Retrieval Augmented Zero-Shot Enzyme Generation for Specified Substrate
Keywords: Protein Design, Diffusion model
Abstract: The ability to generate novel enzymes that catalyze specific target molecules is a critical advancement in biomaterial synthesis and chemical production. However, a significant challenge arises when no recorded enzymes exist for the target molecule, making it a zero-shot generation problem. This absence of known enzymes complicates the training of generative models tailored to the target substrate. To address this, we propose a retrieval-augmented generation method that leverages existing enzyme-substrate data to overcome the lack of direct examples. Since there is no recorded catalytic performance between the enzymes and the new target molecule, the challenge shifts to identifying enzymes that helpful for generation. Our approach tackles this by retrieving enzymes whose substrates exhibit structural similarities to the target molecule, thereby exploiting functional similarities reflected in the enzymes' catalytic capability. This leads to the next challenge: how to utilize the retrieved enzymes to generate a novel enzyme capable of catalyzing the target molecule, given that none of the retrieved enzymes directly catalyze it. To solve this, we employ a conditioned discrete diffusion model that takes the aligned retrieved enzymes to generate a new enzyme. We train the generator with guidance from an enzyme-substrate relationship classifier to make it output the optimal protein sequence distribution for different target molecule. We evaluate our model on enzyme design tasks involving a diverse set of real-world substrates, and our results including catalytic rate predictions, foldability assessments, and docking position analyses, demonstrate that our model outperforms existing protein generation methods for substrate-specified enzyme generation. Additionally, we formally define the zero-shot substrate-specified enzyme generation task and contribute a comprehensive dataset with evaluation methods.
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 674
Loading