Decoding Dendritic Cell Subtypes via Integrated Radiogenomics: A Stacked Ensemble Model for Predicting Immunotherapy...

Area of Science:

• Immunology
• Oncology
• Computational Biology

Background:

• Anti-PD-1 immunotherapy is a key treatment for non-small cell lung cancer (NSCLC).
• Understanding the mechanisms of response, particularly immune cell involvement, is crucial for improving therapeutic outcomes.
• Dendritic cells (DCs) play a significant role in modulating anti-tumor immunity.

Purpose of the Study:

• To develop a multimodal framework integrating single-cell RNA sequencing (scRNA-seq), radiomics, and deep learning.
• To decipher DC-mediated mechanisms underlying anti-PD-1 response in NSCLC.
• To identify predictive biomarkers for immunotherapy response.

Main Methods:

• scRNA-seq analysis of tumor samples from NSCLC patients (responders vs. non-responders).
• Radiomic feature extraction and deep learning models (LSTM, ResNet50).
• Cell-cell communication and pseudotime analyses to identify key DC subsets and marker genes.

Main Results:

• Significant differences in DC subsets were observed between responders and non-responders.
• Conventional DC2 (cDC2) and tolerogenic DC (tDC) were identified as key subsets linked to therapeutic outcomes.
• Six DC-associated genes (FAM3B, TFAP2A, RTKN2, XCL2, KRT6A, RAB27B) showed predictive value.
• A stacked ensemble learning model integrating transcriptomic, clinical, and radiomic data achieved high accuracy (0.97) and AUC (0.99) in predicting response.

Conclusions:

• The multimodal framework combining scRNA-seq and radiomics accurately predicts immunotherapy response in NSCLC.
• Specific DC subsets and their associated marker genes hold potential as prognostic biomarkers for anti-PD-1 therapy.
• This approach offers insights into DC-mediated mechanisms influencing immunotherapy efficacy.