Institute for Protein Design / University of Washington
Diffusion-based backbone generation and sequence design method for programmable asymmetric transmembrane beta-barrel nanopores.
This work introduces a generative deep-learning approach for the de novo design of transmembrane beta-barrel (TMB) nanopores, developed by David Baker's group at the University of Washington Institute for Protein Design and reported as a bioRxiv preprint in June 2026. Transmembrane beta-barrels are the structural class that natural biological nanopores belong to, and they underpin technologies ranging from nanopore DNA sequencing to single-molecule sensing. Yet natural TMBs offer only a fixed repertoire of geometries, and engineering them from natural scaffolds is constrained by the limited diversity nature provides. The method addresses this gap by generating entirely new barrel architectures with programmable shape and pore dimensions.
The approach pairs diffusion-based protein backbone generation, conditioned on beta-barrel structural features, with a TMB-optimized sequence design stage that accounts for the distinct chemical environment of the lipid bilayer. This lets designers specify barrel parameters — such as the number of beta-strands and the diameter of the central pore — and obtain backbones tailored to those targets, rather than adapting whatever natural barrels happen to exist. The result is a design pipeline architecturally related to the RFdiffusion family of structure-conditioned generators but specialized for the geometry and membrane-embedded chemistry of asymmetric beta-barrel pores.
By extending the generative protein design paradigm into the membrane, the work opens the transmembrane beta-barrel fold to the same kind of programmable, target-driven engineering that diffusion models brought to soluble proteins.
The pipeline combines a diffusion model for protein backbone generation, conditioned on beta-barrel structural features, with a transmembrane-beta-barrel-specific sequence design step. It is architecturally related to the RFdiffusion family of structure-conditioned diffusion models from the same institute — distinct from RFdiffusion, RFdiffusion2, and RFdiffusion3 — but specialized for the geometry and membrane chemistry of beta-barrel pores. Generated designs cover barrels of 10 to 16 strands with pore diameters of approximately 0.7 to 1.5 nm. In experimental characterization, 48 designs were tested; high-resolution crystal structures were solved for two of them, confirming that the designed backbones fold as intended, and functional assays demonstrated both ion sensing and DNA translocation through designed pores.
Programmable transmembrane beta-barrels are the structural foundation of nanopore-based technologies, so a method that designs them with custom geometry has direct relevance to nanopore DNA and RNA sequencing, single-molecule biosensing, and molecular transport. Designers can specify pore diameter and strand number to tune a barrel for a particular analyte — for example, sizing a lumen for ion sensing or for threading single-stranded DNA. Beyond sequencing and sensing, custom beta-barrels could serve as building blocks for synthetic membrane channels, controlled-release systems, and engineered cellular interfaces, benefiting groups in synthetic biology, biophysics, and analytical chemistry who previously had to start from the limited palette of natural pore proteins.
This work extends de novo generative protein design from soluble proteins into the membrane, applying diffusion-based backbone generation to one of the most technologically important folds — the transmembrane beta-barrel that natural nanopores are built from. By making barrel geometry and pore size programmable and validating designs experimentally, including two crystal structures and demonstrated ion-sensing and DNA-translocating function, it establishes that custom nanopores can be generated to specification rather than engineered from natural scaffolds. As a 2026 bioRxiv preprint, its long-term influence is still unfolding, and no code or model weights have been released yet, which currently limits direct reuse. Even so, it points toward a future in which nanopore sequencing devices and single-molecule sensors are equipped with purpose-built pores, and it broadens the reach of structure-conditioned diffusion models into membrane protein design.
Philomin, A., et al. (2026) Generative design of programmable asymmetric β-barrel nanopores. bioRxiv.
DOI: 10.64898/2026.06.04.729630Papers that recently cited this model.
The most-cited papers that cite this model.
Not enough data