Transfer learning based on atomic feature extraction for the prediction of experimental ¹³C chemical shifts

Forecasting experimental chemical shifts of organic compounds is a long-standing challenge in organic chemistry. Recent advances in machine learning (ML) have led to routines that surpass the accuracy of ab initio Density Functional Theory (DFT) in estimating experimental $^{13}$C shifts. The extraction of knowledge from other models, known as transfer learning, has demonstrated remarkable improvements, particularly in scenarios with limited data availability. However, the extent to which transfer learning improves predictive accuracy in low-data regimes for experimental chemical shift predictions remains unexplored. This study indicates that atomic features derived from a message passing neural network (MPNN) forcefield are robust descriptors for atomic properties. A dense network utilizing these descriptors to predict $^{13}$C shifts achieves a mean absolute error (MAE) of 1.68 ppm. When these features are used as node labels in a simple graph neural network (GNN), the model attains a better MAE of 1.34 ppm. On the other hand, embeddings from a self-supervised pre-trained 3D aware transformer are not sufficiently descriptive for a feedforward model but show reasonable accuracy within the GNN framework, achieving an MAE of 1.51 ppm. Under low-data conditions, all transfer-learned models show a significant improvement in predictive accuracy compared to existing literature models, regardless of the sampling strategy used to select from the pool of unlabeled examples. We demonstrated that extracting atomic features from models trained on large and diverse datasets is an effective transfer learning strategy for predicting NMR chemical shifts, achieving results on par with existing literature models. This method provides several benefits, such as reduced training times, simpler models with fewer trainable parameters, and strong performance in low-data scenarios, without the need for costly ab initio data of the target property. This technique can be applied to other chemical tasks opening many new potential applications where the amount of data is a limiting factor.