The Promise of Personalized Cancer Vaccines
Every tumor is unique. The mutations that drive one patient's melanoma are different from the mutations driving another patient's melanoma, even though both cancers share the same diagnosis. This fundamental biology has frustrated oncology for decades: a drug that works for 30% of patients fails for the other 70%, and predicting who will respond remains imprecise. Personalized cancer vaccines flip the paradigm. Instead of designing one treatment for all patients, they design one treatment for one patient – custom-built from the unique mutational fingerprint of that individual's tumor.
The concept is elegant. Somatic mutations in tumor cells alter protein sequences, creating peptides that are absent from every normal cell in the body. These mutant peptides – called neoantigens – are displayed on the tumor cell surface by MHC (Major Histocompatibility Complex) molecules. If the immune system recognizes these neoantigens as foreign, it can mount a cytotoxic T cell response that destroys the tumor. A personalized cancer vaccine delivers the patient's specific neoantigens (as synthetic peptides or mRNA) to train the immune system to recognize and attack their unique tumor.
This is not speculation. As of 2026, multiple Phase 2 clinical trials have demonstrated that personalized neoantigen vaccines reduce recurrence risk in cancer patients. BioNTech's autogene cevumeran (an individualized mRNA vaccine encoding up to 34 neoantigens) showed a 44% reduction in recurrence risk in resected pancreatic cancer when combined with the checkpoint inhibitor atezolizumab (ASCO 2024). Moderna's mRNA-4157 (V940), encoding up to 34 patient-specific neoantigens, combined with pembrolizumab reduced recurrence risk by 49% versus pembrolizumab alone in resected melanoma (KEYNOTE-942, Phase 2b). Gritstone Bio's GRANITE program demonstrated neoantigen-specific T cell responses in 100% of evaluable patients with microsatellite-stable colorectal cancer.
The clinical signals are strong enough that the FDA granted Breakthrough Therapy Designation to mRNA-4157 for adjuvant melanoma in 2023, and Phase 3 trials (V940-001, INTerpath-001) are enrolling. The field is transitioning from proof-of-concept to registration-grade evidence, and the computational tools that underpin vaccine design are becoming critical infrastructure.
The Clinical Landscape: Who Is Building What
The personalized cancer vaccine field is dominated by three technology platforms, each with distinct advantages.
BioNTech: Individualized Neoantigen-Specific Therapy (iNeST)
BioNTech's platform encodes up to 34 neoantigens in a single mRNA construct delivered as a lipid nanoparticle. Their manufacturing turnaround is approximately 6 weeks from biopsy to injection. The autogene cevumeran program (in collaboration with Genentech/Roche) is the most advanced, with positive Phase 2 data in pancreatic cancer and melanoma. BioNTech's computational pipeline uses proprietary neoantigen prediction algorithms validated against mass spectrometry data from patient tumors.
Moderna: mRNA-4157 (V940)
Moderna's approach also uses lipid nanoparticle-delivered mRNA encoding up to 34 neoantigens. The KEYNOTE-942 trial in collaboration with Merck (combining V940 with pembrolizumab) produced the strongest clinical signal to date: 49% recurrence-free survival improvement in Stage III/IV melanoma. The Phase 3 V940-001 trial (approximately 1,000 patients) is the most important ongoing study in the field. Moderna's platform leverages their COVID-19 mRNA manufacturing expertise for rapid production.
Gritstone Bio: EDGE and SLATE Platforms
Gritstone uses a heterologous prime-boost approach: a self-amplifying RNA (saRNA) prime followed by an adenoviral vector boost. Their EDGE platform predicts neoantigens using a proprietary neural network trained on mass spectrometry identification of naturally presented MHC ligands. The GRANITE program targets microsatellite-stable colorectal cancer, a traditionally immunotherapy-resistant indication. Their SLATE platform offers "off-the-shelf" shared neoantigen vaccines targeting common driver mutations (KRAS G12C, G12D, G12V, and others).
The Computational Pipeline: From Sequencing to Vaccine Design
Designing a personalized cancer vaccine is fundamentally a computational problem. The wet lab generates the sequence data; the computation identifies the vaccine targets. Here is the end-to-end pipeline.
Step 1: Identify Somatic Mutations
Whole exome sequencing (WES) or whole genome sequencing (WGS) is performed on both the tumor biopsy and a matched normal tissue sample (typically blood). Variant calling algorithms (Mutect2, Strelka2, VarScan2) identify somatic mutations by comparing the tumor to normal. Consensus calling across two or more algorithms reduces false positives. A typical melanoma yields 200 to 2,000 nonsynonymous somatic mutations; a pancreatic cancer yields 20 to 100.
Step 2: Generate Candidate Peptides
Each nonsynonymous mutation alters the amino acid sequence of a protein. For every mutation, generate all possible peptide windows of 8 to 11 amino acids (for MHC class I) and 13 to 25 amino acids (for MHC class II) that contain the mutated residue. A single point mutation in a protein generates approximately 20 candidate MHC-I peptides (sliding window across the mutation site). For a tumor with 500 mutations, this produces roughly 10,000 candidate peptides.
Step 3: HLA Typing
Every person expresses 6 MHC class I alleles (2 HLA-A, 2 HLA-B, 2 HLA-C) and up to 12 MHC class II alleles. The specific alleles determine which peptides can be presented to T cells. HLA typing from the patient's normal tissue sequencing data identifies their allele set. Computational HLA typing (OptiType for class I, HLA-HD for class II) achieves greater than 99% accuracy from standard exome sequencing data.
Step 4: MHC Binding Prediction
For each candidate peptide, predict binding affinity to each of the patient's HLA alleles. Peptides that bind with IC50 below 500 nM are considered binders; those below 50 nM are strong binders. NetMHCpan 4.1 and MHCflurry 2.0 are the standard tools, with AUC values of 0.85 to 0.95 for binding prediction. A typical pipeline filters 10,000 candidate peptides down to 200 to 500 predicted binders.
Step 5: Immunogenicity Scoring
MHC binding is necessary but not sufficient for immunogenicity. Additional factors include proteasomal processing (will the peptide be generated by the proteasome), TAP transport (will it reach the ER for MHC loading), T cell receptor recognition (is the peptide-MHC complex immunogenic), and expression level (is the mutant gene expressed in the tumor). Scoring algorithms combine these factors to rank predicted binders by likelihood of eliciting a T cell response. This typically narrows the list to 20 to 50 top neoantigen candidates.
Step 6: Vaccine Construct Design
For mRNA vaccines, the selected neoantigens (typically 10 to 34) are concatenated into a single polyepitope mRNA construct. The design includes a 5' cap, optimized 5' and 3' UTRs for translational efficiency, signal peptides for MHC processing, linker sequences between epitopes, and a poly-A tail. Codon optimization maximizes expression while maintaining appropriate GC content (40 to 60%) for mRNA stability. Secondary structure prediction ensures the mRNA folds correctly for efficient translation.
SciRouter End-to-End Workflow
SciRouter's Vaccine Design tools provide the complete computational pipeline from mutation list to vaccine construct. Here is the step-by-step workflow using the Python SDK.
Neoantigen Prediction
Start with the list of somatic mutations from variant calling and the patient's HLA alleles from HLA typing. The neoantigen pipeline endpoint processes all mutations, generates candidate peptides, predicts MHC binding, and scores immunogenicity in a single API call.
import os, requests
API_KEY = os.environ["SCIROUTER_API_KEY"]
BASE = "https://api.scirouter.ai/v1"
HEADERS = {"Authorization": f"Bearer {API_KEY}"}
# Patient HLA alleles (from OptiType or clinical HLA typing)
hla_alleles = [
"HLA-A*02:01", "HLA-A*03:01",
"HLA-B*07:02", "HLA-B*44:02",
"HLA-C*05:01", "HLA-C*07:02",
]
# Somatic mutations from variant calling (subset for demo)
mutations = [
{"gene": "BRAF", "mutation": "V600E",
"protein_context": "LATEKSRWSGSHQFEQLS"},
{"gene": "CDKN2A", "mutation": "P114L",
"protein_context": "DPHSGHVARTLDVRDCGPH"},
{"gene": "PTEN", "mutation": "R130Q",
"protein_context": "VNKFHYGGDPHGENILSQFNV"},
{"gene": "TP53", "mutation": "R248W",
"protein_context": "IRPVHHHSVRSPGGRAHSS"},
{"gene": "NRAS", "mutation": "Q61R",
"protein_context": "LLDILDTAGREEYSAMRDQ"},
{"gene": "MAP2K1", "mutation": "P124S",
"protein_context": "GQLPQSAHELLEKGLNKD"},
{"gene": "IDH1", "mutation": "R132H",
"protein_context": "PIIGRHAYGDQYRAT"},
{"gene": "CTNNB1", "mutation": "S45F",
"protein_context": "SYLDSGIHAGATFTPGED"},
]
# Run neoantigen prediction
result = requests.post(f"{BASE}/immunology/neoantigen-pipeline",
headers=HEADERS, json={
"mutations": mutations,
"hla_alleles": hla_alleles,
"peptide_lengths": [8, 9, 10, 11],
"binding_threshold_nm": 500,
"include_class_ii": True,
}).json()
print(f"Mutations analyzed: {len(mutations)}")
print(f"Candidate peptides generated: {result['total_peptides']}")
print(f"Predicted binders (IC50 < 500 nM): {result['num_binders']}")
print(f"Top neoantigens (immunogenicity scored): {len(result['top_neoantigens'])}")
# Display top 10 neoantigen candidates
print("\n=== Top Neoantigen Candidates ===")
for i, neo in enumerate(result["top_neoantigens"][:10]):
print(f"{i+1:>2}. {neo['peptide']:<14} | Gene: {neo['gene']:<8} "
f"| HLA: {neo['hla_allele']:<14} "
f"| IC50: {neo['binding_affinity_nm']:>6.1f} nM "
f"| Score: {neo['immunogenicity_score']:.3f}")MHC Binding Validation
For the top neoantigen candidates, run detailed MHC binding prediction to validate the pipeline results and obtain binding affinity estimates for all patient HLA alleles. This step identifies neoantigens that bind multiple HLA alleles (promiscuous binders), which are preferred because they are more likely to trigger a robust immune response.
# Validate top candidates against all patient HLA alleles
top_peptides = [neo["peptide"] for neo in result["top_neoantigens"][:15]]
binding_matrix = requests.post(f"{BASE}/immunology/mhc-binding",
headers=HEADERS, json={
"peptides": top_peptides,
"hla_alleles": hla_alleles,
"method": "netmhcpan",
}).json()
print("=== MHC Binding Matrix (IC50 nM) ===")
header = f"{'Peptide':<14} " + " ".join(f"{a[-5:]:>8}" for a in hla_alleles)
print(header)
print("-" * len(header))
for peptide_result in binding_matrix["results"]:
row = f"{peptide_result['peptide']:<14} "
allele_count = 0
for allele in hla_alleles:
ic50 = peptide_result["affinities"].get(allele, {}).get("ic50_nm", 99999)
marker = "*" if ic50 < 500 else " "
row += f"{ic50:>7.0f}{marker} "
if ic50 < 500:
allele_count += 1
row += f" ({allele_count} alleles)"
print(row)
# Identify promiscuous binders (bind 3+ alleles)
promiscuous = [
p for p in binding_matrix["results"]
if sum(1 for a in hla_alleles
if p["affinities"].get(a, {}).get("ic50_nm", 99999) < 500) >= 3
]
print(f"\nPromiscuous binders (3+ alleles): {len(promiscuous)}")Vaccine Construct Design
With the final neoantigen selection, design the mRNA vaccine construct. The vaccine design endpoint handles codon optimization, UTR selection, linker insertion, and secondary structure prediction.
# Select top neoantigens for vaccine (prefer promiscuous binders)
selected_neoantigens = [neo["peptide"] for neo in result["top_neoantigens"][:20]]
# Design mRNA vaccine construct
vaccine = requests.post(f"{BASE}/immunology/vaccine-design",
headers=HEADERS, json={
"neoantigens": selected_neoantigens,
"vaccine_type": "mrna",
"optimization": {
"codon_optimization": True,
"target_gc_content": [0.40, 0.60],
"add_signal_peptides": True,
"linker_type": "GPGPG", # Flexible linker between epitopes
"utr_optimization": True,
},
"include_structure_prediction": True,
}).json()
print("=== Personalized mRNA Vaccine Construct ===")
print(f"Neoantigens included: {vaccine['num_epitopes']}")
print(f"mRNA length: {vaccine['mrna_length']} nucleotides")
print(f"GC content: {vaccine['gc_content']:.1%}")
print(f"Codon adaptation index: {vaccine['cai_score']:.3f}")
print(f"Predicted MFE: {vaccine['mfe_kcal_mol']:.1f} kcal/mol")
print(f"\nConstruct elements:")
print(f" 5' cap: {vaccine['five_prime_cap']}")
print(f" 5' UTR: {vaccine['five_prime_utr'][:40]}...")
print(f" Signal peptide: {vaccine['signal_peptide']}")
print(f" Epitope cassette: {len(selected_neoantigens)} epitopes "
f"with {vaccine['linker_type']} linkers")
print(f" 3' UTR: {vaccine['three_prime_utr'][:40]}...")
print(f" Poly-A tail: {vaccine['polya_length']} adenosines")
# Export full sequence for manufacturing
with open("patient_vaccine_construct.fasta", "w") as f:
f.write(f">personalized_neoantigen_vaccine_{len(selected_neoantigens)}ep\n")
seq = vaccine["full_mrna_sequence"]
for i in range(0, len(seq), 80):
f.write(seq[i:i+80] + "\n")
print(f"\nFull sequence exported to patient_vaccine_construct.fasta")Case Study: Demo Melanoma Patient
To illustrate the complete workflow, consider a hypothetical melanoma patient with the following clinical profile: Stage IIIC melanoma, BRAF V600E positive, resected with clear margins, high risk of recurrence. Whole exome sequencing of the tumor identifies 847 somatic mutations, of which 312 are nonsynonymous (changing the protein sequence). The patient's HLA type is HLA-A*02:01, HLA-A*03:01, HLA-B*07:02, HLA-B*44:02, HLA-C*05:01, HLA-C*07:02.
Running the 312 nonsynonymous mutations through the SciRouter neoantigen pipeline generates 6,240 candidate MHC-I peptides (8 to 11-mers containing each mutated residue). Of these, 487 are predicted to bind at least one of the patient's HLA alleles with IC50 below 500 nM. Immunogenicity scoring ranks these 487 binders, and the top 34 are selected for the vaccine construct.
The selected neoantigens include peptides from known melanoma drivers (BRAF V600E, NRAS Q61R, CDKN2A deletions) as well as passenger mutations that happen to produce strong MHC binders. Notably, 8 of the 34 selected peptides are promiscuous binders (binding 3 or more of the patient's HLA alleles), increasing the probability of a robust poly-clonal T cell response.
The mRNA vaccine construct encodes all 34 neoantigens as a polyepitope cassette with GPGPG linkers between epitopes, flanked by optimized 5' and 3' UTRs and a 120-nucleotide poly-A tail. The total mRNA length is approximately 3,800 nucleotides. Codon optimization targets human codon usage with 48% GC content. The construct is ready for lipid nanoparticle formulation and GMP manufacturing.
The Regulatory Pathway
Personalized cancer vaccines present a unique regulatory challenge: every product is different because every patient's tumor is different. The FDA has adapted its framework to accommodate this through the Individualized Neoantigen Therapy (INT) guidance.
The key regulatory principle is that the manufacturing process is the product, not the specific molecular sequence. The sponsor validates the process (tumor sequencing, neoantigen prediction, mRNA design, LNP formulation) through the IND application, and each individual patient's vaccine is manufactured using that validated process without requiring separate approval. This is analogous to autologous cell therapies like CAR-T, where each product is patient-specific but manufactured under a single process validation.
The computational pipeline is part of the validated manufacturing process. The neoantigen prediction algorithm, the MHC binding prediction method, the immunogenicity scoring criteria, and the mRNA design parameters must all be documented, version-controlled, and validated as part of the regulatory submission. Changes to the computational pipeline require process change notifications and may trigger additional validation studies.
Phase 3 registration trials currently underway (V940-001 for melanoma, INTerpath-001 for multiple solid tumors) will generate the evidence needed for regulatory approval. If positive, the first personalized cancer vaccine approvals are expected by 2027 to 2028.
Cost Trajectory and Accessibility
The current cost of a personalized cancer vaccine is estimated at $100,000 to $200,000 per patient, dominated by three components: tumor sequencing ($2,000 to $5,000), computational neoantigen prediction ($500 to $2,000), and mRNA manufacturing ($80,000 to $150,000). As the field scales, each component is expected to decrease significantly.
- Sequencing costs continue to fall. Whole exome sequencing has dropped from $5,000 in 2020 to under $800 in 2026. Whole genome sequencing is approaching $200 per sample with the latest Illumina and Element platforms.
- Computational costs are already low and decreasing. The SciRouter neoantigen pipeline processes a full patient analysis for a few dollars in API credits. As prediction algorithms improve, the computational step adds negligible cost to the total.
- Manufacturing costs are the major bottleneck but are declining with automation. BioNTech's dedicated manufacturing facilities and Moderna's experience from COVID-19 vaccine production are driving costs down. Industry estimates project $30,000 to $50,000 per patient by 2028 and under $20,000 by 2030 as production scales.
For context, CAR-T cell therapies (Kymriah, Yescarta) cost $373,000 to $475,000 per patient at list price. If personalized cancer vaccines demonstrate comparable efficacy in their indications at one-third the cost, they will be among the most cost-effective immunotherapies available.
The Future: Beyond Cancer
The computational pipeline developed for personalized cancer vaccines has applications beyond oncology. The same neoantigen prediction framework can be applied to autoimmune diseases (identifying self-antigens driving autoimmune responses), infectious diseases (predicting T cell epitopes for emerging pathogens), and transplant medicine (predicting alloreactive T cell responses against donor HLA mismatches).
The convergence of affordable sequencing, accurate computational prediction, and rapid mRNA manufacturing is creating a new category of medicine: vaccines designed for a single person based on their unique biology. The tools described in this article – SciRouter's vaccine design, neoantigen pipeline, and MHC binding prediction endpoints – make it possible for any researcher to run the computational portion of this pipeline today, accelerating the translation from tumor biology to therapeutic intervention.
Personalized cancer vaccines represent one of the most promising frontiers in oncology. The clinical evidence is building, the manufacturing challenges are being solved, and the computational tools are already available. The question is no longer whether personalized vaccines work, but how quickly they can be made available to every patient who could benefit.