Use PepMimic Online
Commercially Available PepMimic No-Code Web Server
PepMimic: AI Peptide Binder Design through Binding Interface Mimicry
PepMimic is a deep generative model that performs sequence and structure co-design of all-atom peptide binders. Given a target protein and a known binder, such as a receptor or an existing antibody, it designs a short peptide that mimics the binding interface. It was developed by researchers at Tsinghua University, Peking University, Helixon, Renmin University of China and collaborating hospitals, and published in Nature Biomedical Engineering.
Peptides are attractive as diagnostics and therapeutics because they are small, can reach crowded binding sites and are synthetically accessible. Finding a peptide that binds a chosen target with high affinity is difficult, however, and random library screening is slow and costly. PepMimic starts from a binder that is already known to work and transfers the essence of its interface to a compact peptide.
The authors also describe a route for targets that have no known binder: existing tools first generate a protein binder, and PepMimic then converts it into a peptide.
Property | Detail |
|---|---|
Model type | Deep generative model for all-atom peptide sequence and structure co-design |
Inputs | A target protein and a known binder (receptor or antibody) to mimic |
Output | Short peptide binders with designed sequence and structure |
Targets tested | PD-L1, CD38, BCMA, HER2 and CD4 |
Training and benchmark data | PepBench, released on Zenodo |
Code | Available on GitHub, with checkpoints on Zenodo |
Why Interface Mimicry Matters for Peptide Binder Design
Many therapeutic targets already have an antibody or natural receptor that binds them well, but these molecules are large. PepMimic reuses the knowledge in those complexes, so the design problem becomes how to preserve the key interactions in a much smaller molecule. According to the paper, designs made this way produced strong binders at a rate well above random library screening under identical conditions.
How PepMimic Designs Peptides from Known Binders
The paper describes the method through several components:
Conditional generation: the generative model is conditioned on the target protein and a known binder, and it designs the peptide sequence and all-atom structure together.
Latent interface encoder: the interface of the known complex is encoded so that the generated peptide can reproduce its key features.
AlphaFold-based filtering: a filtering pipeline screens candidates before they are selected for synthesis.
Experimental validation: peptides were tested by surface plasmon resonance imaging, in cellular assays and in vivo.
Reported Results
Measure | Result |
|---|---|
Designed peptides with KD at the 10^-8 M level | 8% |
Peptides reaching KD as low as 10^-9 M | 26 |
In vivo testing | Top peptides tested in breast, myeloma and lung tumour mouse models |
The authors report membrane binding for the top peptides and describe potential for diagnostic imaging and targeted therapy. These are framed as potential rather than demonstrated clinical use.
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 PepMimic on Tamarind Bio
Antibody-to-peptide conversion: turn an existing antibody against a target into a short peptide binder.
Receptor mimicry: convert a known receptor-ligand interface into a compact peptide.
Diagnostic imaging candidates: generate peptides for targets such as PD-L1, HER2 or CD38, which the authors discuss for imaging.
Targets without known binders: start from a protein binder made with other tools, then derive a peptide from it.
How to Use PepMimic on Tamarind Bio
Log in and open the tool: sign in at tamarind.bio and open PepMimic.
Provide the target: upload the target protein structure, if exposed.
Provide the reference binder: supply the known receptor or antibody complex whose interface you want to mimic, as the method requires.
Set parameters: choose the number of peptides to generate and any other options the interface exposes.
Run the job: submit it and wait for the designs to complete.
Download and review outputs: retrieve the designed peptide structures and sequences.
Filter and validate: apply structure-prediction filtering, as in the paper, then synthesize top candidates for binding assays.
Things to Keep in Mind
The method needs a known binder to mimic; for targets without one, a protein binder must be produced first with other tools.
Experimental success was reported for a limited set of targets: PD-L1, CD38, BCMA, HER2 and CD4.
Only a minority of designs, 8%, reached the 10^-8 M affinity level, so plan to screen several candidates.
Clinical use is not demonstrated; the in vivo results are preliminary.
Source: Xiangzhe Kong et al., Peptide design through binding interface mimicry with PepMimic, Nature Biomedical Engineering, 2026, DOI 10.1038/s41551-025-01507-4. Code: https://github.com/kxz18/PepMimic.