Background: Prediction of pathogenic variants is one of the biggest challenges for researchers and clinicians in the time of next-generation sequencing technologies. Stratification of individuals based on truly pathogenic variants might lead to improved, personalized treatments. Results: We present Frequency Conservation Score (FCS) and Frequency Conservation Score for Mitochondrial DNA (FCSMt) two methods for the detection of pathogenic single nucleotide variants in nuclear and mitochondrial DNA, respectively. These scores are based in a random forest model trained over a set of potentially relevant predictors: (i) conservation scores (PhastCons and phyloP); (ii) locus variability at each genomic position built from gnomAD database and (iii) physicochemical distance for amino acids substitutions and the impact/consequence over the canonical transcript. FCS showed an AUC of 98% for deleteriousness in an independent validation dataset, outperforming other scores such as metaLR, metaSVM, REVEL, DANN, CADD, SIFT, PROVEAN or FATHMM-MKL. Moreover, FCSMt presented an AUC=0.92 for pathogenic mitochondrial SNVs detection. The tool is available at http://bioinfo.cnic.es/FCS . Conclusions: FCS and FCS-Mt improve pathogenic mutation detection, allowing the prioritization of relevant variants in Whole Exome and Whole Genome Sequencing Analysis.

Frequency Conservation Score (FCS): the power of conservation and allele frequency for variant pathogenic prediction