Part VI · 3 — Life Sciences

The domain where AI has already changed science — from protein structure (2024 Nobel) to DNA at chromosome scale.

🎨 Figure F-VI.3 — Life, folded and generated. Brief: a protein folding (AlphaFold) and a DNA strand being "read" and generated by a model; in the background, a drug-discovery funnel shortening. Compendium palette.

3.1 Protein structure and design

• AlphaFold 3 (Hassabis + Jumper, 2024 Nobel Prize in Chemistry) — structure

  prediction; transformed structural biology.

• ESM-2/3 — protein language models; fine-tuning ESM-2 for a specific

  domain is feasible on an RTX 4090.

• RFdiffusion / ProteinMPNN — protein design (inverse folding).

3.2 DNA at scale

• Evo 2 (Arc Institute, Nature 2026) — 9T base pairs, 100,000+ species, 40B

  parameters, 1M nucleotides of context; the first model to generate DNA at functional chromosome scale.

3.3 Drug discovery

AI cuts the drug-discovery cycle from ~US$2B and 10–15 years to ~US$500M and 3–5 years in niches. Reference companies: Isomorphic Labs, Recursion, Exscientia, Xaira Therapeutics.

Opportunity (BR): design of tropical enzymes with 1–2 RTX 4090 + a partnership with a lab (UFRJ, UNICAMP, UFMG, Embrapa), on top of ESM-2/RFdiffusion. The real asset is the partnership with the laboratory (data), not the GPU.