Computational Pipeline Yields SART3-Targeted Multi-Epitope Cancer Vaccine Candidate

Researchers have designed and computationally evaluated a multi-epitope vaccine aimed at a tumor-associated protein called SART3, reporting in Genomics & Informatics that the optimized 344-residue construct shows a suite of favorable predicted properties: non-allergenicity, high predicted stability, antigenicity, strong theoretical binding to the immune receptor TLR4, and structural robustness in molecular dynamics simulations. The work, published on 1 May 2026 by Zeynab Bayat, Faezeh Mahdian-Khoo, Lida Samie and Amir Taherkhani, is entirely computational. The authors themselves stress that every result is a prediction that requires laboratory and preclinical validation before the construct can be considered a genuine vaccine candidate.

SART3, short for squamous cell carcinoma antigen recognized by T-cells 3, is a nuclear RNA-binding protein with a half-tetracopeptide repeat region at its N-terminus and two RNA recognition motifs near the C-terminus. It functions as a U4/U6 recycling factor in pre-mRNA splicing and also influences gene regulation, stem cell pluripotency, embryonic development and hematopoiesis. In cancer biology, SART3 overexpression has been linked to poorer prognosis in hepatocellular carcinoma and non-small-cell lung cancer, and the protein has been demonstrated as a tumor-associated antigen in breast cancer, hepatocellular carcinoma, renal cell carcinoma, brain tumors, prostate cancer, musculoskeletal tumors and gastric cancers. Because its expression is minimal in most normal tissues yet consistently upregulated across a range of solid tumors, the authors note that this pattern has supported its development as a target for peptide-based immunotherapy.

The design pipeline began with the SART3 amino acid sequence of 963 residues retrieved from UniProtKB. Potential T-cell epitopes were identified with the Immune Epitope Database (IEDB) platform and screened against a percentile rank cutoff below 1.0 for human MHC alleles, a threshold chosen to enrich for the top 1% of predicted binders. Filtered epitopes were then assessed for immunogenic potential with the VaxiJen v2.0 server and for toxicity with ToxinPred. Promising candidates were docked to MHC alleles using HPEPDOCK 2.0, retaining epitopes with binding energies below −200 kcal/mol, a threshold the authors describe as pragmatic and based on their own comparative experience rather than a universally established cutoff. Experimentally validated SART3 epitopes from the IEDB repository were also incorporated into the selection process.

From this screening, six high-affinity MHC class I epitopes and six MHC class II epitopes were assembled into a single construct. A peptide adjuvant was placed at the N-terminus: the 50S ribosomal protein L7/L12 from Mycobacterium tuberculosis, chosen for its documented agonist activity at Toll-like receptor 4, a pattern recognition receptor that triggers innate immune responses. The adjuvant was connected to the first cytotoxic T lymphocyte epitope via the (EAAAK)₃ linker, which promotes an alpha-helical structure and maintains spatial separation between adjuvant and epitopes. AAY linkers joined the MHC-I epitopes, and GPGPG linkers connected the helper T lymphocyte epitopes and attached a C-terminal histidine tag for purification. In total, 110 distinct candidate constructs of 344 residues each were generated and screened.

The lead candidate emerged from quantitative scoring of physicochemical and biophysical properties. It was confirmed non-allergenic by the AllerTOP tool, showed an instability index of 17.16, an aliphatic index of 76.48, a theoretical isoelectric point of 6.62, a predicted half-life of 30 hours in mammalian reticulocyte cells, and a GRAVY hydropathicity score of −0.315, indicating a hydrophilic character. Its VaxiJen antigenicity score was 0.67, and the SOLpro server assigned a 0.96 probability of favorable aqueous solubility, a property the authors link to ease of overexpression in E. coli. Reverse translation and codon optimization for the E. coli K12 strain yielded a Codon Adaptation Index of 1.0 and a GC content of 59.32%.

Structural modeling was performed with I-TASSER, producing a model with a C-score of −3.34 on a reliability scale from −5 to 2, which was subsequently refined with GalaxyRefine. Quality checks were encouraging: ProSA-web returned a Z-score of −5.37, a value the authors describe as consistent with native proteins, and PROCHECK’s Ramachandran plot placed 83.9% of residues in the most favored regions, with 12.2% in additionally allowed, 2.1% in generously allowed and 1.7% in disallowed regions. Predicted secondary structure comprised approximately 49.13% alpha-helices, 11.05% beta-sheets and 39.83% coils.

Docking against TLR4, using the crystal structure with PDB ID 4G8A at 2.4 Å resolution, predicted a favorable binding mode with a Gibbs free energy of −265.61 kcal/mol and a confidence score of 91%. The authors caution that docking scores predict geometric fit rather than thermodynamic binding affinity or functional immune activation. Molecular dynamics simulations of the vaccine–TLR4 complex ran for 100 nanoseconds with a CHARMM force field. The backbone root-mean-square deviation rose quickly to about 1.4 nm by 10 ns before settling at 1.8 nm at 100 ns, and fluctuation analyses ranged from 0.2 to 1.9 nm for the vaccine and 0.2 to 1.2 nm for TLR4. Normal mode analysis with the iMOD server suggested minimal structural distortion, with an eigenvalue of 2.5 × 10⁻⁵ interpreted as indicating a substantial energetic barrier to deforming the bound complex.

Population coverage analysis using the IEDB tool estimated that the selected MHC class I epitopes would cover 84.72% of individuals across diverse human populations, with an average of 1.93 epitope–HLA hits and a PC90 of 0.65. The MHC class II epitopes showed lower coverage of 37.46%, with 0.49 average hits and a PC90 of 0.16. Cytokine induction predictions for the six MHC-II epitopes indicated that KPKGLAYVEYENES and ESVIQNYNKALQQL may induce IL-4, while FKVFRYSTSLEKHK and KVFRYSTSLEKHKL may induce IL-10; none was predicted to induce IFN-γ. These cytokine analyses were performed solely for descriptive reporting and did not influence epitope selection. ElliPro prediction identified 11 discontinuous B-cell epitopes, the largest comprising 32 residues with a score of 0.765.

An in silico immune simulation with the C-ImmSim server depicted a robust, time-orchestrated response over 35 days: antigen cleared within about five days, antibody titers of IgM and combined IgM plus IgG peaked between days 10 and 15, memory B cells peaked above 200 cells per cubic millimeter by day 6, active T helper cells rose to nearly 1900 cells per cubic millimeter by day 8, and active cytotoxic T cells climbed gradually to roughly 700 cells per cubic millimeter by day 30. The simulated cytokine profile showed IFN-γ peaking above 400,000 ng/mL around day 12 and an early IL-2 spike near 200,000 ng/mL by day 6. Disulfide engineering with Disulfide by Design 2.0 predicted 26 candidate residue pairs, with Val80–Ala123 showing the lowest bond energy at 1.16 kcal/mol, suggesting optimal geometry for a stabilizing cysteine bridge, and Glu68–Thr126 a viable alternative at 2.01 kcal/mol.

The authors place their construct alongside two earlier immunoinformatics designs from their group targeting PRAME and uPAR, noting that the SART3–TLR4 docking energy of −265.61 kcal/mol was slightly better than the PRAME construct’s −250.5 kcal/mol but less favorable than the uPAR complex’s −334.37 kcal/mol. They also flag safety considerations: because SART3 is a nuclear protein involved in essential splicing processes, on-target off-tumor effects and autoimmune phenomena cannot be excluded in principle, although early-phase clinical studies of SART3-derived peptides in solid tumor patients have reported mainly mild local injection-site reactions. TLR4 signaling, too, has been implicated in tumor-promoting inflammation in some cancer models, making its net impact dependent on empirical testing. The team concludes that the construct is a hypothesis-generating candidate whose folding, epitope accessibility, immunogenicity, protective efficacy and safety all remain to be demonstrated experimentally.

Subject of Research: In silico design and evaluation of a multi-epitope cancer vaccine targeting the SART3 tumor-associated antigen

Article Title: A bioinformatics pipeline for the design of a SART3-targeted cancer vaccine with enhanced immunogenicity

Article References: Bayat, Z., Mahdian-Khoo, F., Samie, L., & Taherkhani, A. (2026). A bioinformatics pipeline for the design of a SART3-targeted cancer vaccine with enhanced immunogenicity. Genomics & Informatics, 24(1), Article 11. https://doi.org/10.1186/s44342-026-00068-5

Image Credits: AI Generated

DOI: 10.1186/s44342-026-00068-5

Keywords: SART3, multi-epitope vaccine, cancer immunotherapy, immunoinformatics, TLR4 adjuvant, molecular docking, molecular dynamics, epitope prediction, disulfide engineering, in silico vaccine design, tumor-associated antigen, Genomics & Informatics