Despite advances in therapeutic development, an anthracycline-cytarabine induction regimen remains the gold standard for acute myeloid leukaemia (AML) treatment. However, reliable predictive markers for assessing treatment sensitivity, adjusting therapy intensity, and guiding the use of experimental therapies are still lacking. This study aimed to develop a predictive model of AML chemoresistance.

Transcriptome sequencing and DNA methylation analysis of leukaemic blasts were performed to identify differentially expressed and methylated genes between responding (RES) and non-responding (non-RES) patients. A logistic regression nomogram model was developed using obtained data to predict complete remission (CR) and was further validated.

Compared to RES patients, non-RES patients exhibited a significant overexpression of interferon-related DNA damage resistance signature (IRDS) genes at diagnosis. Based on the expression of three IRDS genes (IFIT5, IFI44L, IFI44), we developed the IRDS score, which demonstrated high predictive accuracy, with calculated probabilities of CR of 0.71 for RES patients and 0.31 for non-RES patients. Downregulation of histone and chromatin remodelling genes following therapy administration was a hallmark of a successful treatment response. Integrative analysis revealed 1108 genes with concordant changes in both gene expression and DNA methylation between RES and non-RES patients, including IRDS genes IFIT5 and IFI44L.

The IRDS score-based model predicts AML chemoresistance with high accuracy and feasibility. It is quick, cost effective, and requires readily available biological material. This tool shows promise for guiding treatment decisions and identifying candidates for intensified or experimental therapies.