US 12,079,730 B2
Transfer learning for molecular structure generation
Enara C Vijil, Millwood, NY (US); Payel Das, Yorktown Heights, NY (US); and Inkit Padhi, White Plains, NY (US)
Assigned to INTERNATIONAL BUSINESS MACHINES CORPORATION, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on May 28, 2020, as Appl. No. 16/886,160.
Prior Publication US 2021/0374551 A1, Dec. 2, 2021
Int. Cl. G06N 3/088 (2023.01); G06N 3/045 (2023.01); G06N 3/047 (2023.01); G16C 20/70 (2019.01)
CPC G06N 3/088 (2013.01) [G06N 3/045 (2023.01); G06N 3/047 (2023.01); G16C 20/70 (2019.02)] 20 Claims
OG exemplary drawing
 
1. A system, comprising:
a memory that stores computer executable components; and
a processor, operably coupled to the memory, and that executes the computer executable components stored in the memory, wherein the computer executable components comprise:
a transfer learning component that trains a machine learning model to design molecules that satisfy defined criteria with respect to binding to proteins while having drug likeliness and synthetic accessibility, wherein the training comprises:
training, using unlabeled training data associated with target attributes of the molecules, an autoencoder of the machine learning model, to predict the target attributes of the molecules, and
jointly training, using first labeled training data for a first attribute of the target attributes having an amount of labeled training data exceeding a defined threshold, the autoencoder and at least one of a regressor or a first classifier of the machine learning model to predict the first attribute, wherein the joint training comprises learning, by the autoencoder, a latent space that is predictive of whether a molecular structure exhibits the first attribute.