Biomolecular Structure Prediction and Co-Folding
Predicting the 3D structure of proteins and their complexes with other proteins, DNA, RNA and drug molecules.
Tier Grades
Tier A: Frontier AI labs
feasible
Google DeepMind, ByteDance and the Boltz team train co-folding models.
Tier B: Indonesian AI labs
stretch
Training is not sourced, but open Boltz and Protenix weights can be run and fine-tuned.
Tier C: Institution units such as ITB
feasible
Running open Boltz or Protenix for local targets fits a university structural biology unit.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
stretch
Inference on open weights supports a drug-discovery proof of concept, but needs a pharma partner with targets.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeProtenix-Mini cuts inference cost with a few-step ODE sampler, pruned blocks and an ESM module in place of the MSA; training compute not stated.
- DataBoltz-2 jointly models complex structures and binding affinities; Boltz-1 is described as the first fully open model to approach AlphaFold3 accuracy.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 6 authors on the Protenix-Mini paper (arXiv 2507.11839).
- Time horizonYears.
- Shelf-life signalsBoltz-1 is described as the first fully open model to approach AlphaFold3 accuracy, and Boltz-2 adds binding affinity.
- Pipeline roleTier A and startups release co-folding models; Tier C runs them on local targets; Tier D could support screening proofs of concept.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Google DeepMind
Isomorphic Labs
Boltz PBC
MIT (Jameel Clinic)
ByteDance
Chai Discovery
OpenFold consortium
Microsoft AILinked models
Protenix-v1scientificexemplarBoltz-2scientificexemplarChai-1scientificexemplarAlphaFold 3scientificexemplar
ESMFold2scientificproduced byProtenix-v2scientificproduced byOpenFold3 (preview)scientificproduced byProtenix-Mini / Mini+scientificproduced byBioEmuscientificproduced byProtenixscientificproduced byBoltz-1scientificproduced byAlphaFold 2scientificproduced byLinks
R&D TopicBiomolecular Structure Prediction and Co-Folding
ModelsAlphaFold 3Google DeepMind, scientificBoltz-2Boltz PBC, scientificProtenix-v1ByteDance, scientificChai-1Chai Discovery, scientific+8 more
AlphaFold 3Google DeepMind, scientificBoltz-2Boltz PBC, scientificProtenix-v1ByteDance, scientificChai-1Chai Discovery, scientific+8 moreWho Runs It
PfizerContract signed, 8 Jan 2026
PfizerContract signed, 8 Jan 2026Models 12
Boltz-2Boltz PBC, MIT (Jameel Clinic), Recursion, scientificexemplarAlphaFold 3Google DeepMind, Isomorphic Labs, scientificexemplarBoltz-1Boltz PBC, MIT (Jameel Clinic), scientificproduced byProtenixByteDance, scientificproduced byProtenix-v1ByteDance, scientificexemplarProtenix-v2ByteDance, scientificproduced byAlphaFold 2Google DeepMind, scientificproduced byChai-1Chai Discovery, scientificexemplarESMFold2EvolutionaryScale, scientificproduced byProtenix-Mini / Mini+ByteDance, scientificproduced byBioEmuMicrosoft AI, scientificproduced byOpenFold3 (preview)OpenFold consortium, scientificproduced bySources
- Protenix-Mini: Efficient Structure Predictor via Compact Architecture, Few-Step Diffusion and Switchable pLM arxiv.org https://arxiv.org/abs/2507.11839
- raw.githubusercontent.com/jwohlwend/boltz/main/README.md https://raw.githubusercontent.com/jwohlwend/boltz/main/README.md
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Biomolecular Structure Prediction and Co-Folding", https://aurion.semai.id/topics/biomolecular-structure-and-co-folding/, records of October 2026.