Use Promera Online
Commercially Available Promera No-Code Web Server
Promera: Unified Model for All-Atom Structure Prediction, Binder Filtering and Protein Design
Promera is a unified generative model for biomolecular structure prediction, binder filtering and controllable design. It was developed by Bowen Jing, Mihir Bafna, Daniel Diaz, Adam Klivans and Bonnie Berger, with affiliations including MIT CSAIL, the MIT Department of Mathematics and UT Austin. The preprint does not list a code repository, model weights or license.
Design pipelines for binders such as miniproteins and nanobodies typically chain separate tools: one to generate candidates, another to predict the complex, and a third to filter. Promera combines all three roles in one model built on an AlphaFold3-style architecture (Pairformer, diffusion and confidence modules).
As a co-folding model it predicts all-atom structures of complexes. As a filter it scores whether a binder really engages its target. As a designer it predicts masked sequence positions, with optional epitope, paratope and template constraints.
Item | Detail from the paper |
|---|---|
Architecture | AlphaFold3-style Pairformer, diffusion and confidence modules |
Inputs | Protein, RNA/DNA or small-molecule identities, with design residues masked |
Optional conditioning | Epitope, paratope, template or partial distogram |
Outputs | Complex structure plus ipTM, ipSAE and interface contact score (iCS) |
Training data | PDB (2023-12-31 cutoff) plus monomer self-distillation from MGnify and AlphaFold DB |
How Promera Works for Co-Folding and Binder Design
Co-fold the target: the target is folded together with a masked binder.
Fill in sequences: the predicted backbone is converted to sequence with ProteinMPNN (soluble weights).
Constrain the design: nanobodies use the caplacizumab framework with random-length masked CDRs flagged as paratope.
Score confidence: a confidence module predicts whether each contact is correct, and iCS averages those probabilities over ground-truth interface contacts.
Promera Benchmark Results for Structure Prediction and Binder Discrimination
Test | Reported result |
|---|---|
1,605 PDB complexes from 2025 | Outperforms OpenFold3-p2 and Boltz-2 across interaction categories |
221 antibody-antigen complexes | Outperforms both; top-ranked-seed success at DockQ above 0.49 rises from 20% to 27% with more seeds |
Nipah minibinders (94 binders, 822 nonbinders) | ipSAE AUROC 0.70 versus 0.66 for the next best |
Shuffled nanobody-antigen pairs | ipSAE AUROC 0.64 versus 0.58; iCS gives about 18x enrichment at 10% recall |
Design Case Studies
About 40% of nanobody designs pass iCS above 0.6 for IL7Ra and InsulinR, and about 75% for PDGFRb. In an Andes hantavirus analysis, six of eight defined epitopes yielded confidently predicted VHH binders. For the beta2-adrenergic receptor, 18 of 25k designs passed the filters. All are in silico results.
What is Tamarind Bio?
Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.
The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:
A user-friendly graphical interface for setting up and launching experiments
A robust API for integration into existing research pipelines
An automated system for managing and scaling computational resources
Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.
Accelerating Discovery with Promera on Tamarind Bio
Nanobody design: generate CDR candidates against a target, with optional epitope constraints.
Minibinder triage: rank candidate binders with ipSAE and iCS to separate likely binders from nonbinders.
Epitope-directed targeting: steer binders to chosen epitopes, as shown for hantavirus.
Complex prediction: model protein, nucleic acid and small-molecule complexes, including antibody-antigen pairs.
How to Use Promera on Tamarind Bio
Open the tool: log in to tamarind.bio and open Promera.
Define the target: provide the target and partner sequences, or ligands, if exposed.
Set design options: mask residues to design and add epitope, paratope or template constraints, if exposed.
Choose seeds and samples: more seeds improved antibody-antigen success in the paper, if exposed.
Run the job: submit and wait for processing.
Download outputs: retrieve predicted structures and confidence scores.
Filter candidates: rank by iCS or ipSAE, then take top designs to downstream validation.
Things to Keep in Mind
Protein-protein accuracy trails Protenix-v1, while protein-ligand accuracy matches it.
Design success is judged mainly through co-folding confidence metrics, with no experimental validation reported.
The Nipah analysis may disadvantage Boltz-2, since its ipSAE was used to select competition sequences.
The scaling law rests on five models and is an observational fit.
The hantavirus analysis ignored glycosylation.
Source: Jing B, Bafna M, Diaz D, Klivans A, Berger B. Promera: a unified model for biomolecular structure prediction, filtering, and design. bioRxiv 2026. DOI: 10.64898/2026.06.07.729267. No code repository is listed in the preprint.