Predictive Minisci and P450 Late Stage Functionalization with Transfer Learning

Abstract

Structural diversification of lead molecules is a key component of drug discovery to explore close-in chemical space. Late stage functionalizations (LSFs) are versatile methodologies capable of installing functional handles on richly decorated intermediates to deliver numerous diverse products in a single reaction. Predicting the regioselectivity of LSF is still an open challenge in the field. Numerous efforts from chemoinformatics and machine learning (ML) groups have made significant strides in this area. However, it is arduous to isolate and characterize the multitude of LSF products generated, limiting available data and hindering pure ML approaches. We report the development of an approach that combines message-passing neural network and an 13C NMR-based transfer learning to predict the atom-wise probabilities of functionalization. We validated our model retrospectively and with a series of prospective experiments, showing that it accurately predicts the outcomes of Minisci-type and P450 transformations, outperforming state-of-the-art Fukui-based reactivity indices.

Transfer learning for a foundational chemistry model

“…8,12,13 The specic graph neural network utilized was a message passing neural network (MPNN), a avor of graph convolutional neural network which have noted success in a variety of chemistry prediction tasks. 8,[13][14][15][16][17] Briey, an MPNN deduces the local chemical environment for each atom within the molecule, preserving the symmetry of chemically identical atoms. The training set for this initial task was comprised of carbon-containing crystal structures in the CCDC.…”

“…Prior research have noted the importance of this "negative data" in predictive modelling. 15,36 Crystal-Yield's average MAE of ∼18% is a step forward towards accurate modeling of noisier data.…”

Transfer Learning for a Foundational Chemistry Model

“…The specific graph neural network utilized was a message passing neural network (MPNN), a flavor of graph convolutional neural network which have noted success in a variety of chemistry prediction tasks. 8,[13][14][15][16][17] Briefly, an MPNN deduces the local chemical environment for each atom within the molecule, preserving the symmetry of chemically identical atoms. The training set for this initial task was comprised of carbon-containing crystal structures in the CCDC.…”

Graph transformer neural network for chemical reactivity prediction

“…Analysis of the screening results revealed that the drugs Loratadine (7), Nevirapine (8), and 11 fragments (26,28,29,(33)(34)(35)(37)(38)(39)(40)(41) were alkylated with different types of alkyl fragments. From this subset, conditions showing reasonable conversion (> 40%, based on UV trace) were subjected to upscaling.…”

“…[10] A similar approach was introduced for the prediction of late-stage alkylation, mainly focusing on the Baran-type diversinate chemistry using alkyl sodium sulfinate salts. [41] A recent study has shown that hybrid machine learning models augmented with quantum chemical information of transition states enable accurate regioselectivity prediction for iridium-catalyzed borylation reactions in very low data regimes. [42] Herein, we introduce the combination of GNN-based in silico reaction screening with a miniaturized HTE setup for identification of LSF alkylations by Minisci reaction, addressing various aspects relevant to drug discovery (Figure 1).…”