Research · Marca 369 Bio
Research
Marca 369 builds and evaluates a compact molecular AI model in public. What follows is what is already measurable — not a claim of clinical use, and not a peer-reviewed paper list we do not have.
For research only, not for clinical decisions.
TDC · Hugging Face
Public artefacts
- 3
Interactive demo
A free Hugging Face Space (benfica306/marca369-bio) accepts SMILES and returns the three ADMET predictions the model is trained for.
- 6
TDC evaluation
Results on Therapeutics Data Commons for microsomal clearance, half-life and CYP3A4 substrate, reported as mean ± std over 5 seeds against the public leaderboard.
- 9
Shop + model link
Customers and members can request extra free predictions. The model remains research-use only.
3 · 6 · 9
Publication status
- There is no peer-reviewed Marca 369 paper published yet. When a preprint or article exists, it will be linked here with its DOI or repository URL.
- Benchmark numbers on this site refer only to the three TDC tasks named above. Other ADMET endpoints are not claimed.
- The model is not a medical device and must not be used for clinical decisions.
SMILES → ADMET
Method, in plain terms
01
Input
Molecular structure as SMILES.
02
Model size
About 2.8 million parameters, trained on the benchmark data for these tasks (no large pre-training corpus claimed).
03
Runtime
About 0.1–0.2 seconds per molecule on CPU in the published demo setup.
04
Limits
Domain shift, sparse labels and assay noise still apply. A strong TDC score does not guarantee behaviour on every proprietary series.
Every Marca 369 customer and member gets extra free predictions from our model.
Use or discuss the work
Try the public demo, request extra predictions as a customer or member, or write if you want to collaborate on evaluation — not on inventing results.