ASSL-Cancer: Pan-cancer Classification Based on Semi-supervised and Active Learning

Cancer is a major disease that threatens human health. Early diagnosis and precise treatment of cancer are crucial to improving patient survival. With the advancement of high-throughput sequencing technology, the use of machine learning in cancer diagnosis is becoming increasingly widespread. However, existing cancer classification methods mainly rely on supervised learning, which requires a large amount of labeled data for model training, while a large amount of unlabeled data containing valuable information is not fully utilized. To address this problem, we propose a pan-cancer classification method based on active semi-supervised learning, ASSL-Cancer, which aims to improve classification performance by prioritizing labeling of high-value samples and effectively utilizing unlabeled data. First, ASSL-Cancer combines active learning with a clustering strategy to ensure that the initial labeled set contains the most informative samples. Then, label spreading module is used to extend the label information to unlabeled samples and generate appropriate pseudo labels. Finally, all available sample information is input into a residual network-based classifier to predict the cancer type label of unlabeled samples. Experimental results show that ASSL-Cancer outperforms other based on supervised and semi-supervised classification methods in multiple classification indicators, with an overall classification accuracy of 90.89%. This confirms that ASSL-Cancer can significantly improve classification accuracy in scenarios with limited labeled data.