PRISM: Parallelized Reaction-rates via Indicator Spectrometry using Machine-vision

This repository contains the complete workflow for predicting the type and amide coupling PRISM reaction rates using machine learning and graph neural networks.

Contents

• build_class_and_bias_models/ - the ML models (classifiers and regressors) with hyperparameter optimization via Optuna to classify the type of PRISM rate and predict the bias and correct the PRISM rates. See README
• build_gnn_model/ - the AIM Graph neural network model for predicting the PRISM reaction rate value. See README
• data/ - Datasets including molecular and atomistic descriptors, reaction rates, and XYZ molecular structures.
• generate_features/ - Scripts for generating molecular/atomistic features from structures using the Morfeus python package and pKa calculators.
• image_analysis/ - Image processing scripts for analyzing the PRISM high-throughput experimental plate data. See README
• predictions_from_class_bias.ipynb - Jupyter notebook for making PRISM classification predictions on new reaction combinations. Open notebook
• predictions_from_gnn.ipynb - Jupyter notebook for making PRISM HTE rate predictions on new reaction combinations. Open notebook

Citation

Research paper coming soon! To cite:

[Citation will be added upon publication]