction programs to assess the potential impact of amino acid substitutions, it is relevant to evaluate the prediction results these programs give in genes that have been well-characterized for Mendelian diseases. Eight genes responsible for neurodegenerative disease with many identified mutations were subgrouped into those that either have a gain or loss of function disease mechanism. Three prediction programs, PolyPhen, Panther and SIFT, were queried for the reported missense mutations. The mean percent of benign mutations was significantly higher in gain of function genes using the PolyPhen program (38% versus 21%, p=0.007). The probability that a gain of function mutation was predicted to have a damaging role was also significantly less using the Panther program (p=4.86x10 -12 ). In contrast, there was no difference between gain and loss of function gene when the SIFT program was used. However, the most accurate distinction between gain and loss of function genes could be obtained when considering the mutations for which all three programs predicted the same result. Further, stratification of SOD1 mutations indicated that only the PolyPhen program could distinguish mutations that impaired enzymatic activity of SOD1 from those with near wildtype activity. The profile of benign and damaging changes from these genes will aid in the interpretation of bioinformatic prediction program results from missense mutations identified in novel genes.

Comparison of Bioinformatics Prediction, Molecular Modeling, and Functional Analyses ofFOXC1Mutations in Patients with Axenfeld-Rieger Syndrome

“…Nevertheless, the results of these predictions should still be interpreted with caution, as previous studies demonstrated the low accuracy of these tools for gain‐of‐function mutations [Valdmanis et al., ; Flanagan et al., ]. Therefore, as per clinical guidelines for the interpretation of missense variants, the predictions made using bioinformatics prediction software should be interpreted together with the results of functional studies, data on population frequency and segregation in affected families.…”

Single nucleotide variations: Biological impact and theoretical interpretation

“…Otherwise, machine learning methods that have used part of the validation data in their training may appear to be more accurate than they really are. When available, comparisons that are performed by independent researchers are preferable 53,85. In one such study, the performance of four commonly used methods (SIFT, Align-GVGD, PolyPhen-2, and Xvar which is now called MutationAssessor) was compared for 267 well-characterized human missense mutations in the BRCA1, MSH2, MLH1, and TP53 genes 85.…”

Assessment of in silico protein sequence analysis in the clinical classification of variants in cancer risk genes

“…With the increased use of clinical genetic testing and the challenges of classifying missense variants, in silico tools are now incorporated into professional society guidelines for variant classification (Easton et al 2007; Plon et al 2008; Richards et al 2015) and are used for clinical variant classification. However, several studies have demonstrated that the specificity (correct identification of benign variants) of these tools can be as low as 13% (Chan et al 2007; Doss and Sethumadhavan 2009; Flanagan et al 2010; Gnad et al 2013; Leong et al 2015; Miosge et al 2015; Ng and Henikoff 2002; Schiemann and Stowell 2016; Thusberg et al 2011; Valdmanis et al 2009) and with poor correlation between results from multiple programs (Thusberg et al 2011). While these limitations may be acceptable in selecting variants for research, false positives and variability between methods may compromise patient care in a clinical setting.…”