Hierarchical Virtual Screening for Plasticizer Discovery with Active Learning

Polymer solubility is a topic widely studied in polymer science; far less is understood about the factors governing the solubility of small molecules within polymer matrices. The dispersion of such molecules, known as plasticizers, in the polymer is critical for the processability of the material and an unavoidable step in the formulation of polymer composites, and yet their selection remains largely empirical, and robust design rules are almost nonexistent. Here, we present a multiscale, data-driven framework for the virtual screening of hydrocarbon-based plasticizers for rubber and polyolefins. Starting from a 730 molecule design space, our active learning strategy narrows the required sampling by an order of magnitude, to only 76 polymer/plasticizer systems, and identifies side-chain rigidity as a primary factor driving phase separation. Importantly, the framework captures packing and entropy driven contributions to miscibility in polymer-rich melts that are not represented by traditional solubility parameter approaches. Finally, by calculating the packing lengths and Voronoi volume distributions, we show that phase behavior in a key area of our design space is primarily governed by the plasticizer’s packing efficiency, providing a basis for straightforward design rules for tuning miscibility. The surrogate model reproduces available experimental data and yields physical, interpretable structure-properties relationships, identifying the key molecular features of the plasticizers that promote miscibility in polymer-rich environments. These insights provide a mechanistic understanding and actionable design rules for tuning polymer/plasticizer miscibility within nonpolar mixtures, and a predictive route for improving polymer/additive compatibility in soft-material formulations.