Use Promera Online

Promera: Unified Biomolecular Structure Prediction, Filtering, and Design

Promera is an open-source, all-atom generative model that unifies biomolecular structure prediction, binder filtering, and controllable design into a single pipeline. Developed by researchers at MIT and UT Austin, Promera overcomes the key limitations of traditional co-folding models by integrating state-of-the-art confidence metrics for highly accurate binder discrimination and controllable, targeted design.

Whether you are performing structural modeling, filtering variants from thousands of samples, or engineering de novo binders, Promera offers a high-signal open-source solution optimized for drug discovery and protein engineering workflows.

Key Features & Capabilities

High-Accuracy Co-Folding

Promera delivers exceptional all-atom structure prediction across therapeutically relevant molecular interaction categories. Its performance exceeds popular open-source alternatives like OpenFold3-p2 and Boltz-2 in crucial categories including:

• Protein-Protein Interactions

• Antibody-Antigen Complexes

• Protein-Ligand Interactions

Superior Binder Filtering and Enrichment

Traditional structure prediction models struggle to differentiate true binders from non-binders. Promera introduces optimized confidence metrics engineered specifically to solve this bottleneck in computational workflows:

• Miniprotein Filtering: Promera’s Interaction Prediction Score from Aligned Errors (ipSAE) yields an AUROC of 0.70 on minibinder datasets, achieving up to a 5x enrichment score at ~10% recall.

• Nanobody Filtering: Introducing the Interface Contact Score (iCS). By classifying predicted contacts as correct or incorrect, iCS achieves up to a 20-fold enrichment of binding pairs at 10% recall—doubling the success rates expected from standard metrics like AF-Multimer ipTM.

Controllable De Novo Design

Promera acts as a versatile design tool by predicting structured backbones for masked sequences using training-time masking protocols. It enables precise biophysical steering through multiple customizable constraints:

• Epitope and Paratope Flagging: Steer binders directly toward a target interface or restrict contacts to specific complementarity-determining regions (CDRs).

• Template Conditioning: Constrain generations to a target receptor geometry.

• Unprecedented Success Rates: Promera significantly outperforms existing generative tools like BoltzGen and matches or exceeds backprop-based hallucination techniques (such as mBER) under co-folding confidence filters.

Real-World In Silico Case Studies

Epitope Targeting of Viral Glycoproteins

Promera was deployed to target varied, spatially separated epitopes on the Andes hantavirus Gn/Gc glycoprotein complex—a major public health threat mediating viral entry. Using epitope conditioning, Promera successfully designed confidently predicted VHH binders across six distinct target epitopes.

Active State Stabilization of GPCRs

Functional antibody design often requires stabilizing a target in a specific conformational state. Promera targeted the extracellular surface of the $\beta_{2}$ adrenergic receptor $(\beta_{2}AR)$. While the receptor fluctuates dynamically in isolation, Promera-designed nanobodies successfully eliminated the inactive structural mode, locking crucial core toggle residues into their active orientation.

What is Tamarind Bio?

Tamarind Bio is a computational biology platform dedicated to making advanced biomolecular modeling tools accessible, scalable, and easy to integrate into modern research workflows. By offering cloud-based deployment of open-source artificial intelligence architectures, Tamarind Bio enables web-based execution, high-throughput batching, and streamlined visualization for structural biology and drug discovery pipelines without requiring specialized local hardware or deep software engineering overhead.

How to Use Promera on Tamarind Bio

Running your structure prediction and binder design campaigns with Promera on Tamarind Bio is simple:

1. Select Your Task Mode: Choose between Co-folding (Structure Prediction) or Binder Design depending on your project goal.

2. Input Molecular Identifiers: Provide your target amino acid sequences, RNA/DNA sequences, or small molecule identities.

   • For design tasks: Provide your binder framework sequence (e.g., a nanobody framework) and mask the specific residues or CDRs you wish to design.

3. Configure Constraints (Optional): Define optional epitope engagement tags, paratope flags, or structural template distograms to condition the generation process.

4. Run Inference Scaling: Set your desired seed and diffusion sample count. Promera benefits from inference scaling; increasing your sample count (e.g., generating 25 or more samples) steadily drives protein-ligand and antibody-antigen success rates higher.

5. Generate Sequences & Structures: Promera will output the atomic 3D coordinates of the complex structure. For design tasks, the structured backbone coordinates of your masked regions are automatically finalized into concrete sequences using an integrated inverse folding model (ProteinMPNN).

6. Filter with High-Signal Metrics: Download your structural ensemble alongside fully tabulated ipSAE, ipTM, and interface contact scores (iCS) to confidently select the top-ranked candidates for downstream experimental validation.

Source