Abstract
Abstract
Designing amphiphilic block copolymers from fossil fuel-derived hydrocarbons is a cornerstone of colloid chemistry, enabling the creation of advanced polymeric surfactants. Despite the high attractiveness of low carbon-emitting plant proteins to defossilise surfactant processing, attempts in identifying plant proteins that will function as appropriate surfactants are somewhat hit and miss. Here, we present a data-driven pipeline to fingerprint plant protein surfactants by harnessing the diblock-like signature of a classic surfactant, integrating protein sequence data with statistical thermodynamics-based calculations, finetuned by machine learning. We demonstrate that protein sequence-level features do not in themselves allow for the prediction of adsorbed configuration. Instead, segmenting adsorbed proteins into blocks enabled identification of hundreds of plant proteins possessing a diblock-like signature, validated experimentally for optimal surface properties. Thus, by embedding adsorption configuration information into the classification process, we streamline the discovery of plant protein surfactants at scale, unlikely to be realised by purely empirical screening.