Multi-cancer classification; An analysis of neural network models

• Published in: Machine learning with applications Vol. 12; p. 100468
• Main Authors: Webber, James W., Elias, Kevin
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 15.06.2023; Elsevier
• Subjects: Cancer prediction; Early detection; Expression analysis; Multi-classification; Neural networks; Expression analysis; Multi-classification; Neural networks; Cancer prediction; Early detection
• ISSN: 2666-8270; 2666-8270
• DOI: 10.1016/j.mlwa.2023.100468

Cancer identification is generally framed as binary classification, normally discrimination of a control group from a single cancer group. However, such models lack any cancer-specific information, as they are only trained on one cancer type. The models fail to account for competing cancer risks. Pan-cancer evaluation requires a model trained on multiple cancer types, and controls, simultaneously, so that a physician can be directed to the correct area of the body for further testing. We investigate neural network models to address multi-cancer classification problems across several data types commonly applied in cancer prediction, including circulating miRNA expression, protein, and mRNA. In particular, we present an analysis of neural network depth and type, and investigate how this relates to classification performance. In our comparisons, we include several state-of-the-art neural networks from the literature. We provide details on the optimal network depth and type, the activation functions and layer sizes. : Our analysis evidences that shallow (i.e., 1 or 2 layer), feed-forward neural network architectures offer greater performance in terms of mean sensitivity and precision when compared to deeper (i.e., >2 layer) feed-forward models, Convolutional Neural Network (CNN), and Graph CNN (GCNN) architectures, across a range of measurement technologies in cancer prediction (e.g., miRNA, mRNA and protein). We also discover that hyperbolic tangent activation functions offer the most consistent performance, and the optimal feed-forward models have descending layer size structure.