Apr 10, 2025
9:00am - 9:15am
Summit, Level 4, Room 421
Yifeng Cai1,Mengqi Sun1,Maxim Ziatdinov2,Sergei Kalinin3,2,David Ginger1,Francois Baneyx1
University of Washington1,Pacific Northwest National Laboratory2,The University of Tennessee, Knoxville3
Yifeng Cai1,Mengqi Sun1,Maxim Ziatdinov2,Sergei Kalinin3,2,David Ginger1,Francois Baneyx1
University of Washington1,Pacific Northwest National Laboratory2,The University of Tennessee, Knoxville3
Elastin-like peptides (ELPs) are low complexity proteins that phase separate (coacervate) into micrometer-size liquid droplets above lower critical solution temperature (LCST). While ELPs have been extensively investigated in biotechnology and medicine, little attention has been paid to their ability to guide hierarchical assembly process by which temperature elevation causes ELP-functionalized nanoparticles to aggregate. Previously, we reported on the integration of dual variational autoencoder (dual-VAE) machine learning model for the rapid and accurate extraction of plasmonic particle geometries from optical scattering response. By incubating ELP-functionalized gold nanoparticles (ELP-AuNPs) below the transition temperature of the ELP, we show that it is possible to stabilize particle clusters of well-defined hydrodynamic diameters. We train dual-VAE with correlative optical scattering spectra and electron microscopy images of these ELP-AuNP assemblies to predict cluster size and average number of particles in each cluster. Interestingly, we further explore the generalizability of this approach to predict the temperature dependence of assembly patterns of AuNPs functionalized with shorter ELPs (higher LCST). This process of building structure-property relationships via dual VAE should be applicable to broader material science and physics problems.