AI-Driven Deep Learning Approach for Pan-Cancer Immune Profiling.

Area of Science:

• Oncology
• Immunology
• Bioinformatics
• Computational Biology

Background:

• The tumor immune microenvironment (TME) is crucial for cancer progression and therapeutic response.
• RNA-sequencing (RNA-Seq) has enabled the identification of distinct immune subtypes within the TME.
• These subtypes include Wound Healing (WH), IFN- Dominant (IFNG), Inflammatory (INF), Lymphocyte Depleted (LD), Immunologically Quiet (IQ), and TGF-β Dominant (TGFb).

Purpose of the Study:

• To develop and evaluate a Convolutional Neural Network (CNN) model for classifying the six known RNA-Seq-defined immune subtypes of the TME.
• To assess the CNN model's performance in handling class imbalance and capturing complex gene expression interactions.

Main Methods:

• Utilized RNA-Seq data to train a Convolutional Neural Network (CNN) model.
• The CNN architecture incorporated ReLU activation and dropout for improved performance.
• Evaluated model performance using 10-fold cross-validation, measuring F1-score and Area Under the Curve (AUC).
• Compared CNN performance against other machine learning models including XGBoost, Random Forest, and TabNet.

Main Results:

• The CNN model achieved a high 10-fold F1-score of 0.9483 and an AUC of 0.9969.
• Demonstrated superior performance in classifying TME immune subtypes compared to XGBoost, Random Forest, and TabNet.
• Confirmed the CNN's capability to effectively manage class imbalance and model intricate gene interaction patterns.

Conclusions:

• Convolutional Neural Networks are highly effective for classifying tumor immune microenvironment subtypes based on RNA-Seq data.
• The developed CNN model offers a robust tool for immune profiling in cancer research.
• Future research will focus on validating these findings on independent datasets and integrating multi-omics data for enhanced accuracy.