One-Person AI Pharma

Infrastructure that changes the economics

In 2006, AWS eliminated the need to buy servers. A developer could rent GPU by the second and deploy within hours. A similar compression is happening in biotech right now — except instead of serving web pages, you're designing proteins.

Cloud GPU (Modal) + protein design tools (biomodals) + automated lab (Adaptyv Bio) = the full stack for AI-driven binder discovery, accessible to a single researcher. — ~$1.7K for 3 rounds of design-validate-iterate vs. $1-3M for traditional CRO

Four-stage closed loop

wget https://files.rcsb.org/download/5JDS.pdb
grep "^ATOM.* A " 5JDS.pdb > 5JDS_chainA.pdb

GPU=A100 uvx modal run modal_bindcraft.py \
  --input-pdb 5JDS_chainA.pdb \
  --number-of-final-designs 5

experiment = client.experiments.create(
    assay_type="binding",
    target_id="comp-pdl1-human",
    sequences=[{"name":"d1","sequence":"QVQLV..."}]
)

# Filter and iterate
hits = [r for r in results if r.kd < 1e-8]
best = min(hits, key=lambda r: r.kd) if hits else None

Modal + biomodals: GPU supercomputer in your terminal

Modal provides serverless A100/L40S GPU compute billed per second. biomodals wraps the best open-source protein engineering tools into one-line Modal commands. Modal free tier: $30/month (~6 complete dry rounds).

Adaptyv Bio: the protein lab as an API

Protein expression, binding assays, real Kd — delivered by API. No lab bench required. — Adaptyv Bio, $8M seed, Lausanne 2025

{
  "sequence_name": "VHH-01",
  "target_name": "HER2 / ERBB2",
  "kd": 8.1e-10,         // M
  "kon": 2400000,         // M^-1 s^-1
  "koff": 0.0019,         // s^-1
  "binding_strength": "strong",
  "r_squared": 0.999
}

Real numbers

Five steps to your first design

curl -LsSf https://astral.sh/uv/install.sh | sh
python -m modal setup   # get token at modal.com ($30 free/month)
pip install adaptyv-sdk # API client for wet lab

git clone https://github.com/hgbrian/biomodals
cd biomodals

wget https://files.rcsb.org/download/5JDS.pdb
GPU=A100 uvx modal run modal_bindcraft.py \
  --input-pdb 5JDS.pdb --number-of-final-designs 5

from adaptyv_sdk import AdaptyvClient
client = AdaptyvClient(api_key="your_key")
exp = client.experiments.create(
    assay_type="binding",
    target_id="comp-pdl1-human",
    sequences=[{"name":"d1","sequence":"QVQLV..."}]
)

results = client.results.list(experiment_id=exp.id)
for r in results:
    print(f"{r.sequence_name}: Kd = {r.kd:.2e} M")

Everything in one place