Unlocking reproducible transcriptomic signatures for acute myeloid leukaemia: Integration, classification and drug repurposing

Abstract

Acute myeloid leukaemia (AML) is a highly heterogeneous disease, which lead to various findings in transcriptomic research. This study addresses these challenges by integrating 34 datasets, including 26 control groups, 6 prognostic datasets and 2 single-cell RNA sequencing (scRNA-seq) datasets to identify 10,000 AML-related genes (ARGs). We focused on genes with low variability and high consistency and successfully discovered 191 AML signatures (ASs). Leveraging machine learning techniques, specifically the XGBoost model and our custom framework, we classified AML subtypes with both scRNA-seq and bulk RNA-seq data, complementing the ELN2022 classification approach. Our research also identified promising treatments for AML through drug repurposing, with solasonine showing potential efficacy for high-risk AML patients, supported by molecular docking and transcriptomic analyses. To enhance reproducibility and customizability, we developed CSAMLdb, a user-friendly database platform. It facilitates the reuse and personalized analysis of nearly all results obtained in this research, including single-gene prognostics, multi-gene scoring, enrichment analysis, machine learning risk assessment, drug repositioning analysis and literature abstract named entity recognition. CSAMLdb is available at http://www.csamldb.com.