LadderGen: A Large-Scale Generative Library of Ladder Polymers for Membrane Separations

Abstract

Ladder polymers possess exceptional rigidity, microporosity, and thermal stability, yet their exploration as gas separation membranes is hindered by the scarcity of synthetically accessible chemical structures. Here, we introduce LadderGen, a large-scale hypothetical ladder polymer library constructed using template polymerization reactions and advanced generative machine learning (ML) models, expanding the chemical space of ladder polymers to 0.8 million chemical structures. Existing ML models were used to predict key membrane properties, such as glass transition temperature, fractional free volume, and O₂/N₂ ideal selectivity and permeability, enabling rapid screening of the generated library to identify promising candidate ladder polymers that may approach or exceed the 2008 Robeson upper bound. Finally, high-fidelity molecular dynamics simulations provide validation of the predicted gas permeabilities of the top candidates. This work provides a framework for accelerated de novo ladder polymer discovery to identify promising next-generation gas-separation membranes for sustainable energy and environmental applications.