KR-20260065059-A - DDPM AND GRAPH ATTENTION NETWORK-BASED AI DRUG RESPONSE PREDICTION SYSTEM AND METHOD THEREOF

KR20260065059AKR 20260065059 AKR20260065059 AKR 20260065059AKR-20260065059-A

Abstract

The artificial intelligence drug responsiveness prediction system based on DDPM and graph attention networks according to the present invention comprises a genetic data input unit that collects genetic data of a patient and It includes a gene data preprocessing unit that selects important genes by calculating the proximity between target genes and biological pathway genes, a drug responsiveness prediction artificial intelligence model that receives preprocessed gene data, generates augmented data similar to the original data through DDPM, and predicts drug responsiveness through a graph attention network, and a drug responsiveness output unit that outputs predicted drug responsiveness.

Inventors

박상현
최승연
서상민

Assignees

연세대학교 산학협력단

Dates

Publication Date: 20260508
Application Date: 20241031

Claims (12)

In an artificial intelligence drug responsiveness prediction system based on DDPM (Denoising Diffusion Probabilistic Models) and graph attention networks, Gene data input unit for collecting patient's genetic data; A gene data preprocessing unit that selects important genes by calculating the proximity between target genes and biological pathway genes; A drug responsiveness prediction AI model that receives preprocessed genetic data as input, generates augmented data similar to the original data through DDPM, and predicts drug responsiveness through a graph attention network; and Drug responsiveness output unit that outputs predicted drug responsiveness A system characterized by including
In paragraph 1, A system characterized by the above-mentioned gene data input unit including patient gene data comprising cancer cell line information data, GE, and reactivity information data between the cancer cell line and the drug compound.
In paragraph 1, A system characterized by the above-mentioned gene data preprocessing unit accessing at least one database among GDSC (Genomics of Drug Sensitivity in Cancer), TCGA (The Cancer Genome Atlas), and PDX (Patient-Derived Xenograft) to obtain data.
In paragraph 1, A system characterized by the above-mentioned gene data preprocessing unit analyzing biological pathways to select important genes and generating a latent space reflecting biological relationships through a graph autoencoder (GAE).
In paragraph 1, A system characterized by the above-described drug responsiveness prediction artificial intelligence model performing generalization of augmented data using the above-described DDPM.
In paragraph 2, The above drug responsiveness prediction AI model predicts drug responsiveness based on the biological pathway between target proteins and pathway genes through a graph attention network, and A system characterized by the above graph attention network forming a graph structure based on biological pathways related to drug target genes, each gene being represented as a node in the graph, graph learning being performed by reflecting the relationships between genes through connection information between nodes, and generating a prediction result for drug responsiveness by combining information from each node after performing independent graph attention operations for each biological pathway.
In a drug responsiveness prediction system, regarding a method for predicting drug responsiveness, Step of collecting patient's genetic data; A step of selecting important genes by calculating the proximity between target genes and pathway genes; A step of generating augmented data from selected gene data using DDPM (Denoising Diffusion Probabilistic Models); A step of predicting drug responsiveness through a graph attention network using augmented data; and Step of outputting the predicted drug responsiveness above A method for predicting drug responsiveness characterized by including
In Paragraph 7, A method characterized by the step of collecting genetic data of the above-mentioned patient, which includes collecting genetic data including cancer cell line information data, GE, and reactivity information data between the cancer cell line and the drug compound.
In Paragraph 7, The above gene data preprocessing step is characterized by obtaining data by connecting to at least one of the databases GDSC (Genomics of Drug Sensitivity in Cancer), TCGA (The Cancer Genome Atlas), and PDX (Patient-Derived Xenograft).
In Paragraph 7, A method characterized by the above gene data preprocessing step analyzing biological pathways to select important genes and generating a latent space reflecting biological relationships through a graph autoencoder (GAE).
In Paragraph 7, A method characterized by a step of predicting drug responsiveness based on biological pathways between target proteins and pathway genes through a graph-based attention network.
In Paragraph 11, A method characterized by the above graph attention network forming a graph structure based on biological pathways related to drug target genes, each gene being represented as a node in the graph, graph learning being performed by reflecting the relationships between genes through connection information between nodes, and generating a prediction result for drug responsiveness by combining the information of each node after performing independent graph attention operations for each biological pathway.

Description

DDPM and Graph Attention Network-Based AI Drug Response Prediction System and Method Thereof The present invention relates to an artificial intelligence drug responsiveness prediction system and a prediction method based on DDPM and a graph attention network, and more specifically, to a drug response prediction system that effectively processes high-dimensional data and improves prediction performance by establishing a data augmentation and drug response prediction model utilizing genetic features through biological pathway analysis. Predicting drug responsiveness is a crucial element in modern precision medicine, which aims to develop new drugs and find effective treatments tailored to patients' genetic characteristics. This is because different effects are observed among patients diagnosed with the same type of cancer and identified as having similar progression, even when the same treatment method is applied. It is known that the primary reason for differing responsiveness to treatment—such as varying therapeutic effects or side effects among patients even when similar treatment methods and prescriptions are applied—is the unique genetic characteristics that differ from person to person. Therefore, there is a need to research treatment methods that consider the patient's genetic characteristics, and with the recent increase in genomic research, studies analyzing the patient's genetic characteristics are continuing. FIG. 1 is a diagram illustrating a drug response prediction system that predicts a patient's unique drug responsiveness according to conventional technology. A conventional drug response prediction system (101) used omics data (11) containing patient genome data and various genetic information as training data for a machine learning model (21). In addition, models are being researched to more accurately predict a patient's unique drug responsiveness by integrating or applying various machine learning methods. However, conventional prediction models face problems that hinder predictive efficacy due to the inherent characteristics of the data. In particular, biological data contains high-dimensional information, and it is difficult to train accurate prediction models when the sample size is small. In the case of genetic data, the sample size is limited, and due to the complex relationships between genes, data analysis becomes more complex and a risk factor of overfitting arises during model training. To overcome these issues, it is necessary to simulate new data points similar to actual gene expression profiles for AI training methods using genetic data. Furthermore, as biological information is incorporated into drug responsiveness prediction, there is a need for prediction methods capable of capturing complex gene interactions. FIG. 1 is a diagram illustrating a conventional drug responsiveness prediction system (101). FIG. 2 is a diagram illustrating an artificial intelligence drug responsiveness prediction system (100) according to an embodiment of the present invention. FIG. 3 is a flowchart illustrating the process of establishing an artificial intelligence model for predicting drug responsiveness according to an embodiment of the present invention. FIG. 4 is a flowchart illustrating a gene data preprocessing process according to an embodiment of the present invention. FIG. 5 is a flowchart illustrating the DDPM execution process of the drug responsiveness prediction artificial intelligence model (130) according to an embodiment of the present invention. FIG. 6 is a flowchart illustrating the drug responsiveness prediction process of the drug responsiveness prediction artificial intelligence model (130) according to an embodiment of the present invention. FIG. 7 is a diagram illustrating the prediction performance of an artificial intelligence model (130) for predicting drug responsiveness according to an embodiment of the present invention. FIG. 8 is a flowchart illustrating a method for predicting drug responsiveness according to an embodiment of the present invention. FIG. 9 illustrates a computing device that implements a drug responsiveness prediction system (100) according to an embodiment of the present invention. The present invention will be described below with reference to the attached drawings. However, the present invention may be implemented in various different forms and is therefore not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification have been given similar reference numerals. Throughout the specification, when it is stated that a part is "connected (connected, in contact, combined)" with another part, this includes not only cases where they are "directly connected," but also cases where they are "indirectly connected" with other members interposed between them. Furthermore, when it is stated that a