AutoTFCNNY: A multi-instance neural network for enhanced early cancer detection using TCR data.

Area of Science:

• Oncology
• Immunology
• Bioinformatics
• Artificial Intelligence

Background:

• Early cancer diagnosis is critical for improving patient survival rates and treatment efficacy.
• Peripheral blood T-cell receptors (TCRs) offer a non-invasive biomarker for cancer detection.
• Extracting cancer-specific information from complex TCR repertoires presents a significant challenge.

Purpose of the Study:

• To develop an advanced deep learning model for accurate and sensitive early cancer detection.
• To leverage T-cell receptor (TCR) repertoire data for identifying diverse cancer types.
• To address the challenge of extracting meaningful cancer-associated signals from heterogeneous TCR data.

Main Methods:

• Development of AutoTFCNNY, a multi-instance deep neural network integrating Transformer and Convolutional Neural Network (CNN) architectures.
• Utilizing a multi-instance learning (MIL) framework to model global dependencies and local features within TCR sequences.
• Training and validation of the model on peripheral blood TCR repertoires from patients with various cancer types.

Main Results:

• AutoTFCNNY achieved an average area under the ROC curve (AUC) exceeding 0.94 across 22 cancer types.
• Exceptional performance was observed in 18 cancer types, with AUCs surpassing 0.99, including brain and non-small cell lung cancer.
• The model demonstrated high accuracy, stability, and strong generalization capabilities in early cancer detection.

Conclusions:

• AutoTFCNNY effectively extracts crucial cancer-related information from peripheral blood TCR repertoires.
• The model shows significant potential as a non-invasive tool for early cancer prediction and diagnosis.
• Findings highlight the utility of deep learning approaches for analyzing complex immunological data in oncology.