Predicting Indium Phosphide Quantum Dot Properties from Synthetic Procedures Using Machine Learning

Prediction of chemical reaction outcomes using machine learning (ML) has emerged as a powerful tool for advancing materials synthesis. However, this approach requires large and diverse datasets, which are extremely limited in the field nanomaterials synthesis due to inconsistent nonstandardized reporting literature lack understanding synthetic mechanisms. In study, we extracted parameters InP quantum dot (QD) syntheses our inputs resultant properties (absorption, emission, diameter) outputs from 72 publications. We “filled in” missing data imputation method prepare complete dataset containing 216 entries training testing predictive ML models. defined descriptor space two ways (condensed extended) based on either identity or role reagents explore best categorizing input features. achieved mean absolute errors (MAEs) low 20.29, 11.46, 0.33 nm absorption, diameter, respectively, with model. used these models deploy an accessible interactive web app designing (https://share.streamlit.io/cossairt-lab/indium-phosphide/Hot_injection/hot_injection_prediction.py). Using app, investigated trends syntheses, such effects common additives, like zinc salts trioctylphosphine. also designed conducted new experiments extensions procedures compared experimentally measured predictions, thus evaluating “real-life” accuracy Conversely, inverse design obtain QDs specific properties. Finally, applied same train, test, launch CdSe by expanding previously published dataset. Altogether, preprocessing implementations demonstrate ability targeted underlying mechanisms even when faced resources.