Table 1 X-CAP features. violetItalicized features are novel and have not been used in previous stopgain pathogenicity predictors. Specifically, no features related to zygosity, stop codon read-through, or alternative translation reinitiation are present in earlier classifiers

Binary variable distinguishing homozygous (and hemizygous) variants from heterozygous variants, inputed when known or predicted as a function of benign stopgain alleles at the same position in training set when unknown

Number of benign stopgains in training set along gene divided by gnomAD’s expected number of loss-of-function variants

Measure of gene intolerance to functional variation

Two non-exclusive, binary features indicating whether a recessive or dominant disease listed in the OMIM Gene Map is caused by mutations in this gene

Transcript or exon contains no benign variants and at least one pathogenic variant within training set

Variant is skipped in at least one isoform of the gene

Number of coding nucleotides from CDS start and end

Distance from CDS start divided by CDS length

Number of coding nucleotides from exon start and end

Distance from exon start divided by exon length

Number of nucleotides in overlapped exon

Index of the exon that the variant overlaps

Number of exons in overlapped transcript

Ternary variable indicating if the variable is located on an autosomal, X, or Y chromosome

Number of coding nucleotides upstream from last exon-exon junction (negative if downstream of junction)

Percentage of overlapped transcripts in which the variant is >50 bp upstream of the last exon-exon junction

One-hot encoding of the new stop codon introduced by the stopgain

Number of base pairs between the variant and the next potential downstream start codon within the mRNA

Base-pair conservation across vertebrates of upstream, downstream, and overlapped exon regions

Regional conservation across vertebrates of upstream, downstream, and overlapped exon regions