WD2-1
Kynurenine Pathway in Depression: A Systematic Review and Meta-analysis
うつ病患者におけるキヌレニン経路の代謝産物濃度に関する体系的レビューとメタ解析
Ogyu Kamiyu（尾久 守侑）1，久保 馨彦1，野田 賀大1，岩田 祐輔1,2，津川 幸子3，大村 祐貴3，和田 真孝1，垂水 良介1，エリック プリットマン2，森口 翔1,2，宮崎 貴浩1，内田 裕之1，アリエル グラフゲレーロ2，三村 將1，中島 振一郎1
1Department of Neuropsychiatry, Keio University School of Medicine, Tokyo, Japan
2Centre for Addiction and Mental Health, University of Toronto, Toronto, Canada
3Keio University School of Medicine, Tokyo, Japan
Background: Evidence suggests that abnormalities of the kynurenine (KYN) pathway may be implicated in the pathophysiology of depression. However, the relationships between depression and each metabolite of the KYN pathway have not been systematically evaluated.Methods: A literature search was conducted using PubMed and Embase with the defined terms. Studies measuring levels of metabolites of the KYN pathway in patients with depression and healthy controls (HCs) were included. Standardized mean differences (SMDs) were calculated to determine differences in metabolite levels between the groups. Subgroup analyses were performed for metabolite measurement techniques and antidepressant-free status. Influences of patients' age, male ratio, and proportion of those medicated on study SMDs were assessed through meta-regression.Results: Out of 899 initial records, 22 articles were identified to form the empirical basis of this analysis. Seventeen, 10, and 18 studies examined levels of kynurenic acid (KYNA), quinolinic acid (QUIN), and KYN, respectively. Nineteen studies measured the metabolites in blood. KYNA and KYN levels were lower in patients with depression in comparison to HCs, while QUIN levels did not differ between the two groups. Subgroup analyses showed decreased KYNA levels and increased QUIN levels in antidepressant-free patients. Meta-regression noted that male ratios of the samples were negatively associated with study SMDs for KYNA.Conclusion: This meta-analysis revealed decreases in KYNA and KYN levels, and increases in QUIN levels, in patients with depression. However, given the small number of the included studies and heterogeneity among their sample characteristics, further research is clearly needed. | | WD2-2
Prediction of susceptibility to depression utilizing a machine learning model based on genome wide single nucleotide polymorphism data
ゲノムワイド遺伝子多型データを用いた機械学習によるうつ病脆弱性の予測
Takahashi Yuta（高橋 雄太）1,2,3，植木 優夫2，田宮 元2，荻島 創一2，福本 健太郎4，大塚 耕太郎4，富田 博秋2,3
1Department of Psychiatry, Tohoku University
2Tohoku Medical Megabank Organization, Tohoku University
3Department of Disaster Psychiatry, Tohoku University
4Department of Neuropsychiatry, Iwate Medical University
Genome -based prediction of the susceptibility to depression can be achieved by modeling with effective incorporation of a number of small effects. One of the most frequently utilized prediction models is genetic risk score methods. In this model, target phenotype-associated single nucleotide polymorphisms (SNPs) were selected out of univariate analysis information of training data set based on an arbitrary p-value threshold, and genetic risk score for each subject of test data set is calculated by summing products of regression coefficients of the selected SNPs and the standardized number of the respective alleles. However, the prediction accuracy of this method is low due to uncertainty of appropriate p-value threshold, ignorance of correlations among selected SNPs, and Winner’s Curse (i.e. the overestimation of selected SNPs). To overcome above mentioned problems, the Smooth-Threshold Multivariate Genetic Prediction (STMGP) method was developed, which sets best p-value screening threshold for prediction accuracies by estimating prediction errors, and utilizes generalized Ridge regression to selected SNPs, which enables to efficiently utilize both of correlated SNPs for better prediction accuracies, and decrease Winner’s Curse with weighting SNPs by statistics of univariate analyses. In this study, STMGP prediction model, along with the conventional approaches (genetic risk score, and G-BLUP) were trained utilizing genome data of 4,354 subjects and validated by an independent data from 3,456 subjects. Prediction accuracy estimated by area under the curve of STMGP, genetic risk score, and G-BLUP were 0.64, 0.51, 0.49, respectively. The novel machine learning method, STMGP, showed the best prediction accuracy compared with the conventional genetic prediction methods. |
