The Guardian of the Genome Needs an Army

A Manifesto for DrugDiscovery@Home

By Ashish Makani (tp53) · March 30, 2026

Also published on onco.substack.com

I never met my nani (maternal grandmother).

Leukemia took her before I was born, when my mother was barely a teenager. I grew up with the shape of that absence — the stories my mother told, the photographs on the wall, the holidays where "all my friends would go to naani ghar during summer holidays in school".

Years later, while in grad school at Purdue, I watched Randy Pausch deliver "The Last Lecture" at CMU. He was dying of pancreatic cancer at 47. He had months to live. And he stood in front of four hundred people and said: "We cannot change the cards we are dealt, just how we play the hand."

I have read (& re-read) Paul Kalanithi's When Breath Becomes Air — a neurosurgeon diagnosed with stage IV lung cancer at thirty-six, writing his way toward meaning as his body failed him. "You can't ever reach perfection, but you can believe in an asymptote toward which you are ceaselessly striving."

These weren't abstractions for me. They were instructions.

In 2020, my uncle was diagnosed with hepatocellular carcinoma — liver cancer. He's alive today because of immunotherapy, because of bevacizumab, because someone designed a molecule that tells his immune system to fight. Last year, two of my aunts were diagnosed — one with breast cancer, one with ovarian cancer. I've been at the hospital. I've been in the waiting rooms. I've been on the phone calls where the oncologist explains what's next.

I named myself tp53 online — after the p53 tumor suppressor gene, the "guardian of the genome." It's the most commonly mutated gene in human cancer. When p53 works, it kills damaged cells before they become tumors. When it doesn't, cancer grows.

My friends say I have an unhealthy obsession with cancer.

I'm proud of it.

The Problem: We're Losing a War of Attrition

Ten million people die of cancer every year. That's more than HIV, malaria, and tuberculosis combined. It's a 9/11 every eight hours, every day, forever.

We have drugs that work. Osimertinib, a targeted therapy for EGFR-mutant lung cancer, gives patients months or years they wouldn't otherwise have. Immunotherapy has turned some terminal diagnoses into chronic conditions. The science is real. The progress is real.

But here's what the press releases don't say: the cancer adapts.

In 12 to 18 months, most patients on osimertinib develop resistance. The cancer mutates — specifically, a single amino acid change called C797S — and the drug stops working. And right now, for that mutation, there is no approved drug. The oncologist looks at the patient and says: "We've run out of targeted options."

This isn't a failure of science. It's a failure of throughput. We know what mutations cause resistance. We have the tools to design new drugs. But designing, testing, and approving each new molecule takes years and costs billions. The cancer mutates faster than we can respond.

We are losing a war of attrition — not because we lack intelligence, but because we lack speed.

And then there's the other crisis.

Nipah virus kills 40-75% of the people it infects. It's a BSL-4 pathogen — so dangerous that you need a full spacesuit to handle it in a lab. There is no vaccine. There is no drug. The WHO lists it as a priority pathogen for pandemic preparedness. If Nipah ever achieved efficient human-to-human transmission, the consequences would dwarf COVID.

COVID taught us what happens when we're unprepared. The scramble for vaccines. The months of waiting. The millions who died while the pipeline churned.

What if we could have drug candidates designed on Day Zero?

The Insight That Changes Everything

Here's the thing almost nobody realizes:

Designing a drug candidate is expensive. Checking if it works is cheap.

Let me explain with an analogy. Imagine you need to write a novel. Writing the novel takes months of creative effort. But spell-checking it? That takes seconds. Any computer can do it.

Now imagine you could get a million people to each write one page. Most of the pages would be terrible. But you could spell-check all million in minutes and keep only the ones that are perfect.

That's exactly how drug design works now:

That's a 30x cost difference between designing and checking.

This isn't a new idea. It's the same structural property that powered one of the greatest citizen science projects in history.

The Precedent: One Million People Built the World's Most Powerful Computer

In the year 2000, a professor at Stanford named Vijay Pande had a crazy idea. He was inspired by Napster — the music-sharing service. If people could share music files peer-to-peer, why couldn't they share computing power peer-to-peer for science?

He created Folding@home — a project where anyone with a computer could donate their machine's idle time to simulate how proteins fold. Each person's computer would run one tiny simulation. The results would be stitched together into a complete picture.

It worked.

For twenty years, Folding@home grew. Then COVID hit. In February 2020, the project had about 30,000 active volunteers. By March, it had over one million. People trapped at home, terrified, wanted to do something. And they could: just install the software and let their computer fight the pandemic while they slept.

Together, those million volunteers built the world's first exascale computer — aggregate computing power greater than any supercomputer on Earth. Five times more powerful than Summit, the previous record holder.

And they used it. Folding@home simulations discovered "cryptic pockets" in the SARS-CoV-2 virus — hidden drug targets invisible to any other method. The COVID Moonshot project, powered by Folding@home, designed a patent-free protease inhibitor that is progressing toward clinical trials.

One million people donated their idle computers. They built the most powerful computer humanity had ever assembled. And they used it to fight a pandemic.

What if we did the same thing — but instead of simulating proteins, we designed them?

The Vision: DrugDiscovery@Home

Andrej Karpathy — former head of AI at Tesla, one of the most influential minds in machine learning — recently proposed something he called "AutoResearch at Home." His idea: a global swarm of AI agents, running on untrusted computers across the internet, each contributing to research in parallel.

"A global swarm of agents could potentially run circles around Frontier Labs by leveraging the massive amount of untrusted compute available across the Earth."

His key insight is about verification: while it takes 10,000 attempts to find a breakthrough, it's very cheap for a trusted central node to verify whether a single attempt worked.

This is exactly the verification asymmetry in drug design.

DrugDiscovery@Home takes this insight and applies it to the most urgent problems in medicine:

  1. You pick a "track" — a disease target. EGFR-C797S (drug-resistant lung cancer). Nipah virus. Influenza. Ebola.
  2. Your GPU designs antibody binders against that target using open-source AI tools (RFdiffusion, ProteinMPNN — the same tools that won a Nobel Prize).
  3. A trusted central server checks every candidate — scoring binding affinity, safety, and drug-likeness in seconds.
  4. The best candidates appear on a public leaderboard — open for the world to see, open for any lab to test.

Anyone with a GPU can contribute. A gamer's RTX 4090. A university's compute cluster. A cloud instance between jobs. Every idle GPU becomes a drug design factory.

This needs a dedicated organization — not a university lab (too slow), not a pharma company (too secretive). A Focused Research Organization: time-limited, mission-driven, open-source. Build the infrastructure, seed the community, then let it run.

Use Case 1: The Cancer Drug That Doesn't Exist Yet

Fifteen to twenty percent of lung cancer patients have EGFR mutations. That's hundreds of thousands of people every year, worldwide.

Osimertinib works — until it doesn't. The C797S mutation makes it fail, and there is no approved replacement.

DrugDiscovery@Home would create dedicated "tracks" for every known EGFR resistance mutation:

If 1,000 people donated GPU time for one day, we'd generate 10,000 candidate binders — each computationally verified for binding, selectivity, and drug-likeness. The top candidates would go to an experimental lab for physical testing.

Ten thousand shots on goal, in one day, for the cost of a few hundred dollars in electricity.

Use Case 2: Day Zero Pandemic Readiness

We can't predict when the next pandemic will come. But we can prepare.

DrugDiscovery@Home would pre-compute antibody candidates against every WHO priority pathogen: Nipah, Ebola, MERS, novel influenza strains. Design them now. Store them. Have them ready.

We already demonstrated this is feasible. Using the abdesign platform we built, we ran a campaign against the Nipah virus G protein on a single NVIDIA H100 GPU:

If — despite the best efforts of public health agencies — there is ever another pandemic, DrugDiscovery@Home will ensure that we have drug designs on Day Zero and medicines ready to combat pathogen pathology within a week or two of identification.

Not months. Not years. Weeks.

The Exhortation

My grandmother died of leukemia before I was born. My uncle fights liver cancer today. My aunts battle breast and ovarian cancer as I write this sentence.

I sat in Dave Patterson's course at Berkeley in 2011, reading "The Cartoon Guide to Genetics," and for the first time understood what DNA actually does. I watched Randy Pausch's last lecture and understood what it means to fight. I read Paul Kalanithi and understood what it means to lose.

And I decided: this is what I build for. Not chatbots. Not recommendation engines. Tools that fight cancer. Tools that prepare for pandemics. Tools that any lab on Earth can use.

There are GPUs sitting idle all over the world right now. Gaming PCs in bedrooms. Workstations in university basements. Cloud instances between jobs. Every one of them could be designing the next cancer drug.

It's time to get citizens involved.

It's time to leverage the compute lying unused all over the world to dramatically expand the arsenal of cancer drugs and bring succor to cancer patients and their families — everywhere.

It's time to build the infrastructure that ensures the next pandemic finds us ready — not scrambling, not waiting, not burying our dead while the pipeline churns.

This is not a pipe dream. This is not "pie in the sky."

Folding@home proved that citizen science works at exascale — one million people, the world's most powerful computer, a patent-free drug candidate heading to trials.

Karpathy proved that AI agents can research in parallel — thousands of directions simultaneously, verified by a single trusted node.

The tools — RFdiffusion, ProteinMPNN, Boltz-2, ESM-2 — are open-source. Free. Ready.

What if humanity made a collective resolve to end most major diseases in the next decade?

Not as a slogan. Not as a TED talk applause line. As an engineering project. With milestones. With benchmarks. With a leaderboard that the whole world can see.

The proteins that guard our genome — p53, BRCA1, the antibodies our immune system manufactures — have been doing this work for billions of years. They're good at it. But they need help.

The guardian of the genome needs an army.

Will you join?

tp53 (Ashish Makani) is a research engineer at KCDH-A, Ashoka University. MS Computer Engineering from Purdue. Grand Challenges India Award (Gates Foundation / BIRAC). Builder of rfab-harness, DADB, Virtual Tumor Board, and Onco-TTT. His friends say he has an unhealthy obsession with cancer. He's proud of it.

Full portfolio: inventcures.github.io/projects

Contact: spiff007@gmail.com | @tp53 | inventcures.github.io