MOTIFNet: Automating the Analysis of Amphiphile and Block Polymer Self-Assembly from SAXS Data

Daoyuan Li, Shuquan Cui, Mahesh Mahanthappa, Frank Bates, Timothy Lodge, Joern Ilja Siepmann

Submission Track: Short Paper

Submission Category: Automated Material Characterization

Keywords: Small-Angle X-ray Scattering (SAXS), Self-Assembly, Mixture of Experts (MoE), Temporal Convolutional Network (TCN), Self-Attention, Order-Disorder Transition (ODT)

TL;DR: MOTIFNet automates the analysis of SAXS data for amphiphile and block polymer self-assembly by using a simplified mixture of experts model with temporal convolution and self-attention for morphology classification and ODT prediction.

Abstract: Accurately classifying morphology and assessing stability in soft matter self-assembly often require specialized analysis of small-angle X-ray scattering (SAXS) data, creating an obstacle to automation. To address this, we introduce MOTIFNet, a simplified sparse mixture of experts (MoE) model with top-1 routing. By combining temporal convolution and self-attention, MOTIFNet effectively processes SAXS time series data, enabling morphology classification, SAXS pattern prediction, and the estimation of order-disorder transition (ODT) probabilities. This model advances automated characterization, accelerating experimentation and high-throughput studies in soft matter self-assembly.

Submission Number: 16