Nucleophilicity Prediction Using Graph Neural Networks

Abstract

The quantitative description between chemical reaction rates and nucleophilicity parameters plays a crucial role in organic chemistry. In this regard, the formula proposed by Mayr et al. and the constructed reactivity database are important representatives. However, the determination of Mayr’s nucleophilicity parameter N often requires time-consuming experiments with reference electrophiles in the solvent. Several machine learning (ML)-based models have been proposed to realize the data-driven prediction of N in recent years. However, in addition to DFT-calculated electronic descriptors, most of them also use a set of artificially predefined structural descriptors as input, which may result in a biased representation of the nucleophile’s structural information depending on descriptors’ definition preference. Compared with traditional ML algorithms, graph neural networks (GNNs) can naturally take the molecule’s structural information into account by applying the message passing technique. We herein proposed a SchNet-based GNN model that only takes the molecular conformation and solvent type as input. The model achieves a comparable performance to the previous benchmark study on 10-fold cross-validation of 894 data points (R 2 = 0.91, RMSE = 2.25). To enhance the model’s ability to capture the molecule’s electronic information, some DFT-calculated parameters are then incorporated into the model via graph global features, and substantial improvement is achieved in the prediction precision (R 2 = 0.95, RMSE = 1.63). These results demonstrate that both structural and electronic information are important for the prediction of N, and GNN can integrate these two kinds of information more effectively.

The computational road to reactivity scales

“…In recent years, the data set size has been expanded to a number of reference structures greater than 750 to learn correlations between different descriptors and reactivity parameters, which enable highly accurate predictions of reactivity. Saini et al 59 report their best result using a NN for 752 structures, while Tavakoli et al 60 neural was trained by Nie et al 61 on nearly 900 nucleophiles from Mayr's database, in which electronic and, additionally, solvent descriptors were employed.…”

“…Saini et al 59 report their best result using a NN for 752 structures, while Tavakoli et al 60 use methyl anion/cation affinities in solution (MAA* and MCA*, see below) as inputs for over 2421 structures to train a graph attention network. A graph neural network was trained by Nie et al 61 on nearly 900 nucleophiles from Mayr's database, in which electronic and, additionally, solvent descriptors were employed.…”

Boosting Graph Neural Networks with Molecular Mechanics: A Case Study of Sigma Profile Prediction

“…Thus, because molecules are easily translated to graphs, the use of GNNs in chemistry-related fields is gaining popularity. 15,16 Recent examples include the prediction of chemical shifts in Nuclear Magnetic Resonance, 17 the prediction of nucleophilicity parameters, 18 and, central to this work, the prediction of sigma profiles reported by Zhang et al 13 Zhang et al 13 developed and trained a message passing GNN to predict sigma profiles. However, due to the initial poor performance of the method, their graph embedding latent space had to be supplemented with 200 additional moleculelevel features, leading to a composite feedforward neural network rather than a neat GNN.…”

Prediction of Nucleophilicity and Electrophilicity Based on a Machine‐Learning Approach

“…Several ML models have been reported for the prediction of N or E [3] . Very recently, using Nguyen's dataset, Fu, Yu, and Li [4] established a graph neural network (GNN) model for nucleophilicity prediction, achieving R 2 =0.95 and RMSE=1.63 for N prediction when adding quantum chemical descriptors to the graph. Despite these advances, all these models rely on DFT‐calculated descriptors that are carefully selected based on expertise.…”