Abstract LB111: Convolutional neural network-based artificial intelligence for immune subtype classification in pan-cancer

Abstract Background: Pan-cancer analysis offers a holistic perspective on cancer, spanning 33 types and over 11,000 tumor samples from The Cancer Genome Atlas (TCGA). This analysis provides key insights into molecular features shared across cancers and aids in identifying immune subtypes, which are critical for developing personalized immunotherapies. Six distinct immune subtypes were identified in the immune landscape of cancer, offering a framework to explore tumor-immune interactions and their impact on prognosis. Methods: RNA sequencing (RNA-Seq) data were extracted from the TCGA Pan-Cancer Atlas, normalized, and processed for classification. Convolutional neural networks (CNNs), a subset of artificial intelligence, were employed to classify immune subtypes. The CNN model utilized convolutional layers with ReLU activation, dropout regularization, and batch normalization to enhance robustness and prevent overfitting. The dataset was split into training, validation, and testing subsets to evaluate model performance. Performance metrics, including accuracy and the area under the curve (AUC), were used to benchmark the CNN against traditional machine learning methods such as XGBoost, Random Forest, and TabNet. Results: The CNN model achieved an AUC of 0.993 and an overall accuracy of 98.85%, outperforming traditional classification methods. It demonstrated resilience to class imbalance and accurately classified underrepresented immune subtypes. The approach was validated against independent datasets, demonstrating consistent performance across diverse processing pipelines and normalization methods. Conclusions: This study highlights the utility of artificial intelligence, particularly CNNs, in accurately classifying immune subtypes across pan-cancer datasets. By leveraging RNA-Seq data, this approach offers a scalable and robust framework for pan-cancer analysis, paving the way for advancements in precision oncology and personalized immunotherapy. Future efforts will incorporate multi-omics data and validate these findings in clinical settings to enhance their translational potential. Citation Format: Minh Huu Nhat Le, Ha-Hieu Pham, Huy Quoc Nguyen, Hong Xuan Ong, Hien Quang Kha, Phat Ky Nguyen, Thanh-Huy Nguyen, Han Hong Huynh, Dang Nguyen, Thanh-Minh Nguyen, An Thuy Vo, Thuy Vu Minh Nguyen, Trung Minh Tu Tran, Lam Huu Phuc Nguyen, Nguyen Quoc Khanh Le. Convolutional neural network-based artificial intelligence for immune subtype classification in pan-cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_2):Abstract nr LB111.