The human leukocyte antigen B27 (HLA-B27) gene has been observed to significantly increase the risk of developing ankylosing spondylitis (AS). The HLA-B27 allele B*27:05, a genetic ancestor of all other HLA-B27 alleles, is strongly associated with AS, whereas its close relative B*27:09 does not predispose for AS, although they differ by only a single residue (D116H substitution). In the present study, an integrative biology strategy that combines in silico calculations with in vitro assays is described to perform quantitative sequence-activity model (QSAM)-based high-throughput virtual screening, molecular dynamics/energetics-based large-scale mutagenesis analysis and FACS-based HLA stabilization assay against 28,244 9-mer self-peptides derived from 17 human cartilage proteins as the antigen candidates of HLA-B27. It is revealed that the D116H mutation, which brings B*27:05 to B*27:09, does not substantially shift the peptide binding profile of HLA-B27, suggesting that the B*27:05 and B*27:09 share a similar peptide repertoire. However, few self-peptides that bind differently to B*27:05 and B*27:09 are identified; their sequences are primarily differed by C-terminus (but basically consistent in N-terminus and middle region). HLA stabilization assays substantiate that three identified self-peptides, 2515KRSSRHPRR2523 (Aggrecan), 200TRSVSHLRK208 (Annexin) and 131EKAFSPLRK139 (Biglycan), exhibit high affinity to B*27:05 (BL50 = 76.3, 28.5 and 8.6 nM, respectively) and strong selectivity for B*27:05 over B*27:09 (8.9-fold, 4.3-fold and 12.7-fold, respectively), which are highly promising as the potential candidates involved in AS elicitation.