Romero LabDuke University

Romero Lab

We build AI systems to understand and engineer proteins.

We seek to uncover the principles that govern protein function and use them to design molecules for chemistry, biotechnology, and medicine.

Explore the work
01SEQUENCE → STRUCTURE → FUNCTION

Learning the language of proteins.

Proteins encode remarkable molecular functions in their amino acid sequences. We develop machine learning methods that learn the relationships between sequence, structure, chemistry, and function, with the goal of understanding how proteins work and how to design new ones.

LEARNING PROTEIN FUNCTIONFIG. 01 / SEQUENCE → STRUCTURE → FUNCTION

Representative work

Functional alignment of protein language models via reinforcement learningRLXF · BIORXIV · 2026Biophysics-based protein language models for protein engineeringMETL · NATURE METHODS · 2025
02AUTONOMOUS EXPERIMENTATION

AI that experiments.

What if AI could interact directly with biology? We build systems that can design and perform experiments, learn from the results, and decide what to do next. By coupling AI to the laboratory, we aim to create a new kind of scientific intelligence that learns through direct interaction with the physical world.

FIG. 02 / AUTONOMOUS EXPERIMENTATION

Representative work

Learning protein function through autonomous experimental interactionAUTONOMOUS AI · BIORXIV · 2026Self-driving laboratories to autonomously navigate the protein fitness landscapeSAMPLE · NATURE CHEMICAL ENGINEERING · 2024
03MOLECULAR ENGINEERING

Engineering new biological function.

Nature has evolved proteins with an extraordinary range of functions. We seek to expand this molecular repertoire, designing proteins with new activities and properties for challenges in chemistry, biotechnology, and medicine.

FORM → FUNCTIONFIG. 03 / MOLECULAR ENGINEERING

Biocatalysis

Engineering enzymes to perform useful chemical transformations.

Synthetic biology

Reprogramming biological systems to produce new molecules and functions.

Therapeutics

Designing proteins with new molecular activities for applications in human health.

Representative work

Machine learning-guided olivetolic acid cyclase engineering enables tailored cannabinoid biosynthesis in yeastBIOCATALYSIS · BIORXIV · 2026Multiobjective learning and design of bacteriophage specificitySYNTHETIC BIOLOGY · BIORXIV · 2026Directed evolution of angiotensin-converting enzyme 2 peptidase activity profiles for therapeutic applicationsTHERAPEUTICS · PROTEIN SCIENCE · 2023

Latest.

Duke Center for Biomolecular and Tissue Engineering
Lew Predoctoral Fellowship

Jeonghyeon Kim was named a Lew Predoctoral Fellow in Duke’s Center for Biomolecular and Tissue Engineering.

Jeonghyeon Kim, Nathaniel Brooks, and Ameya Joshi at ICML 2026 in Seoul
ICML, Seoul

Jeonghyeon, Nathaniel, and Ameya presented the lab’s PRISM work at ICML 2026 in Seoul.

International Conference on Learning Representations logo
ICLR, Rio de Janeiro

Ben Perry presented AlphaFast at ICLR 2026 in Rio de Janeiro.

Follow @romerolab1