Frontiers in Genetics (Oct 2023)
Predicting leukemic transformation in myelodysplastic syndrome using a transcriptomic signature
Abstract
Background: For prediction on leukemic transformation of MDS patients, emerging model based on transcriptomic datasets, exhibited superior predictive power to traditional prognostic systems. While these models were lack of external validation by independent cohorts, and the cell origin (CD34+ sorted cells) limited their feasibility in clinical practice.Methods: Transformation associated co-expressed gene cluster was derived based on GSE58831 (‘WGCNA’ package, R software). Accordingly, the least absolute shrinkage and selection operator algorithm was implemented to establish a scoring system (i.e., MDS15 score), using training set (GSE58831 originated from CD34+ cells) and testing set (GSE15061 originated from unsorted cells).Results: A total of 68 gene co-expression modules were derived, and the ‘brown’ module was recognized to be transformation-specific (R2 = 0.23, p = 0.005, enriched in transcription regulating pathways). After 50,000-times LASSO iteration, MDS15 score was established, including the 15-gene expression signature. The predictive power (AUC and Harrison’s C index) of MDS15 model was superior to that of IPSS/WPSS in both training set (AUC/C index 0.749/0.777) and testing set (AUC/C index 0.933/0.86).Conclusion: By gene co-expression analysis, the crucial gene module was discovered, and a novel prognostic system (MDS15) was established, which was validated not only by another independent cohort, but by a different cell origin.
Keywords
