Use PepMimic Online

PepMimic: AI-Powered Peptide Design through Binding Interface Mimicry

Transform known protein receptors, antibodies, or AI-generated binders into high-specificity short peptide therapeutics with atomic precision.

What is PepMimic?

PepMimic is a breakthrough geometric deep generative artificial intelligence model built to achieve the sequence and structure co-design of all-atom peptide binders.

Traditional small-molecule discovery and biologic engineering struggle to effectively mimic discontinuous or complex binding surfaces. PepMimic solves this bottleneck using a novel all-atom latent diffusion process. By conditioning its generation on a target protein and an existing binder (such as a natural receptor, antibody, or nanobody) , PepMimic automatically learns key interaction patterns to connect amino acids into a single, high-affinity peptide structure.

Key Performance Capabilities

• Atomic Precision Sequence & Structure Co-Design: Captures the true joint distribution of amino acid types and all-atom geometries within a well-formed latent space. It models essential side-chain geometry directly, outperforming backbone-only alternatives.

• Unprecedented Wet-Lab Success Rates: In robust Surface Plasmon Resonance imaging (SPRi) experiments, PepMimic generated binders for critical drug targets like PD-L1, CD38, BCMA, HER2, and CD4, achieving target hits up to 20,000 times higher than random experimental library screening.

• Validated In Vivo Targeting: Top-ranked PepMimic designs demonstrate effective cell membrane binding and exceptional contrast accumulation at tumor sites in vivo during preclinical diagnostic and imaging evaluations.

How PepMimic Works

PepMimic operates using an innovative computational framework comprised of three core modules:

1. All-Atom Autoencoder: Establishes a completely reversible, standardized mapping from all-atom geometries of peptides to compressed residue-level latent representations.

2. Latent Diffusion Model: Learns the joint distribution of both sequences and all-atom structures, transforming standard Gaussian noise into meaningful latent point clouds conditioned on a specific target binding site.

3. Latent Interface Encoder (Gradient Guidance): Projects reference binding interfaces into a numerical vector space. By calculating vector distances, the encoder guides the generative diffusion process using backpropagation gradients to align the generated peptide with the exact reference interaction footprint.

How to Use PepMimic on Tamarind Bio

Running your first peptide-mimicry design campaign takes only a few simple steps through our web interface:

1. Upload Target Complex: Provide a solved 3D complex structure (PDB format) containing your target protein and an existing binder template (e.g., an antibody, nanobody, or natural receptor).

2. Select the Interface: Highlight the structural residues forming the binding site to establish the structural conditions for the model.

3. Configure the Run: Specify your desired peptide length constraints (supporting sequences from 4 to 25 residues).

4. Generate & Rank: The generative diffusion model will co-design and relax candidate structural variations. Results are organized dynamically utilizing our prioritized ranking strategy across Rosetta interface energy (ΔG), FoldX energy, and AlphaFold Multimer pLDDT metrics to bubble up the highest-probability wet-lab candidates.

What is Tamarind Bio?

Tamarind Bio is a premier web platform dedicated to making state-of-the-art computational biology and molecular design tools accessible to therapeutic researchers, medicinal chemists, and biologists worldwide. By removing complex local hardware dependencies and command-line execution requirements, Tamarind Bio wraps groundbreaking algorithms - like PepMimic - into an intuitive, cloud-accelerated environment. Our mission is to democratize high-resolution structural biology, bridging the gap between advanced deep generative AI models and immediate benchtop experimentation.

Frequently Asked Questions

Can PepMimic design peptides for drug targets that have no known natural binders?

Yes. For targets lacking established binders, PepMimic introduces a dual AI design paradigm. Users can first utilize protein binder design systems like RFDiffusion to create initial artificial mini-proteins, and then pass those high-pLDDT mini-binders into PepMimic to generate streamlined, lower-immunogenicity short peptide analogs.

Does PepMimic require retraining for every new target or binding interface?

No. The latent space guidance mechanism endows PepMimic with exceptional generalizability across entirely distinct interfaces, avoiding the need for target-specific tuning or model retraining.

What types of binding interfaces are easiest to mimic?

Preclinical evaluations reveal that interface mimicry success correlates strongly with compact binding interactions. Tighter interfaces with lower average distances between residues and a higher number of distinct interaction segments are optimally positioned for successful generative outputs. Additionally, utilizing loop structures to mimic natural loops achieves the most reliable biophysical translation.

Source