Diabetes is a common disease that affects 1 in 6 Koreans over the age of 30. It is a dangerous chronic disease that causes complications such as stroke or cardiovascular disease, so prevention is important. Lifestyle including diet and genetic conditions are important for the onset of diabetes, and research into predictive models based on such information is steadily progressing.
Existing diabetes risk prediction model studies have mainly targeted Western populations. Even for Asians, demographic information such as height, weight, and family history, or clinical information such as glycated hemoglobin (HbA1c) and cholesterol levels were mainly used. As such, there were limitations in predicting diabetes that reflected genetic and environmental factors specific to Koreans.
In addition, a joint research team led by Professor Jeonghye Lee of UNIST’s Department of Industrial Engineering (President Yonghoon Lee) and Professor Jihoon Kang of the Department of Family Medicine at Kosin University Gospel Hospital (Director Kyungseung Oh) developed artificial intelligence ( Director Kyungseung Oh). AI) machine that improved the predictive performance of the onset of type 2 diabetes based on a large-scale Korean cohort. Developed a learning model (machine learning).
The machine learning model developed this time can identify the risk of developing diabetes, specific to Korean people, and provide event factors. The research team expects that if this model is used in clinical settings, type 2 diabetes can be effectively prevented and responded to.
The research team developed a Genome-wide Polygenic Risk Score (gPRS) specific to Koreans and used a combination of demographic, clinical and metabolome information.
In particular, the research team took on the challenge of developing a predictive model using information specific to Korea. It is based on large-scale cohort data from the Korea Genome Epidemiology Survey (KoGES) collected by the National Institutes of Health under the Centers for Disease Control and Prevention. This cohort has been tracked and collected since 2001 to study chronic diseases such as diabetes, hypertension, obesity, and metabolic syndrome, which are common among Koreans.
The finally developed type 2 diabetes onset machine learning prediction model had approximately 11 percentage points (%c) higher predictive performance than using demographic information alone. Compared to the case where demographic information and clinical information were also used, it showed a better predictive performance of about 4 percentage points (%c) or more. The predictive performance of diabetes improved by adding genetic and environmental information to demographic and clinical data.
First author Seok-Joo Han, a doctoral researcher in the Department of Industrial Engineering at UNIST, said, “To obtain genetic information on the onset of type 2 diabetes, a ‘polygenic risk score’ was newly calculated according to the genetic characteristics of the Koreans and used . in the prediction model.” By reflecting it as a ‘sieve’, we supplemented the information that genetic information could not explain.”
Co-author Soo-Hyun Kim, a Ph.D. in the Department of Industrial Engineering at UNIST, “As we obtain demographic and clinical information from the Korean cohort and add the newly developed polygenic risk score and metabolomic information, the prediction accuracy of the model increases.” he emphasized.
Professor Jeonghye Lee said, “It is very meaningful to change the method from a Western study that focuses on a cohort to a Korean cohort,” and he expected that it could be used in various subsequent studies using Asian cohort data.
Meanwhile, this study was published in eBioMedicine, a sister journal of The Lancet, a leading academic journal in the medical field, ‘Predicting Type 2 Diabetes Using Genome-Wide Polygenic Risk Scores and Metabolic Profiles: A 10-Year Prospective Population-Based Study ‘. Predicting Type 2 Diabetes Using Genome-Wide Polygenic Risk Score and Metabolic Profiles: Machine Learning Analysis of a 10-Year Population-Based Cohort Studydown)’ published on the 1st.
