← Back to papers

RareFold: Structure prediction and design of proteins with noncanonical amino acids

★ ★ ★ ★ ☆

Overview

Paper Summary › Explain Like I'm Five › Conflicts of Interest › Identified Limitations › Rating Explanation › Good to know › Topic Hierarchy › File Information ›

Paper Summary

Paperzilla title
Building with Weird Amino Acids: A New Tool for Protein Design

This paper introduces RareFold, a deep learning model that predicts the 3D structures of proteins containing both common and unusual amino acids. Researchers used RareFold to create a new tool called EvoBindRare, which successfully designed linear and cyclic peptides that bind to a target protein, demonstrating the potential for developing next-generation peptide therapeutics.

Explain Like I'm Five

Scientists made a computer program that can predict the shape of proteins made with special building blocks. They used it to design tiny protein-like molecules that stick to other proteins, which could be useful for making new medicines.

Possible Conflicts of Interest

P.B. is a cofounder of and shareholder in Cyclic Therapeutics.

Identified Limitations

Limited Experimental Validation of RareFold
While the authors provide experimental validation for EvoBindRare, the direct validation of RareFold's structure prediction accuracy is limited to a small test set of 174 structures, which might not be fully representative of the diversity of proteins with NCAAs. More extensive benchmarking on larger and diverse datasets is needed to fully assess RareFold's performance, especially for rare or novel NCAAs.
Limited Scope of NCAAs
The study focuses on 29 NCAAs, which is a small subset of the over 500 known NCAA types. The model's ability to generalize to other, less common, NCAAs remains unclear and requires further investigation.
Computational Cost of EvoBindRare
Although RareFold itself is relatively efficient, the design process using EvoBindRare still requires significant computational resources, especially for longer peptides and larger design spaces. This computational cost could limit its applicability for high-throughput design or complex design tasks.
Limited Generalizability of EvoBindRare
The EvoBindRare framework is currently only demonstrated on a single target protein (Ribonuclease). Further validation on different targets and with different binding modes is needed to assess its general applicability for peptide binder design.

Rating Explanation

The development of RareFold and EvoBindRare represents a significant advance in protein structure prediction and design with NCAAs. The combination of a novel token-based architecture with experimental validation makes this a strong contribution to the field. However, the limited scope of NCAAs included in the study and the need for further benchmarking and validation on larger and more diverse datasets prevent a rating of 5. The acknowledged conflict of interest is noted but does not appear to directly impact the scientific validity of the work.

Good to know

This is the Starter analysis. Paperzilla Pro fact-checks every citation, researches author backgrounds and funding sources, and uses advanced AI reasoning for more thorough insights.

Explore Pro →

File Information

Original Title: RareFold: Structure prediction and design of proteins with noncanonical amino acids
Uploaded: September 08, 2025 at 08:01 PM
Privacy: Public