The selection of experimental conditions leading to a reasonable yield is an important and essential element for the automated development of a synthesis plan and the subsequent synthesis of the target compound. The classical QSPR approach, requiring one-to-one correspondence between chemical structure and a target property, can be used for optimal reaction conditions prediction only on a limited scale when only one condition component (e.g., catalyst or solvent) is considered. However, a particular reaction can proceed under several different conditions. In this paper, we describe the Likelihood Ranking Model representing an artificial neural network that outputs a list of different conditions ranked according to their suitability to a given chemical transformation. Benchmarking calculations demonstrated that our model outperformed some popular approaches to the theoretical assessment of reaction conditions, such as k Nearest Neighbors, and a recurrent artificial neural network performance prediction of condition components (reagents, solvents, catalysts, and temperature). The ability of the Likelihood Ranking model trained on a hydrogenation reactions dataset, (~42,000 reactions) from Reaxys^® database, to propose conditions that led to the desired product was validated experimentally on a set of three reactions with rich selectivity issues.

Reaction conditions (RCs) are a crucial part of reaction definition, and their accurate prediction is an important component of chemical synthesis planning. The existence of multiple combinations of RCs capable of achieving the desired result complicates the task of condition recommendation. Herein, we propose a conditional variational autoencoder (CVAE) generative model to predict suitable RCs. The CVAE model has been customized to generate diverse sets of valid conditions, ensuring high flexibility and accuracy, while circumventing the necessity for enumeration or combinatorial search of potential RCs. The efficacy of the CVAE approaches was evaluated using hydrogenation reactions and other H₂-mediated reactions, predicting the set of catalysts, additives (acid, base, and catalytic poison), ranges of temperature, and pressure. The CVAE models predicted conditions with different “heads”, each corresponding to specific condition components, and their respective losses. CVAE models were tested on two datasets: a small one containing 31K reactions with 2232 potential conditions’ combinations and a big one having 196K reactions with ~7 × 10⁴² potential conditions’ combinations to evaluate the model’s ability to predict varying complexity and diversity conditions. To optimize the accuracy of the models, we experimented with three latent distribution variants—Gaussian (g-CVAE), Riemannian Normalizing Flow (rnf-CVAE), and Hyperspherical Uniform (h-CVAE). In our experiments, the h-CVAE model demonstrated robust overall performance, making it the optimal choice for scenarios requiring high accuracy across multiple top-k predictions. Benchmarking analyses demonstrated the high performance of the CVAE models compared to state-of-the-art reaction condition prediction approaches.