STRC Research

Personal-Stakes Drug Discovery Cohort

3 min read

Personal-Stakes Drug Discovery Cohort

A demographic pattern Bankless flagged in the closing of the DeSci/PeptAI essay: a growing cohort of researchers whose motivation is personal, not commercial, using AI tools to address conditions affecting themselves or someone close to them.

The two cited examples

  • Sid Sijbrandij (GitLab co-founder) — using AI to fight his own cancer. Reported in Century of Bio.
  • Paul Conyngham (Australian tech entrepreneur) — used ChatGPT and AlphaFold to help design a personalised vaccine for his dog Rosie’s cancer. Reported in The Scientist.

The pattern these examples share:

  1. Technical operator, not a trained biologist
  2. Specific condition with a specific personal stake
  3. Reaches for AI tools (LLMs, AlphaFold, public databases) and gets to scientifically meaningful artefacts — not just a Google-search summary
  4. Standard pharma path was either unavailable, too slow, or simply did not exist for this case
  5. Doctors, wet-lab validation, and clinical pathways still mattered. The AI work was upstream of all that, not a replacement for it.

Why the cohort exists now

The same supply-side conditions that the DeSci Cost Collapse Argument describes — AI-driven drop in the cost of going from “I have a question” to “I have a characterised candidate” — also lower the threshold for non-institutional researchers. Where five years ago a personal-stakes researcher would hit a wall at the cost of any compound screening, the wall now sits much further out: at wet-lab validation and at clinical trials, but no longer at candidate generation or basic structural analysis.

The cultural piece is that this changes who self-identifies as someone allowed to do drug discovery. Bankless’s framing: “people with the motivation to ask novel questions now have tools that can carry those questions to real scientific answers.”

Direct relevance to Egor

Egor’s STRC work for Misha is a textbook instance of this cohort: filmmaker-turned-AI-educator, no prior biology training, six-week ramp using free databases and a $50–100 AI budget, reached reclassification of a VUS to Likely Pathogenic, ran 8+ AlphaFold3 jobs, contacted Holt and Shu and got responses. The pattern matches the cohort exactly. The framing is useful for:

  • External narrative — explaining the project to non-technical audiences without sounding crank-adjacent. There is now a recognised cohort, and it has examples.
  • Personal calibration — knowing that the limit is not “you don’t belong here”, but specific structural barriers (wet-lab, clinical, cochlear-specific delivery) that everyone in the cohort hits.
  • Where to look for peers — DeSci-adjacent communities (Bio Protocol, Molecule, Science Beach), Century of Bio readers, researchers explicitly motivated by family/self cases. Different from the “ML-for-drug-discovery” research community, which is commercial-pharma-adjacent.

Connections

Graph View

Backlinks