Generating information regarding chemical compound based on latent representation
AssigneePreferred Networks, Inc.
InventorsKenta Oono, Justin Clayton, Nobuyuki Ota
Life of this patent
Estimating the term…
Abstract
A computer system for generating information regarding chemical compound includes one or more memories and one or more processors configured to generate information regarding chemical compound based on a latent variable, and to evaluate the generated information regarding chemical compound based on desired characteristics, wherein generating the information regarding chemical compound is restricted by the desired characteristics.
Claims36
A method of generating information regarding a chemical compound, comprising:
- causing, by at least one processor, a neural network to generate the information regarding the chemical compound by inputting a latent representation and label information into the neural network,
- wherein the information regarding the chemical compound includes information that expresses, and uniquely identifies, a chemical structure of the chemical compound, and
- wherein the neural network is a decoder, and the decoder has been trained based on differences between training information that expresses, and uniquely identifies, chemical structures of chemical compounds and reconstructions of the training information, the reconstructions being generated by the decoder by inputting latent representations into the decoder, the latent representations being generated by inputting the training information into an encoder.
An apparatus comprising:
- one or more memories;
- and
- one or more processors configured to:
- cause a neural network to generate information regarding a chemical compound by inputting a latent representation and label information into the neural network,
- wherein the information regarding the chemical compound includes information that expresses, and uniquely identifies, a chemical structure of the chemical compound, and
- wherein the neural network is a decoder, and the decoder has been trained based on differences between training information that expresses, and uniquely identifies, chemical structures of chemical compounds and reconstructions of the training information, the reconstructions being generated by the decoder by inputting latent representations into the decoder, the latent representations being generated by inputting the training information into an encoder.
A non-transitory recording medium having at least one program embodied therein for causing at least one computer to cause a neural network to generate information regarding a chemical compound by inputting a latent representation and label information into the neural network,
- wherein the information regarding the chemical compound includes information that expresses, and uniquely identifies, a chemical structure of the chemical compound, and
- wherein the neural network is a decoder, and the decoder has been trained based on differences between training information that expresses, and uniquely identifies, chemical structures of chemical compounds and reconstructions of the training information, the reconstructions being generated by the decoder by inputting latent representations into the decoder, the latent representations being generated by inputting the training information into an encoder.
A method of generating information regarding a chemical compound, comprising:
- causing, by at least one processor, a neural network to generate the information regarding the chemical compound by inputting a latent representation into the neural network;
- generating another latent representation based on the information regarding the chemical compound;
- and
- causing, by the at least one processor, the neural network to generate second information regarding another chemical compound by inputting the another latent representation into the neural network,
- wherein the information regarding the chemical compound includes information specifying a chemical structure of the chemical compound, and
- wherein the second information regarding the another chemical compound includes information specifying a chemical structure of the another chemical compound.
An apparatus comprising:
- one or more memories;
- and
- one or more processors configured to perform:
- causing a neural network to generate information regarding a chemical compound by inputting a latent representation into the neural network;
- generating another latent representation based on the information regarding the chemical compound;
- and
- causing the neural network to generate second information regarding another chemical compound by inputting the another latent representation into the neural network,
- wherein the information regarding the chemical compound includes information specifying a chemical structure of the chemical compound, and
- wherein the second information regarding the another chemical compound includes information specifying a chemical structure of the another chemical compound.
A non-transitory recording medium having at least one program embodied therein for causing at least one computer to perform:
- causing a neural network to generate information regarding a chemical compound by inputting a latent representation into the neural network;
- generating another latent representation based on the information regarding the chemical compound;
- and
- causing the neural network to generate second information regarding another chemical compound by inputting the another latent representation into the neural network,
- wherein the information regarding the chemical compound includes information specifying a chemical structure of the chemical compound, and
- wherein the second information regarding the another chemical compound includes information specifying a chemical structure of the another chemical compound.
Dependencies are inferred from the claim text.