SYNEVO: towards synthetic evolution of biomolecules via aligning protein language models to biological hardware
Track: Biology: datasets and/or experimental results
Nature Biotechnology: Yes
Keywords: Wet-dry lab integration, Zinc Fingers, cis-coupling, protein language models, Reinforcement Learning, DPO
TL;DR: We propose SYNEVO: an AI-driven, closed-loop system integrating automated protein design, and real-time experimental feedback to iteratively optimize biomolecular function.
Abstract: Applications of biomolecular systems span gene writing, drug discovery, and environmental remediation. Despite their potential, biodesign remains slow and labor-intensive, often relying on trial and error. Recent advances in high-throughput sequencing, automated synthesis, and generative AI offer new opportunities but remain fragmented. We propose SYNEVO: an AI-driven, closed-loop system integrating automated protein design, and real-time experimental feedback to iteratively optimize biomolecular function. SYNEVO does not use template-based DNA replication, enabling a constraint-free generation of new genotypes, which departs from conventional evolution. We aim to validate our platform studying Zinc Finger proteins, a versatile class of DNA-binding proteins with significant therapeutic potential. Preliminary results showed that, by iteratively generating large libraries with autoregressive protein language models and experimentally testing their phenotypes, we optimized sequence selection. The measured features were fed back into the model via Reinforcement Learning to maximize protein enrichment scores, achieving a progressive improvement of generated phenotypes. This method, compared with Directed Evolution, shows higher efficiency in sampling high fitness protein candidates, and broader exploration of the sequence space. Potentially, by continuously refining its designs with minimal human intervention, this approach will accelerate protein engineering and provide a scalable solution for engineering new biomolecules with broad use across biotechnology and synthetic biology.
Anonymization: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and identifying URLs.
Presenter: ~Maria_Artigues-Lleixa1
Format: Yes, the presenting author will attend in person if this work is accepted to the workshop.
Funding: Yes, the presenting author of this submission falls under ICLR’s funding aims, and funding would significantly impact their ability to attend the workshop in person.
Submission Number: 113
Loading