Therapeutic Discovery — Foundation Learning Guide¶
For a reader with no background in this area. If you have watched the Foundations film series, this follows directly from it.
The one idea to hold¶
Most medicines work by locking onto one specific protein. If you know the protein's shape, you can search for molecules that fit it.
What this subject does with that idea¶
Designs new candidate molecules shaped to fit a specific protein, and scores how well each might bind.
It attacks the hardest question in drug discovery — which molecule to even try — at the very front of a 10-15 year pipeline.
Decision support for a qualified clinician — never autonomous diagnosis or prescribing. Every output on this page is intended to inform a clinician's judgement, not replace it.
What it cannot do¶
Molecule generation (MolMIM) and docking (DiffDock) are gated NVIDIA NIMs and are not installed. Candidates shown are pre-computed. This is the flagship demo and the easiest to overclaim.
Stating the limit is not a disclaimer bolted on the end — it is how you tell a tool that helps from a tool that misleads.
Vocabulary you will meet¶
| Term | Plain meaning |
|---|---|
| Capability | One named thing the platform can do, registered in capabilities.json |
live / planned |
Whether a capability actually answers today, or is intended |
| Decision support | Output that informs a clinician; it never decides |
| LIVE / REPRESENTATIVE / BURST | Whether a demo ran now, was pre-computed, or ran on remote GPUs |
Try it¶
The demo for this subject is E3:
.venv/bin/python scripts/run_demo.py E3
If it reports BLOCKED, the message names exactly what is missing. That is intentional — a demo
that cannot run says so rather than showing you something canned.