Method, device, and computer program for predicting interaction between compound and protein

ChangeBridge: Patent Grants - AI & Computing (G06N)

Published April 7th, 2026

Detected April 8th, 2026

Summary

USPTO granted Patent US12597489B2 to ONCOCROSS CO., LTD. for a deep learning method predicting compound-protein interactions. The patent covers training data acquisition, deep-learning model construction, and interaction prediction by excluding amino acid sequences associated with protein domains that negatively influence interactions. The patent contains 12 claims and has a filing date of December 6, 2021.

View original document View source feed page

What changed

USPTO issued Patent US12597489B2 to ONCOCROSS CO., LTD. for a method, device, and computer program using deep learning to predict compound-protein interactions. The invention acquires training data including compound data, protein data, and interaction scores, constructs a deep-learning model, and predicts interactions while excluding amino acid sequences from protein domains that negatively influence binding. This is a standard patent grant establishing exclusive rights.

For competitors in biotechnology and pharmaceutical research, this patent creates potential licensing requirements or design-around obligations for any commercial deep learning systems predicting compound-protein interactions using similar exclusion methods. The granted patent provides ONCOCROSS with 20-year exclusivity from the 2021 filing date for this specific computational approach to drug interaction prediction.

What to do next

Monitor for updates

Source document (simplified)

← USPTO Patent Grants

Method, device, and computer program for predicting interaction between compound and protein

Grant US12597489B2 Kind: B2 Apr 07, 2026

Assignee

ONCOCROSS CO., LTD.

Inventors

Jin Woo Choi, Yi Rang Kim

Abstract

A method, a device, and a computer program for predicting the interaction between a compound and a protein are provided. A method for predicting the interaction between a compound and a protein, according to some embodiments of the present disclosure, may include: acquiring compound data for training, protein data for training, and training data including interaction scores; constructing a deep-learning model by using the acquired training data; and predicting the interaction between the given compound and protein by using the constructed deep-learning model. The interaction between the given compound and protein in an in vivo environment can be accurately predicted by training the deep-learning model, while excluding, from an amino acid sequence of the protein for training, amino acid sequences associated with a protein domain having a negative influence on the interaction.