606 – Biomarkers and Endpoint Validation 2
Identifying Metabolic Signatures for Chronic Kidney Disease in Type II Diabetic Patients
Minya Pu
University of California, San Diego
Rintaro Saito
University of California, San Diego
Youyi Zhang
University of California, San Diego
Yurong Guo
Kumar Sharma
University of California, San Diego
Loki Natarajan
University of California, San Diego
Diabetic patients with chronic kidney disease (CKD) are at much higher risk of morbidity, so it is of great importance to understand the disease mechanism. We used metabolomics data to explore the association between estimated glomerular filtration rate (GFR) values, a kidney function outcome, and a panel of urine biomarkers consisting of about 100 metabolites. A total of 114 type II diabetic patients were included in this analysis. We used two approaches, LASSO and k-TSP, to classify patients into DM+CKD and DM-CKD groups. We also used LASSO to explore the metabolites that were associated with disease severity in which eGFR values were used as a continuous outcome. A bootstrap-permutation based stability analysis was performed to assess the reproducibility of each variable in a LASSO model. We also showed that LASSO produced a lower leave-one-out cross-validation error rate than k-TSP in the training data set (1.3% vs. 6.6%), and also a slightly lower prediction error rate in the validation set (5.3% vs. 7.9%). A newer top scoring method may help to improve the error rates.