Introduction

IMPRESSION is a series of machine learning models which predict NMR parameters.

The first generation of IMPRESSION models predict Carbon and Proton chemical shifts, as well as 1-bond Carbon-Proton scalar coupling constants.

The second generation models (currently being tested) will predict all 1-3 bond Carbon,Proton, and Nitrogen (all combinations) scalar coupling constants as well as H, C, and N chemical shifts.

First Generation

The first generation of IMPRESSION models are based on Kernel Ride Regression, and use molecules obtained from the Cambridge structural database. They are discussed fully in the recent publication in chemical science

Second Generation (coming summer 2020)

The second generation of IMPRESSION models are based on graph transformer networks, and are trained using the molecules from generation one, as well as molecules from the QM9 dataset.