The wide heterogeneity of osteoarthritis (OA) is the biggest challenge for classifying OA, predicting disease progression and developing effective therapeutics. Present OA clinical classifications fail to subset the disease and this results in inconsistent response to therapeutics. This is most likely due to the fact that the different OA clinical phenotypes consist of overlapping molecular endotypes, which are yet to be defined. The application of metabolomics approach in our research has generated much promising results. Specifically, we found that the conversion pathway of phosphatidylcholines (PCs) to lysophosphatidylcholines (lysoPCs) was overactivated in OA patients and plasma/serum lysoPCs to PCs ratio was significantly associated with OA risk. Patients with higher plasma lysoPCs to PCs ratio was 2.3 times more likely to undergo total knee joint replacement surgery in 10 years follow-up. The ratio was also significantly associated with knee cartilage volume loss measured by MRI over two years, and the ratio can predict who would respond well to symptomatic drugs including licofelone and naproxen. We found that high blood phenylalanine level was associated with both unilateral and bilateral radiographic knee OA progression in 5 years follow-up. Thus, OA patients should be advised to avoid any food/drinks containing large amount of aspartame which could raise blood phenylalanine levels. Arginine deficiency was also found in OA patients, suggesting arginine supplement may help slow down OA progression. Total joint replacement therapy (TJR) is by far the most effective treatment for end-stage OA patients, however, up to one third of TJR patients either do not achieve symptomatic improvement or deteriorate after TJR. We found that three metabolic ratios (C2 to PC ae C40:1, PC aa C36:4, and glutamine to isoleucine) related to inflammation and muscle breakdown could predict non-responders to TJR. More recently, we found in a large OA cohort that OA had at least three distinct endotypes characterized by different metabolic markers. The results are novel and have potential in developing precision medicine tools for OA management.