Deep learning identifies a T-cell exhaustion-dependent transcriptional signature for predicting clinical outcomes and...

Area of Science:

• Immunology
• Oncology
• Computational Biology

Background:

• Immune checkpoint blockade (ICB) therapy shows promise in cancer treatment but benefits only a subset of patients.
• Understanding T-cell exhaustion (TEX) dynamics during ICB is crucial for improving treatment efficacy.
• Developing predictive biomarkers for ICB response remains a significant clinical challenge.

Purpose of the Study:

• To investigate dynamic molecular profiles of T-cell exhaustion (TEX) during ICB treatment.
• To identify a transcriptional signature associated with ICB response.
• To develop a predictive model for clinical ICB response and patient survival.

Main Methods:

• Single-cell and bulk transcriptome analysis to study TEX molecular profiles.
• Ensemble deep-learning computational framework to identify an ICB-associated gene signature (ITGs).
• Machine-learning model (MLTIP) incorporating ITGs to predict ICB response and survival across multiple cohorts.

Main Results:

• Distinct TEX molecular profiles were identified and associated with ICB response.
• A 16-gene signature (ITGs) predictive of ICB response was discovered.
• The MLTIP model demonstrated reliable predictive power (AUC 0.778) for clinical ICB response and improved overall survival prediction (HR=0.093).
• MLTIP outperformed existing markers, showing a 21.5% average increase in AUC.

Conclusions:

• The identified TEX-dependent transcriptional signature (ITGs) shows potential as a tool for precise patient stratification.
• The MLTIP model offers a promising approach for personalized immunotherapy strategies.
• These findings support the clinical translation of TEX signatures in precision medicine for ICB therapy.