Abstract Abstract 3857 The diagnosis of patients with myelodysplastic syndromes (MDS) is largely dependent on morphologic examination of bone marrow aspirates. Several criteria that form the basis of the classifications and scoring systems most commonly used in clinical practice are affected by operator-dependent variation. In order to identify more standardized molecular markers that would allow a more reliable prediction of prognosis, we have used gene expression profiling (GEP) data on CD34+ cells from MDS patients to determine the relationship between gene expression levels and prognosis in this disorder. GEP data on CD34+ cells from 125 MDS patients with a minimum 12-month follow-up since date of bone marrow sample collection were included in this study. Prediction for overall survival was performed using supervised principal components (“SuperPC”) and lasso penalized Cox proportional hazards regression applying the “Coxnet” algorithm. Supervised principal components analysis was performed on patients randomly split in a training set (n=84) and a test set (n=41), and 139 genes were identified the expression of which was significantly associated with MDS patient survival, including LEF1, CDH1, WT1 and MN1. In order to identify a smaller set of genes associated with patient survival, a second approach aiming at building sparse prediction models was used. A model was generated using the Coxnet algorithm and a predictor consisting of 20 genes was identified. Eight genes identified by the supervised principal components method were in common with the genes identified by the Coxnet model: ADHFE1, BTBD6, CPT1B, LEF1, FRMD6, GPR114, C7orf58 and LOC100286844. The Coxnet predictor outperformed other predictors including one which additionally used clinical information. To validate our findings, we evaluated the performance of our prognostic Coxnet gene signature in an independent gene expression profiling dataset on MDS bone marrow mononuclear cells (Mills et al, Gene Expression Omnibus series GSE15061). Our Coxnet gene signature based on CD34+ cells significantly identified a low-risk patient group in this independent GEP dataset based on unsorted bone marrow mononuclear cells, demonstrating that our signature is robust and may be applicable to bone marrow cells without the need to isolate CD34+ cells. These GEP-based signatures correlating with clinical outcome may significantly contribute to a refined risk classification of MDS. Disclosures: No relevant conflicts of interest to declare.