Over the past 30 years researchers described varied metabolic phenotypes in overweight and obesity with diverse disease outcomes. Well-conducted randomized large cohort clinical trials in individuals with overweight and prediabetes or type 2 diabetes suggest mixed response to caloric restriction or metformin, the first-line medication in type 2 diabetes [1]. In the Diabetes Prevention Program, 21% of individuals with prediabetes treated with metformin progressed to diabetes at 3 years [2]. Similarly, 55% of individuals with prediabetes in the Tubingen Lifestyle Intervention Program did not revert to normal glucose tolerance in response to energy restriction and moderate exercise intervention [3]. Using gold-standard phenotyping tools of hepatic and peripheral glucose regulation, we find different levels of insulin resistance in liver and muscle in individuals with obesity. Using latest-generation plasma metabolomic and lipidomic analyses combined with machine learning, random forest-based feature selection and classification we identified three plasma lipids that can classify sub-cohorts of liver versus muscle insulin resistance in obesity with remarkable accuracy. We propose that therapy guided by plasma biomarkers will be more therapeutically effective, have less side-effects, and be more cost-effective.