Interpretable Machine Learning Models for Phase Prediction in Polymerization-Induced Self-Assembly

doi:10.1021/acs.jcim.3c00460

Interpretable Machine Learning Models for Phase Prediction in Polymerization-Induced Self-Assembly

2023-05-19

5

OA

AI

DOI：10.1021/acs.jcim.3c00460

Original

Save

View PDF

Abstract

En 中文

While polymerization-induced self-assembly (PISA) hasbecome apreferred synthetic route toward amphiphilic block copolymer self-assemblies,predicting their phase behavior from experimental design is extremelychallenging, requiring time and work-intensive creation of empiricalphase diagrams whenever self-assemblies of novel monomer pairs aresought for specific applications. To alleviate this burden, we develophere the first framework for a data-driven methodology for the probabilisticmodeling of PISA morphologies based on a selection and suitable adaptionof statistical machine learning methods. As the complexity of PISAprecludes generating large volumes of training data with insilico simulations, we focus on interpretable low variancemethods that can be interrogated for conformity with chemical intuitionand that promise to work well with only 592 training data points whichwe curated from the PISA literature. We found that among the evaluatedlinear models, generalized additive models, and rule and tree ensembles,all but the linear models show a decent interpolation performancewith around 0.2 estimated error rate and 1 bit expected cross entropyloss (surprisal) when predicting the mixture of morphologies formedfrom monomer pairs already encountered in the training data. Whenconsidering extrapolation to new monomer combinations, the model performanceis weaker but the best model (random forest) still achieves highlynontrivial prediction performance (0.27 error rate, 1.6 bit surprisal),which renders it a good candidate to support the creation of empiricalphase diagrams for new monomers and conditions. Indeed, we find inthree case studies that, when used to actively learn phase diagrams,the model is able to select a smart set of experiments that lead tosatisfactory phase diagrams after observing only relatively few datapoints (5-16) for the targeted conditions. The data set aswell as all model training and evaluation codes are publicly availablethrough the GitHub repository of the last author.

Keywords:

AQUEOUS DISPERSION POLYMERIZATION

COPOLYMER NANO-OBJECTS

POLYELECTROLYTE-STABILIZED NANOPARTICLES

CROSS-LINKING

MOLECULAR-WEIGHT

WORM GELS

BLOCK

RAFT

PISA

VESICLES

Interpretable Machine Learning Models for Phase Prediction in Polymerization-Induced Self-Assembly

Abstract

AI Summary

Journal

Organization