Fig. 1From: MetaRNN: differentiating rare pathogenic and rare benign missense SNVs and InDels using deep learningData preparation and model development steps for MetaRNN. First, the target variant and affect codon were identified, e.g., g.1022225G>T. Second, the flanking sequences were retrieved as well as all possible alternative alleles, as illustrated on the nucleotides to the right of the up-pointing arrow. Third, only alleles that result in a missense variant were kept, and annotation scores were averaged across these alleles within the same locus. For example, the annotation scores for variants g.1022230A>T and g.1022230A>C will be averaged, to get the locus-specific annotations. Lastly, the model will be trained using the annotations for context variants and annotations for the target variantBack to article page