A parsimonious 3-gene signature predicts clinical outcomes in an acute myeloid leukemia multicohort study

• DOI: 10.1182/bloodadvances.2018030726
• Publication Date: 2019-04-23T16:55:22Z
• Authors: Sarah Wagner; Jayakumar Vadakek...; Sarah K. Tasian; Heidi Altmann; Martin Bornhäuser; A. Graham Pockley; Graham R. Ball; Sergio Rutella
• Keywords: Signature (topology); Gene signature

Abstract Acute myeloid leukemia (AML) is a genetically heterogeneous hematological malignancy with variable responses to chemotherapy. Although recurring cytogenetic abnormalities and gene mutations are important predictors of outcome, 50% 70% AMLs harbor normal or risk-indeterminate karyotypes. Therefore, identifying more effective biomarkers predictive treatment success failure essential for informing tailored therapeutic decisions. We applied an artificial neural network (ANN)–based machine learning approach publicly available data set discovery cohort 593 adults nonpromyelocytic AML. ANN analysis identified parsimonious 3-gene expression signature comprising CALCRL, CD109, LSP1, which was event-free survival (EFS) overall (OS). computed prognostic index (PI) using normalized gene-expression levels β-values from subsequently created Cox proportional hazards models, coupled clinically established prognosticators. Our PI separated the adult patients in each European LeukemiaNet risk category into subgroups different probabilities very high–risk features, such as those high either FLT3 internal tandem duplication nonmutated nucleophosmin 1. The remained significantly associated poor EFS OS after adjusting prognosticators, its ability stratify validated 3 independent cohorts (n = 905 subjects) 1 childhood AML 145 subjects). Further silico analyses that only tumor type among 39 distinct malignancies concomitant upregulation LSP1 predicted survival. our ANN-derived refines accuracy patient stratification potential improve outcome prediction.