Ready to bring this to a patient?
We're in closed beta, accepting a small number of pilot cases per quarter through veterinary oncologists. Join the waitlist and we'll be in touch.
From a single DNA sequencing file to a synthesis-ready personalized mRNA vaccine blueprint. The complete bioinformatics pipeline, explained.
When a tumor grows, its DNA accumulates unique mutations that healthy cells don't have. These mutations create abnormal proteins called neoantigens, which are invisible to a dog's immune system. RosieVaccine teaches the immune system to recognize and attack them.
A VCF file from your vet's DNA sequencing lab captures every place the tumor's genome differs from the dog's healthy cells.
The pipeline identifies which mutant peptides will bind strongly to the dog's immune display molecules (DLA), making them visible for attack.
A deterministic builder assembles the top binders into a codon-optimized mRNA sequence ready to hand off to a synthesis lab for manufacturing.
Each stage is automated. The whole process runs in the background while you wait, and no bioinformatics experience is required.
The Variant Call Format (VCF) file is the output from the sequencing lab: a structured text file listing every position in the genome where the tumor differs from normal tissue. Each line encodes a chromosome, position, reference base, and alternate (mutant) base.
Only missense mutations proceed, meaning changes that swap one amino acid for another. Silent mutations (same protein) are filtered out.
Each mutation's genomic coordinate is mapped to its protein context. Using the dog's reference genome, we look up the codon affected, determine which amino acid is present in normal tissue (wild-type) and which one the mutation creates (mutant).
Val → Glu substitution at position 6. The mutant version is what the tumor expresses.
MHC Class I molecules (DLA in dogs) display 9-amino-acid fragments on cell surfaces. For each mutation, a sliding window extracts all 9-mer peptides that contain the mutation site, creating both a wild-type (WT) version and a mutant version for comparison.
DLA (Dog Leukocyte Antigen) is the canine equivalent of human HLA, the immune system's display molecule. Each dog has specific DLA alleles that determine which peptide shapes fit into the MHC binding groove and get presented to T-cells.
We submit every 9-mer peptide pair to NetMHCpan-4.2, the gold-standard binding affinity predictor from DTU Health Tech. It returns a %Rank score for each peptide: lower is better.
After prediction, we apply two filters to select the most promising vaccine targets:
Absolute %Rank threshold
Mutant peptide must have mut_rank < 0.5%. It must be a Strong Binder to the dog's DLA allele.
Fold improvement score
We compute wt_rank / mut_rank. A high fold score means the mutant peptide is dramatically better at binding than the normal version, the ideal vaccine target: specific to the tumor, ignored in healthy tissue.
The top Strong Binders are passed to a deterministic assembly step (codon optimization → procedural builder) that designs a complete, synthesis-ready mRNA molecule. No generative model touches the therapeutic sequence, and the same inputs always produce the same blueprint. This is the actual deliverable your vet hands to the manufacturing lab.
The mRNA is assembled with all the structural elements required for stability and expression inside the dog's muscle cells after injection:
Codon optimization
DNA codons are selected to maximize translation efficiency in mammalian (canine) cells, yielding more protein from the same sequence.
Neoantigen chaining
Top binders are concatenated with optimized linkers so one mRNA molecule trains the immune system against multiple targets simultaneously.
The pipeline compiles everything into a structured PDF report containing:
Everything above is the method. Here is the output. This is an actual pipeline run on representative canine tumor sequencing data. Every number, peptide, and binding score below is exactly what a veterinary oncologist would receive.
Sample Patient
Worked example · Mutation discovery
The pipeline compared Rosie's healthy DNA to tumor DNA and found three somatic missense mutations, spots where the tumor protein differs from normal tissue. These are the targets.
R175H
Arginine → Histidine at position 175
TP53 is the most commonly mutated tumor suppressor gene in canine cancers. The R175H hotspot disrupts DNA damage response, allowing cancer cells to proliferate unchecked.
Best peptide binding rank
✓ Excellent Strong Binder
G12D
Glycine → Aspartate at position 12
KRAS G12D is a driver mutation found in many aggressive canine cancers. It locks the KRAS protein in an "always on" state, sending constant growth signals to the cancer cell.
Best peptide binding rank
✓ Excellent Strong Binder
R12C
Arginine → Cysteine at position 12
TRPM7 mutations regulate ion channel activity linked to tumor cell migration and invasion. R12C affects the kinase domain, altering cellular signaling pathways that drive metastatic spread.
Best peptide binding rank
✓ Excellent Strong Binder
Worked example · Binding prediction
For each mutation, the pipeline generated multiple 9-amino-acid peptide windows and scored each one's binding to Rosie's immune receptor. Binding is predicted by NetMHCpan-4.2 against the dog's specific DLA allele. %Rank below 0.5 = Strong Binder. Lower is better. Every candidate here is a confirmed Strong Binder (% Rank < 0.5).
| Mutation | Mutant Peptide 9-amino-acid sequence containing the mutated residue. This is what the vaccine teaches the immune system to recognize. | Binding Rank %Rank from NetMHCpan; lower is better. <0.5% = Strong Binder. <2% = Weak Binder. | Normal Tissue How well the normal (healthy) version of the peptide binds. High normal rank = the immune system ignores it naturally. Good. | Fold Improvement How much better the mutant peptide binds vs. normal. Higher = the immune system will target the tumor specifically, not healthy cells. | Verdict |
|---|
DLA allele: DLA-88*508:01 · Prediction method: NetMHCpan-4.2 · Peptide length: 9-mer
Worked example · Vaccine design
From all candidates, the pipeline selects the strongest binders, one unique peptide per mutation, to maximize immune coverage while minimizing off-target risk.
AAAAHAAAA
GPGTSCERS
VVVGADVGK
A binding rank of 0.066% means Rosie's immune receptor binds to this peptide better than 99.93% of all random peptides, an exceptionally strong signal. The 77× fold improvement on TRPM7 means the mutant peptide binds 77 times better than the normal version of the same sequence, so the immune system will strongly prefer attacking tumor cells over healthy tissue.
All 7 candidates cleared the 0.5% Strong Binder threshold, the same cutoff used in human neoantigen vaccine research. This is an unusually clean result and suggests these mutations are highly immunogenic targets.
Worked example · The deliverable
After selecting the top targets, a deterministic codon-optimization step generates a complete, synthesis-ready mRNA sequence. Here's the structure of Rosie's personalized vaccine blueprint.
Codon Optimized
Each neoantigen sequence is rewritten with canine-preferred codons for maximum translation efficiency in dog cells.
Pseudouridine Modified
Uridine residues are replaced with N1-methylpseudouridine (m1Ψ) to evade innate immune detection, the same modification used in Pfizer's COVID-19 vaccine.
Lab-Ready Format
The final report includes the full sequence in FASTA format, a synthesis spec sheet, and a cover letter for your vet and synthesis facility.
Not all dogs display the same peptides. The DLA allele a dog inherits determines the shape of the binding groove, and therefore which 9-mers fit. Getting the allele right is critical.
DLA-88*508:01
Most common; used as default in beta
DLA-88*501:01
High prevalence in retriever breeds
DLA-88*034:01
Common in herding and working breeds
A future enhancement will auto-select alleles by breed and run the full pipeline across all probable alleles simultaneously.
Every nucleated cell in a dog's body continuously samples its own proteins, chops them into 9-mer fragments, and loads them onto MHC Class I molecules that sit on the cell surface. Cytotoxic T-cells patrol these displays. If they see a foreign or mutant peptide they recognize, they destroy the cell.
The MHC Class I binding groove has a fixed length and accommodates exactly 9 amino acids (occasionally 8 or 10). NetMHCpan-4.2 was trained on thousands of experimentally confirmed binding events at this length, making 9-mer prediction the most accurate in the field.
This approach is not theoretical. Personalized mRNA cancer vaccines using neoantigen pipelines are in active Phase II/III clinical trials in humans. RosieVaccine applies the same methodology to canine oncology.
Sahin et al. (Nature, 2017) demonstrated individualized neoantigen vaccines in melanoma patients using the same VCF → peptide → binding prediction pipeline. T-cell responses were observed against 60% of targeted neoantigens.
Sahin U et al., Nature 547, 222–226 (2017)
Dogs share significant MHC biology with humans, making them a validated translational model. DLA allele frequencies and binding groove structures have been characterized enabling direct use of NetMHCpan architecture.
Kennedy LJ et al., Tissue Antigens (2002); Wagner JL, J Hered (2003)
The state-of-the-art pan-specific MHC-peptide binding predictor from DTU Health Tech. Trained on hundreds of thousands of experimentally measured binding events using machine learning. The %Rank threshold of 0.5 is the field standard for Strong Binder classification.
Reynisson B et al., Nucleic Acids Research 48, W449–W454 (2020)
The mRNA vaccine format, validated by the COVID-19 pandemic at scale, is ideal for personalized cancer vaccines: fast to manufacture, no live virus required, strong innate immune stimulation, and straightforward codon optimization for target species.
Pardi N et al., Nature Reviews Drug Discovery 17, 261–279 (2018)
Processing time: Beta mode
Currently the pipeline submits to the DTU Health Tech public NetMHCpan web server, which takes 5–20 minutes per job. A local binary installation (pending commercial license) will reduce this to under 2 minutes.
DLA allele input required
The pipeline currently requires manual DLA allele selection. We default to DLA-88*508:01 if unknown. Breed-based automatic allele selection is on the roadmap and will improve accuracy for known breeds.
Investigational use only
RosieVaccine is an investigational bioinformatics tool. It generates a blueprint for a licensed synthesis lab and veterinary oncologist to act on. It is not an approved veterinary medical device or treatment. Results are not guaranteed.
VCF quality matters
The pipeline output is only as good as the sequencing input. We require a tumor–normal paired VCF (somatic mutations only). Germline-only VCFs or low-coverage sequencing will produce poor neoantigen candidates.
We're in closed beta, accepting a small number of pilot cases per quarter through veterinary oncologists. Join the waitlist and we'll be in touch.