Leveraging AI predictive models to develop thymic tumor–specific peptides mimicking COL17A1 alterations.

e20138 Background: We recently identified COL17A1 protein as a novel target in thymic epithelial tumors (TETs). Here, we designed tumor-specific peptides to mimic mutation regions in COL17A1. These peptides are intended as tools to generate antibodies targeting mutation-specific epitopes on COL17A1. Methods: In TCGA dataset, E391Kfs12, V809A, and P1409L alterations were frequently observed in the COL17A1 gene across various tumors, including TETs. The COL17A1 sequence was retrieved from UniProt, and mutations were introduced individually and in combination into the sequence. BLAST alignment analysis enabled direct comparison between the wild type and altered amino acid sequences. Next, we developed peptides of 8-10 Amino Acid (AA) around each mutation. We assessed the biophysical properties, immunogenicity (NetMHCpan), B-cell epitope prediction (Kolaskar), surface accessibility (Emini), solubility (NetSolP), thermal stability (TempStatPro) and enzyme function (TFPC). AlphaFold-3 analyzed predicted Template Modeling (pTM) and Local Distance Difference Test (pLDDT) scores for global folding and local stability, respectively. Results: Among the mutations, E391Kfs12 was located intracellularly, while V809A, and P1409L were extracellular. Using MutaGene, P1409L & E391Kfs12 were functionally deleterious. Next, a total of 144 peptides were designed to mimic mutations, including single and combined alteration. COL17A1 peptides had stronger binding affinity to MHC class I compared to class II. HLA-A24:02 & HLA-A26:01 (Score-BA 0.01) the strongest peptide-HLA-I affinity. The T-cell Immunogenicity predictions indicated the highest score for peptides near P1409L (Max 0.1). Emini surface accessibility showed the highest cell surface expression for E391Kfs12 (2.03) and P1409L (1.89). Epitope Conservancy Analysis evaluated matches with target and related sequences, with P1409L achieving the best score of 80% (range 55%–90%). Next, we divided peptides into 6 groups based on mutation sites. We assessed pLDDT (0–100), where high scores indicate strong local structural confidence. Single peptides had pLDDT >90, while most grouped peptides had scores <70. Peptides mimicking E391Kfs12 showed highest solubility (mean: 0.64, 0.60–0.78) and strong thermal stability at 45°C. P1409L peptides exhibited high enzymatic activity (mean: 0.48) but lower thermal stability (>70°C). V809A peptides demonstrated intermediate solubility (mean: 0.54) and stability, requiring optimization. Conclusions: AI predictive models enabled the design of thymic-tumor-specific peptides mimicking COL17A1 alterations. E391Kfs12 peptides with superior solubility and stability, and P1409L peptides, with strong immunogenicity, are promising candidates for antibody epitope generation. These findings provide a foundation for developing mutation-specific antibodies targeting thymic epithelial tumors.