Iris
← Keeping up with Protein Design - Monthly Roundups by Iris
October 2026 roundupPeriod scanned: 2 October – 1 November 20264 min listen

Nvidia and UW release RFD4-Proteina for protein design

Audio briefing

4 min

In this report
October's developments move protein design beyond structure generation towards functional testing, programmable antibody engineering and experimental structure constraints. Several systems produced lab-validated proteins, but the evidence remains assay-specific rather than a general demonstration of reliable design at scale. The month also brought separate advances in sequence optimisation and multi-component antibody generation.
  1. Nvidia and the University of Washington released RFD4-Proteina, reporting 5 to 10-fold gains in design success and speed across several tested tasks:
  2. Pinal linked natural-language functional descriptions to designed sequences and structures, with experimental tests including a PET hydrolase, an alcohol dehydrogenase and a 1.7-fold titre increase:
  3. De novo cysteine proteases reached rate enhancements of up to 3 × 107, with six crystal structures matching their design models and activity in mammalian cells:
Generative design

Nvidia and UW release RFD4-Proteina for design

General
Nvidia and the University of Washington's Institute for Protein Design released RFD4-Proteina, an open model that jointly refines protein sequences, structures and interacting molecules. The reported 5 to 10-fold gains in design success and generation speed were accompanied by laboratory tests across binding, scaffolding and DNA-binding tasks.
  • The system combines RosettaFold Diffusion with Proteina-Complexa and is available through Amazon, Google Cloud and Hugging Face.
  • The reported validation spans protein binding, small-molecule binding, enzyme scaffolding and DNA binding, but the release does not establish how these results transfer across broader design targets.
RFD4-Proteina extends diffusion-based structure generation into concurrent sequence and interaction design, with an open distribution model.

Pinal designs proteins from natural-language functions

General
Pinal is a 16-billion-parameter model that turns natural-language functional descriptions into protein sequences and structures. Experimental tests of its candidates included a PET hydrolase and an alcohol dehydrogenase, while a designed H-protein increased pathway product titre 1.7-fold against the native E. coli protein.
  • Pinal was trained on 1.7 billion protein-text pairs and was evaluated for foldability, alignment with the requested function and sequence diversity.
  • The titre result is reported for a CO2-fixation pathway; it is evidence for a particular engineered context, not a general measure of natural-language design accuracy.
Pinal makes functional intent the starting constraint for sequence and structure generation rather than requiring manual specification of those constraints.

Researchers design cysteine proteases from scratch

General
Researchers designed cysteine proteases with rate enhancements up to 3 × 107 over the uncatalysed reaction. The proteins have folds distinct from natural proteases, and their activity in mammalian cells gives the design result a functional test beyond structural agreement.
  • Thirteen of 69 designs were successful, and six crystal structures matched the design models at Cα RMSD below 1.2 Å.
  • The designed enzymes remain less active than native cysteine proteases, with the authors attributing the gap to imperfect active-site preorganisation and substrate positioning.
The study tests whether de novo structural design can produce catalytic function, while making the remaining active-site limitations explicit.

RFdiffusion3 designs proteins and nucleic acids atomwise

General
RFdiffusion3 applies atom-level generative design to proteins and nucleic acids, with control over molecular interactions and catalytic-site organisation. Laboratory experiments reported functional designs including a paraoxon-degrading enzyme and a protease targeting transthyretin amyloidosis.
  • The reported scope includes protein structures, enzymes, custom ribozymes and RNA sensors.
  • Its biological evidence is task-specific, so the experiments show that atom-level control can yield working examples rather than establishing uniform performance across the design space.
RFdiffusion3 broadens generative biomolecular design from protein backbones to coordinated atom-level designs across proteins and nucleic acids.
Antibody and structural methods

AbOptimizer edits antibody sequences for multiple properties

General
AbOptimizer uses reinforcement learning and protein language models to edit several antibody sequence sites in parallel. In vitro variants improved affinity and specificity in emibetuzumab and siltuximab, and enabled indel-inclusive affinity optimisation for trastuzumab.
  • The framework first identifies candidate editing sites and then recommends mutations against multiple property objectives.
  • Its reported in-silico advantage over baseline methods is complemented by selected in-vitro variants, but the experiments cover three antibody contexts rather than a broad therapeutic panel.
AbOptimizer addresses antibody engineering as a multi-property sequence-editing problem rather than a single affinity optimisation task.

OFAntibody generates monoclonal and bispecific antibodies

General
OFAntibody is an all-atom generative framework for epitope-conditioned monoclonal and bispecific antibody design. The preprint reports a Top-5 enrichment rate of 41.5%, 5.39 times the result for RFantibody, alongside generation across several multi-component formats.
  • Training uses distilled antigen-antibody complexes and multi-component structural supervision to model CDR-epitope interactions.
  • The supported formats include tandem VHH, diabody and CODV designs; the reported evidence is a benchmark result from a preprint, not a demonstration of therapeutic activity.
OFAntibody extends de novo antibody generation from binary complexes to molecules designed to recognise distinct targets within one construct.

CryoBoltz guides structure prediction with cryo-EM density

General
CryoBoltz combines cryo-EM density maps with structure-prediction models to steer sampling towards experimentally consistent conformations. It is designed to build atomic models into heterogeneous maps, including dynamic transporters and antibodies.
  • The method derives global and local structural constraints from density maps and applies them during prediction rather than using prediction alone.
  • Its stated scope is heterogeneous cryo-EM modelling across dynamic systems; the report does not provide a general accuracy estimate comparable with a benchmark across map qualities.
CryoBoltz connects experimental density constraints to generative structure sampling, narrowing the gap between cryo-EM observation and atomic modelling.
How Iris covered your topics this month
Every topic you track was searched over the last month.
TopicSources foundIn this briefing
General327 stories
New stories that we're watching. Keep or drop these on your dashboard.
Generative protein designEnzyme and antibody engineering
Appendix: 32 sources relevant to you, from 51 scanned
General · 27
CBAT-Net: a unified CNN–BiLSTM–MultiHead transformer architecture for joint Q3/Q8 protein secondary structure prediction | Scientific Reports · nature.com · 2026-10-27Research
EquiDesign: a versatile computational framework for de novo protein sequence design | Briefings in Bioinformatics | Oxford Academic · academic.oup.com · 2026-10-26Research
A Compact Generative Language Model Learns the Compositional Grammar of Phase-Separating Proteins | Chemical Science | The Royal Society of Chemistry · pubs.rsc.org · 2026-10-26Research
Multiproperty optimization of therapeutic antibodies through sequence editing via large language model and reinforcement learning | Nature Communications · nature.com · 2026-10-26Research
Nvidia, UW Institute for Protein Design Unveil Open-Source Protein Design AI Model · genengnews.com · 2026-10-22 · 6 outletsAnnouncement
Programmable design of functional proteins from natural language | Nature · nature.com · 2026-10-21 · 6 outletsResearch
Does a Fab antibody need a specialist model? | Cradle · cradle.bio · 2026-10-20News
RFdiffusion3 designs proteins and nucleic acids atom by atom with striking · bioengineer.org · 2026-10-16News
Programming biological behavior with AI-guided de novo protein design: Molecular Cell · cell.com · 2026-10-15Research
Biologically-driven generative chemistry: using biological data to guide de novo drug design | Digital Discovery | The Royal Society of Chemistry · pubs.rsc.org · 2026-10-15Research
InstructPro2: Natural Language and Biological GroundingGuided De Novo Protein Design · arxiv.org · 2026-10-14Research
Efficient Protein Structure Prediction from Desktop to Datacenter with OpenFold-TRT | Computational and Structural Biotechnology Journal · spj.science.org · 2026-10-14Research
HDOCK-Multimer: improved structure prediction of large protein complexes by integrating docking and combinatorial assembly | Nature Computational Science · nature.com · 2026-10-14Research
LoL-align: sensitive and fast probabilistic protein structure alignment | Nature Methods · nature.com · 2026-10-14Research
De novo design of cysteine proteases | Nature · nature.com · 2026-10-14Research
Designing Protein Structures From Natural Language - Astrobiology · astrobiology.com · 2026-10-14News
BioByte 174: Protein Design - by Pranay Satya · decodingbio.substack.com · 2026-10-14News
Orthogonal Fc engineering enables multiplexed delivery of ... · science.org · 2026-10-14Research
FunFOLD5: combining localised templates and structural homology for accurate predictions of biologically relevant protein–ligand interactions | Bioinformatics | Oxford Academic · academic.oup.com · 2026-10-10Research
De novo protein design in an expanded chemical space · biorxiv.org · 2026-10-09Research
Advances in Engineering Immune Proteins as Therapeutics - ScienceDirect · sciencedirect.com · 2026-10-09Research
Multiscale guidance of protein structure prediction with heterogeneous cryo-EM data | Proceedings of the 39th International Conference on Neural Information Processing Systems · dl.acm.org · 2026-10-08Research
Machine Learning Empowers Enzyme Discovery, Structure ... · pubs.acs.org · 2026-10-07Research
De novo design of monoclonal and bispecific antibodies with OFAntibody · arxiv.org · 2026-10-07Research
A systematic review of protein language models for protein structure prediction and a proposed biological readiness taxonomy | Discover Artificial Intelligence | Springer Nature Link · link.springer.com · 2026-10-05Research
Generating Structural Ensembles of Disordered Proteins ... · biorxiv.org · 2026-10-04Research
Developability-by-design for bispecific antibodies in discovery: combining computational prediction and early screening for candidate selection: Expert Opinion on Biological Therapy: Vol 26 , No 10 - Get Access · tandfonline.com · 2026-10-03Research
19 more scanned and set aside
Politics: Latest and breaking political news today - POLITICO · politico.com · 2026-11-01News
UK | Latest News & Updates | BBC News · bbc.com · 2026-10-31News
NBC News - Breaking Headlines and Video Reports on World, U.… · nbcnews.com · 2026-10-29News
News | Today's latest from Al Jazeera · aljazeera.com · 2026-10-27News
Associated Press News: Breaking News · apnews.com · 2026-10-21News
antibody-service · exreprotein.com · 2026-10-21News
Latest news & breaking headlines · thetimes.com · 2026-10-21Low signal
Antibody-Antibiotic and Antibody-Enzyme Conjugate Design – C… · creative-biolabs.com · 2026-10-19News
News and current affairs from Germany and around the world · dw.com · 2026-10-19News
The 2026 Kofi Annan Geneva Peace Address: Choose Peace - Kof… · kofiannanfoundation.org · 2026-10-16News
AI-driven de novo protein design coupled with high-throughpu… · researchoutput.ncku.edu.tw · 2026-10-15Research
secondary structure prediction peptide · cetaf.org · 2026-10-12News
Protein Engineering Market Size, Share, Growth, Report, ... · fortunebusinessinsights.com · 2026-10-12Low signal
protein structure prediction__2026_v3 · web.stanford.edu · 2026-10-06Research
Protein Structure Prediction Activity - Chemistry LibreTexts · chem.libretexts.org · 2026-10-06News
AI in Protein Engineering Market worth $3.82 billion by 2031 · marketsandmarkets.com · 2026-10-05Low signal
Xinhua world news summary at 0030 GMT, Oct. 2 -Xinhua · english.news.cn · 2026-10-02News
TOLOnews - 6pm News - 02 Oct 2026 | TOLOnews · tolonews.com · 2026-10-02News
Thousands of troops head to Mideast, Christa Pike in critica… · apnews.com · 2026-10-02News
Tell Iris what to cover more, less, or differently. It shapes your next briefing.
This briefing was generated by Iris Briefings.
Iris uses artificial intelligence to synthesise recent developments in your field. It can make mistakes, so please verify anything critical against the linked primary sources before acting on it. Thank you for supporting Iris.

Want this for your field?

Iris builds a briefing agent around your own profile: your field, your focus areas, your context. Delivered on your schedule.

Get your own briefing

Iris is a tool that uses AI to find publicly available information and extract the insights that matter for you. Iris Briefings have taken precautions to ensure our systems are robust but these are not infallible. Nothing in your briefings should be taken for medical, legal or professional advice and Iris cannot take responsibility for your decisions. We encourage you to do your own due diligence on the primary sources that Iris references for critical decisions.