Performance evaluation of deleteriousness prediction methods for intronic SNVs in next generation sequences
![Thumbnail](/xmlui/themes/Mirage2//images/thumbnails/thumn_cientif.png)
Identificadores![](/xmlui/themes/Mirage2//images/despl_identif.png)
Identificadores
Visualización o descarga de ficheros![](/xmlui/themes/Mirage2//images/despl_contenidos.png)
Visualización o descarga de ficheros
Fecha de publicación
2020Título de revista
EUROPEAN JOURNAL OF HUMAN GENETICS
Tipo de contenido
Journal Article
Resumen
Introduction: Alterations in splicing sites (ss) are estimated to explain approximately 10% of human disease causal variants. Mutations outside the ss but affecting “regulatory elements” can be up to 25%. Accurate deleteriousness prediction for intronic variants is crucial for diagnostic purposes. Many deleteriousness prediction methods have been developed, but their relative values are still unclear in practical applications. We comprehensively evaluated the predictive performance of two complementary deleteriousness-scoring methods using information from real patients.
Material and Methods: We selected the dbscSNV (both ADA and RF scores) and SPIDEX algorithms, that study variants in splicing consensus regions or in regulatory regions respectively. The tools, either alone or in combination, were tested on 29294 gene intronic SNVs that have previously been characterised by ClinVar as either “pathogenic” (430) or “benign” (28864). The sensitivity, specificity and positive and negative predictive values were calculated. Moreover, we applied the algorithms to WES data from undiagnosed patients, and we analysed the mRNA sequence from genes that fitted the patient’s phenotype.
Results: The highest sensitivity corresponds to dbscSNV with 96.55% while the best specificity is for SPIDEX with 95.78%. When considering the 3 scores (SPIDEX, dbscSNV ADA and RF Score), the sensitivity and specificity values were 60.7% and 94.6%. The Positive and Negative Predictive Value were 14.45% and 99.39%. The results for 20 undiagnosed cases are presented.
Conclusions: Besides the low positive predictive value, the combination of both algorithms leads less than 1% of false negatives, so their routine use can be recommended for diagnostic purposes.
![](/xmlui/themes/Mirage2//images/final_lines.png)