Generative design of programmable asymmetric β-barrel nanopores

Abstract

Protein nanopores are powerful tools for molecular sensing, sequencing, and separation, but designing pores with programmable function remains challenging. Native homo-oligomeric transmembrane {beta} barrels (TMBs) are used for these applications, but their uniform lumens limit spatial resolution and analyte discrimination. Although monomeric TMBs can be designed using energy-based methods, these approaches remain highly manual and limited to structural design rather than function. Here, we present a generative AI framework for TMB design, with backbone and sequence design models trained on a curated distillation set. We characterized 48 designs spanning 0.7-1.5 nm in diameter, diverse lumen chemistries, and hydrophobic thicknesses. Crystal structures closely match the design models. We demonstrate that our method produces customizable nanopores for ion sensing, DNA translocation, and transport across polymer membranes.