mHA prediction algorithm schema. (A) The alleles at cSNPs for all 101 DRPs in the patient cohort were determined using the Illumina NS-12 array platform (n = 13 917 cSNPs). (B) Each allele for each cSNP was mapped to the reference coding transcriptome. (C) The peptide sequences for all 8-, 9-, 10-, 11-, 16-, 20-, and 24-mer amino acid sequences that contained the allele of interest were computationally generated (n = 2 189 712 amino acid sequences covering both alleles for all 11 172 cSNPs). (D) All peptides that could be mHAs for a given DRP (ie, one allele was present in the recipient, but not the donor) were tested for predicted binding affinity to the DRP’s HLA types. (E) All predicted mHAs were classified as GVL or GVHD associated based upon the tissue expression of the source protein for the mHA.