Method and apparatus for optimizing synthetic conditions for generation of target products

A method of optimizing synthetic conditions, comprising:

• receiving a graph-based descriptor comprising structural information of at least one reactant and structural information of a target product to be synthesized by the at least one reactant, and a synthetic path of the at least one reactant, and receiving a reaction mechanism of the at least one reactant;
• determining a search space comprising combinations of synthetic conditions for generating the target product and the synthetic path that satisfy the reaction mechanism by applying the graph-based descriptor to a prediction neural network model;
• sampling at least one initial condition combination among the combinations based on a first confidence corresponding to a yield of the combinations in the search space;
• conducting a chemical synthesis experiment using the sampled at least one initial condition combination, and obtaining a ground-truth yield based on a result of the chemical synthesis experiment;
• updating the prediction neural network model based on a difference between the ground-truth yield and a predicted yield of the prediction neural network model;
• determining a priority of the combinations for a subsequent experiment based on the updated prediction neural network model, a model uncertainty, and a yield obtained via a yield inference method;
• and
• redetermining the search space to include subsequent combinations of synthetic conditions and a subsequent synthetic path based on the determined priority, and repeating the chemical synthesis experiment and the updating of the prediction neural network model until a target yield is achieved,
• wherein, as the prediction neural network model is updated based on a result of an experiment with the subsequent combinations, the priority is redetermined, and the search space is determined to include next subsequent combinations of synthetic conditions that follow the subsequent combinations based on the redetermined priority.

An apparatus for optimizing synthetic conditions, the apparatus comprising:

• a user interface (UI) configured to receive a graph-based descriptor comprising at least one of structural information of at least one reactant and structural information of a target product to be synthesized by the at least one reactant;
• a memory storing one or more instructions;
• and
• a processor configured to execute the one or more instructions to:
• determine combinations of synthetic conditions for generating the target product by applying the graph-based descriptor to a prediction neural network model, select at least one initial condition combination from among the combinations based on a first confidence corresponding to a yield of the combinations, conduct a chemical synthesis experiment using the selected initial condition combination, obtain a ground-truth yield based on the result of the chemical synthesis experiment, update the prediction neural network model based on based on a difference between the ground-truth yield and a predicted yield of the prediction neural network model, determine a priority of the combinations for a subsequent experiment based on the updated prediction neural network model, a model uncertainty, and a yield obtained via a yield inference method, determine subsequent combinations of synthetic conditions based on the determined priority, and repeat the chemical synthesis experiment and the updating of the prediction neural network model until a target yield is achieved,
• wherein, as the prediction neural network model is updated based on a result of an experiment with the subsequent combinations, the priority is redetermined, and next subsequent combinations of synthetic conditions are determined based on the redetermined priority.

An apparatus for optimizing synthetic conditions, the apparatus comprising:

• a user interface (UI) configured to receive a graph-based descriptor comprising structural information of at least one reactant, structural information of a target product to be synthesized by the at least one reactant, and a synthetic path comparing at least one of substituents and reactors of the at least one reactant, and receive a reaction mechanism of the at least one reactant;
• a memory storing one or more instructions;
• and
• a processor configured to execute the one or more instructions to:
• determine a search space comprising combinations of synthetic conditions for generating the target product and a synthetic path, which satisfy the reaction mechanism, by applying the graph-based descriptor to a prediction neural network model;
• sample at least one initial condition combination among the combinations based on a first confidence corresponding to a yield of the combinations in the search space;
• conduct a chemical synthesis experiment using the sampled at least one initial condition combination, and obtain a ground-truth yield based on a result of the chemical synthesis experiment,
• update the prediction neural network model based on a difference between the ground-truth yield and a predicted yield of the prediction neural network model;
• determine a priority of the combinations for a subsequent experiment based on the updated prediction neural network model, a model uncertainty, and a yield obtained via a yield inference method;
• and
• redetermine the search space to include subsequent combinations of synthetic conditions based on the determined priority, and repeat the chemical synthesis experiment and the updating of the prediction neural network model until a target yield is achieved,
• wherein, as the prediction neural network model is updated based on a ground-truth yield obtained from a result of an experiment with the subsequent combinations of the synthetic conditions, the priority is redetermined, and the search space is determined to include next subsequent combinations of synthetic conditions based on the redetermined priority.