中文
中文版

Using AI to Design a Custom mRNA Cancer Vaccine for a Dog

Paul Conyngham's Complete Process Breakdown | Can an Ordinary Person Replicate This? | 2026-04-05
Video source: Today Show Australia | Cross-referenced with multiple reports

An engineer with zero biomedical background spent $3,000 on DNA sequencing,
used ChatGPT + AlphaFold to design the world's first custom dog cancer mRNA vaccine --
the tumor shrank 75% within two months.
Table of Contents
  1. Background Story
  2. Complete 7-Step Process (Replicable SOP)
  3. Treatment Timeline
  4. Tools & Technologies Used
  5. Cost Breakdown
  6. Can an Ordinary Person Do This? Step-by-Step Feasibility
  7. How to Do This in Taiwan? Local Resource Inventory
  8. The Science: Why This Approach Works
  9. Limitations & Risks
  10. Historical Parallels
  11. Business Insights

1. Background Story

Protagonist: Paul Conyngham, Sydney, Australia. Data scientist / machine learning engineer (17 years of tech experience, zero biomedical background)
Patient: Rosie, an 8-year-old Staffordshire Bull Terrier, rescue dog
Diagnosis: Mast Cell Cancer, multiple tumors. Vet gave her only months to live
Prior treatment: Chemotherapy + surgery both failed; tumors continued spreading
Turning point: Paul decided to use the only thing he was good at -- AI and data -- to try to save Rosie

Paul said in an interview: "Rosie is my best friend. She was there for me through breakups, through tough business negotiations, through walks in the forest. When she was given a death sentence, I felt I had to do something for her."

2. Complete 7-Step Process

Below is the complete process reconstructed from cross-referencing multiple reports. This is the most detailed step-by-step breakdown available in public information.

1
Use ChatGPT to Develop a Research Plan

Paul described Rosie's condition to ChatGPT (mast cell cancer, failed chemo and surgery, multiple tumors) and asked what treatment options remained.

ChatGPT suggested the immunotherapy route, specifically a personalized cancer vaccine, and directed him to get genomic sequencing done.

Key output: A complete research roadmap -- from DNA sequencing to vaccine design, step by step.

Anyone can do this Anyone can use ChatGPT for this step. The key is asking the right questions: don't just ask "my dog has cancer, what should I do?" Instead ask "if I want to use immunotherapy to treat canine mast cell cancer, what steps are needed? What data do I need from a genomics perspective?"
2
DNA Sequencing (Healthy + Tumor)

Paul contacted UNSW's (University of New South Wales) Ramaciotti Centre for Genomics and spent approximately $3,000 USD to sequence both:

  • Healthy tissue DNA (from blood) -- as the "normal baseline"
  • Tumor tissue DNA (from the tumor) -- to find where the mutations are

Purpose of sequencing: compare the two DNA sets to identify tumor-specific genetic mutations (these mutations become the attack targets).

External help needed You need to find a genomic sequencing company or university lab. Taiwan has several options (see Section 7). Cost: approximately NT$50,000-150,000. You'll need a vet to collect tumor and blood samples.
3
Use AI to Analyze Mutations and Identify Neoantigens

With sequencing results in hand, Paul used his own data pipeline and ChatGPT to:

  1. Compare healthy vs. tumor DNA to find all mutation sites
  2. Filter for "meaningful mutations" (non-synonymous mutations that alter protein structure)
  3. Evaluate which mutant proteins are most likely to be recognized by the immune system (neoantigens)
Doable with tech background If you can write Python, you can use bioinformatics tools (like VarScan, MuTect) for mutation comparison. ChatGPT can guide you through each step. No biology degree needed, but patience and willingness to learn are required.
4
Use AlphaFold to Predict Protein 3D Structure

After identifying the mutant proteins, Paul used Google DeepMind's AlphaFold (a free tool) to predict the 3D structures of these mutant proteins.

Purpose: understand the shape of mutant proteins and determine which parts are "exposed on the surface" -- most easily recognized and attacked by immune system antibodies.

From dozens of mutation candidates, he selected the most promising neoantigens as vaccine targets.

Anyone can do this AlphaFold is free and open source. Google also provides AlphaFold Server (online version). You just input the amino acid sequence and it predicts the 3D structure. ChatGPT can teach you how to use it.
5
Design the mRNA Vaccine Sequence

Based on the selected neoantigens, Paul designed an mRNA sequence that, when injected into the body, would instruct cells to produce fragments of these mutant proteins, teaching the immune system to recognize and attack them.

Paul spent three months preparing a 100-page ethics review document and obtained ethics approval from UNSW.

Final output: a half-page mRNA formula -- this is the vaccine's "blueprint."

High difficulty but learnable This step requires understanding mRNA codon optimization, UTR design, and related concepts. ChatGPT can assist, but results need repeated verification. Paul's ML background helped him evaluate the reasonableness of AI output.
6
Vaccine Manufacturing (Lab Work)

UNSW RNA Institute Director Professor Pall Thordarson (a pioneer in nanomedicine) took Paul's design and synthesized the actual mRNA vaccine in the lab.

From receiving the sequence design to completing manufacturing: less than two months.

Not possible for individuals This is the biggest bottleneck. Synthesizing an mRNA vaccine requires specialized RNA lab equipment, lipid nanoparticle (LNP) encapsulation technology, cleanroom environments, and quality control. You must find a university or biotech company willing to collaborate.
7
Injection + Combined Immunotherapy

The vaccine was administered by a veterinarian, combined with a checkpoint inhibitor -- a drug that "releases the brakes" on the immune system, amplifying the immune response.

  • December 2025: First dose administered
  • February 2026: Booster dose
  • Result: 75% of tumors showed significant shrinkage
Requires vet cooperation The injection itself requires a veterinarian. Checkpoint inhibitors require a prescription. If you can find a cooperative veterinary oncologist, this step is feasible.

3. Treatment Timeline

2024

Rosie diagnosed with mast cell cancer. Chemotherapy and surgery failed. Vet gave her only months to live.

2024-2025 (approx. 3-6 months)

Paul used ChatGPT to develop plan -> DNA sequencing -> AI mutation analysis -> AlphaFold protein modeling -> mRNA sequence design -> prepared 100-page ethics review document

October-November 2025

UNSW RNA Institute began manufacturing the vaccine (completed in less than two months)

Early December 2025

Rosie received her first mRNA vaccine dose. At this point, her mobility had severely declined; she was starting to "shut down."

Late January 2026

Six weeks later, Rosie jumped over a fence at the dog park chasing a rabbit. Paul: "I cried right there on the spot."

February 2026

Second booster dose administered. Tumors continued shrinking.

March 2026

The tennis-ball-sized tumor on her leg shrank by 75%. Most tumors were shrinking. However, one tumor did not respond -- Paul is sequencing that tumor and preparing to design a second vaccine.

4. Tools & Technologies Used

ToolPurposeCostAccessible to anyone?
ChatGPTResearch planning, literature review, mutation analysis, mRNA design assistance$20/moFully accessible
AlphaFoldPredict 3D structure of mutant proteinsFreeFree & open source
Genomic sequencing serviceWhole genome sequencing of healthy DNA + tumor DNA$3,000 USDRequires external service
Custom data pipelineProcessing genomic data, comparing mutationsFree (self-built)Requires coding skills
Bioinformatics tools
VarScan / MuTect / GATK etc.
Variant detectionFree & open sourceRequires learning
RNA laboratorymRNA synthesis, LNP encapsulationUndisclosedRequires lab access
Checkpoint inhibitorEnhance immune responseUndisclosedRequires prescription

5. Cost Breakdown

Known Costs

  • DNA sequencing: $3,000 USD (~NT$90,000)
  • ChatGPT Plus: ~$20/mo x 6 months = $120
  • AlphaFold: Free

"Design side" total: approx. $3,200 USD (~NT$100,000)

Undisclosed Costs

  • mRNA vaccine manufacturing fees (unclear if UNSW charged)
  • Checkpoint inhibitor drug costs
  • Veterinary consultation and injection fees
  • Ethics review administrative costs

"Manufacturing + treatment side" estimate: tens of thousands to hundreds of thousands NT$

6. Can an Ordinary Person Do This? Step-by-Step Feasibility

StepDifficultyFeasible for individuals?What you need
1. Use AI to develop planLowYesChatGPT + good questioning skills
2. DNA sequencingMediumOutsource neededFind a sequencing company + vet for sampling
3. Mutation analysisMedium-HighLearning requiredPython + bioinformatics basics
4. AlphaFold protein modelingMediumYesAlphaFold Server + ChatGPT guidance
5. mRNA sequence designHighVerification neededChatGPT assistance + expert review
6. Vaccine manufacturingVery HighNo -- lab requiredFind a cooperative RNA laboratory
7. Injection + monitoringMediumVet neededVeterinary oncologist cooperation

Verdict: Design Is Doable; Manufacturing Is the Bottleneck

Steps 1-5 (research plan -> DNA analysis -> vaccine design) can indeed be done independently given a technical background. Paul is living proof.

But Step 6 (manufacturing) is an insurmountable barrier for individuals -- you must find a lab willing to collaborate. Paul succeeded because a UNSW professor was convinced by his data.

So the real question isn't "am I capable of designing it," but "can I find a lab willing to manufacture it for me?"

7. How to Do This in Taiwan? Local Resource Inventory

DNA Sequencing Services

Estimated cost: Whole genome sequencing NT$50,000-150,000 (depending on depth and scope)

RNA / mRNA Research Capabilities

Key challenge: Convincing a lab to synthesize an mRNA vaccine for your "pet" case. Academic labs may be more willing to frame it as a "research collaboration."

Veterinary Oncology Specialists

8. The Science: Why This Approach Works

How mRNA Vaccines Work

  1. mRNA is injected into the body -> absorbed by cells
  2. Cells follow mRNA instructions to produce fragments of the tumor's mutant proteins (neoantigens)
  3. The immune system detects these "foreign proteins" -> generates targeted antibodies and T cells
  4. These immune cells search the entire body to find and attack cancer cells carrying the same mutant proteins
  5. Combined with checkpoint inhibitors -> disables the cancer cells' "cloaking device" against the immune system

In essence, it's "teaching the immune system what the enemy looks like, then unleashing it to hunt."

Why Does It Need to Be Personalized?

Every tumor's genetic mutations are different, even within the same type of cancer. So generic vaccines have limited effectiveness. A personalized vaccine is designed based on "this specific patient, this specific tumor's" unique mutations, achieving far greater precision than any one-size-fits-all approach.

This is also the technology that Moderna and BioNTech are spending billions of dollars to develop -- and Paul used AI to accomplish the same design work in just a few months.

9. Limitations & Risks

Facts We Must Honestly Face

10. Historical Parallels

Hua Tuo and "Fighting Poison with Poison" -- A Millennium of Immunotherapy Intuition

Hua Tuo, a legendary physician in ancient China (~200 AD), is regarded as a pioneer of surgery. But the deeper metaphor is this: the ancient Chinese already had the concept of "fighting poison with poison" -- using small amounts of the "enemy" to train the body's defense system. Variolation (the practice of inoculating with smallpox material) has centuries of history in China, and its essence is "showing the enemy's features to the immune system."

What Paul did is perfectly consistent with this ancient intuition -- the only difference is that the tools evolved from "ground-up smallpox scabs" to "AI-designed mRNA sequences." The principle hasn't changed; the tools have evolved.

Zhuge Liang's "Empty Fort Strategy" -- Maximum Impact with Minimum Resources

Zhuge Liang (the legendary strategist of China's Three Kingdoms era, ~200 AD), facing Sima Yi's massive army with no troops of his own, used "information asymmetry" to make his opponent retreat on their own. Paul faced the multi-billion-dollar pharmaceutical industry without a lab, without medical training, without a research team -- but he had AI, a tool that made him "look like he had an entire research team."

The essence of AI is giving individuals institution-level capabilities. This is the most profound significance of this case.

11. Business Insights

1. Personalized Pet Medicine Is a Blue Ocean Market

The global pet healthcare market exceeds $30 billion, and pet cancer treatment is the fastest-growing segment. Yet there are virtually no "personalized gene therapy" services available. If Paul's process could be productized (design-side SaaS + manufacturing partner network), this represents a massive opportunity.

2. The "AI + Biotech" Democratization Platform

UNSW Professor Thordarson said: "This technology is democratizing the entire process." Imagine a platform where: users upload genomic sequencing data -> AI automatically analyzes mutations -> automatically designs vaccine sequences -> connects to RNA manufacturing labs. This would be "the Shopify of gene therapy."

3. Separating "Design" from "Manufacturing" = A New Business Model

Paul proved that vaccine design can be done by AI + an individual, while manufacturing is handed off to specialized labs. This "fabless model" is already mature in the semiconductor industry (TSMC = foundry, MediaTek = design). The fabless era of biotech is dawning.

Sources: Fortune | Newsweek | Yahoo Finance | The Conversation | Creati.ai | Today Show Australia transcript