Abstract 1931: Multi-cancer classification: An analysis of neural network complexity

Abstract

Abstract Background: Often, in analysis of expression data, binary classification problems are considered, with the aim to separate a control group from a single cancer group (e.g., ovarian cancer). Such models lack any cancer specific information, as they are only trained on one cancer type. It is more useful and efficient to train a model on multiple cancer types, and controls, simultaneously, so that a physician can be directed to the correct area of the body for further testing. Results: We introduce novel, data-driven neural network models to address the multi-cancer classification problem across a number of data types commonly applied in cancer prediction, including circulating miRNA expression (blood draw), protein (tissue sample), and mRNA (tissue sample). In particular, we present an analysis of neural network depth and complexity and investigate how this relates to classification performance. Comparisons of our models are also given to the state-of-the-art neural network models from the literature. The proposed models yield high accuracy (mean AUC greater than 0.95) across all data types considered and are shown to offer greater performance when compared to the models from the literature. Conclusion: We have introduced a simple neural network framework which can be applied to multiple data types and which yields the best performance when compared to similar models from the literature. Upon analysis of the neural network complexity, we discover that shallow, feed-forward neural net architectures offer greater performance when compared to more complex deep feed-forward, Convolutional Neural Network (CNN), and Graph CNN (GCNN) architectures considered in the literature. The depth analysis indicates that shallow neural nets (e.g., 1 or 2 layers) are more favorable for this problem, when compared to deep architectures (e.g., more than 3 layers). The results show that multiple cancers and controls can be classified simultaneously, and accurately using the proposed models, across a range of expression technologies in cancer prediction. Citation Format: James W. Webber, Kevin M. Elias. Multi-cancer classification: An analysis of neural network complexity [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1931.