Abstract
Developing bio-based formulations is essential for reducing the environmental impact of modern surfactants. Peptide amphiphile (PA) surfactants offer a modular platform for developing materials with tailored structures and functions, making them highly attractive to replace existing petrochemical surfactants. However, rationally designing peptide sequences with predictable and tunable surface activity remains challenging. Existing sequence-based models primarily target aggregation and antimicrobial activity and cannot predict surface activity or account for environmental conditions such as pH and temperature. To address this shortcoming, we combine experimental and computational approaches to investigate sequence–property relationships of PAs. We employ automated experimental workflows to generate a consistent dataset of PA surface tension across a range of pHs. Our analyses reveal that the presence of proline most strongly induces surface activity at all pHs. Positively charged residues such as lysine and arginine more strongly introduce surface activity than negatively charged residues. Experimental validation of a machine learning classifier trained on the data collected for our library showed that the machine learning model correctly predicts surface activity in 97% of the sampled predicted surface active sequences. Thereby, we show that machine learning models trained on our dataset generalize across the sequence space, and enable targeted design of surface active PAs in diverse environments.