Systems Biology and Machine Learning Approach Stratifies Patient Risk in Pediatric Acute Myeloid Leukemia
Researchers at the Herbert Irving Comprehensive Cancer Center (HICCC) have identified a new risk stratification for patients with pediatric acute myeloid leukemia (AML), which could potentially direct treatment options. In the first study combining a systems biology approach looking at regulatory networks with machine learning for pediatric AML risk stratification, the team from Andrea Califano’s lab in the Department of Systems Biology identified key proteins that predicted risk in a dataset of pediatric patients with AML. They presented the results at this year’s American Association for Cancer Research (AACR) held virtually April 10-15.
Instead of traditional approaches using raw gene expression data to stratify patients, the team ‘reverse-engineered’ the gene regulatory network of AML, using two algorithms developed by the Califano lab, ARACNe and VIPER. ARACNe and VIPER work together to infer transcription factor activity, identifying relationships between regulators (the transcription factors), and the target genes that they control.
Transforming the vast array of raw gene expression data into the inferred transcription factor (TF) activity, ARACNe identified targets for each transcription factor, and VIPER inferred the TF activity based on the expression of those targets. From there, the researchers used machine learning techniques to identify the TFs that are considered master regulators, or key drivers of, pediatric AML.
The team validated their findings in external datasets, identifying nine different risk groups with significantly different survival rates. In addition, the master regulator framework the team identified helped to reclassify FLT3-ITD, a subtype of pediatric AML, into two different subtypes, high-risk and a low-risk. Currently, the recommended treatment for FLT3-ID patients includes bone marrow transplant, a major medical procedure that can have serious complications, including graft-versus-host disease (GVHD), where the donor cells attack the host cells, a potentially life-threatening condition. This new re-classification could mean that low-risk FLT3-ID subtype patients may not need to undergo this higher-risk therapy.
Next steps for the team include validating the results in vitro or in vivo. The master regulators identified in the study also represent potential new therapeutic targets, and the team plans on embarking on new research to predict drugs that may be used as therapies for the high-risk pediatric AML patients.