Machine Learning Assistants Construct Oxidative Stress-Related Gene Signature and Discover Potential Therapy Targets for Acute Myeloid Leukemia

Background. Oxidative stress (OS) is associated with the development of acute myeloid leukemia (AML). However, there is lack of relevant research to confirm that OS-related genes can guide patients in risk stratification and predict their survival probability. Method. First, we Data from three public databases, respectively. Then, we use batch univariate Cox regression and machine learning to select important characteristic genes; next, we build the model and use receiver operating characteristic curve (ROC) to evaluate the accuracy. Moreover, GSEAs were performed to discover the molecular mechanism and conduct nomogram visualization. In addition, the relative importance value was used to identify the hub gene, and GSE9476 was to validate hub gene difference expression. Finally, we use symptom mapping to predict the candidate herbs, targeting the hub gene, and put these candidate herbs into Traditional Chinese Medicine Systems Pharmacology (TCMSP) to identify the main small molecular ingredients and then docking hub proteins with this small molecular. Results. A total of 313 candidate oxidative stress-related genes could affect patients’ outcomes and machine learning to select six potential genes to construct a gene signature model to predict the overall survival (OS) of AML patients. Patients in a high group will obtain a short survival time when compared with the low-risk group (HR=3.97, 95% CI: 2.48-6.36; p