Deep learning-based transcriptome model predicts survival of T-cell acute lymphoblastic leukemia

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Accurate prognostication in T-cell acute lymphoblastic leukemia (T-ALL) is crucial for treatment decisions.
• Current methods for predicting T-ALL survival across diverse patient groups are limited.
• Identifying distinct T-ALL subgroups with poor prognosis can enhance patient outcomes.

Purpose of the Study:

• To develop a deep learning (DL) model for prognostic staging of T-ALL patients.
• To identify T-ALL subgroups with significantly different survival rates.
• To validate the DL model's performance in predicting T-ALL patient survival.

Main Methods:

• Utilized transcriptome sequencing data from the TARGET initiative for model development.
• Constructed a DL-based survival model using data from 265 T-ALL patients.
• Validated the model's predictive accuracy on an independent clinical cohort.

Main Results:

• The DL model successfully stratified T-ALL patients into two subgroups (K0 and K1) with statistically significant survival differences (P<0.0001).
• The more aggressive subgroup was linked to tumor-related signaling pathways including PI3K-Akt, cGMP-PKG, and TGF-beta.
• The DL model demonstrated robust performance in the clinical validation cohort (P=0.0248).

Conclusions:

• A DL-based model can effectively predict T-ALL patient survival.
• The identified subgroups and associated pathways offer insights into T-ALL malignancy.
• The developed DL model holds potential for clinical application in T-ALL prognostication.