Hierarchical Classification for Predicting Metastasis Using Elastic-Net Regularization on Gene Expression Data

TL;DR

This paper presents a hierarchical classification approach using elastic-net regularization on gene expression data to accurately predict cancer tissue origin and metastasis status, offering insights into biological mechanisms.

Contribution

It introduces a novel elastic-net-based hierarchical model that improves prediction accuracy for metastasis and tissue origin from gene expression profiles.

Findings

Achieved 97% accuracy in tissue-of-origin prediction.

Achieved 90% accuracy in metastasis prediction.

Identified mitochondrial gene expression as negatively correlated with metastasis.

Abstract

Metastasis is a leading cause of cancer-related mortality and remains challenging to detect during early stages. Accurate identification of cancers likely to metastasize can improve treatment strategies and patient outcomes. This study leverages publicly available gene expression profiles from primary cancers, with and without distal metastasis, to build predictive models. We utilize elastic net regularization within a hierarchical classification framework to predict both the tissue of origin and the metastasis status of primary tumors. Our elastic net-based hierarchical classification achieved a tissue-of-origin prediction accuracy of 97%, and a metastasis prediction accuracy of 90%. Notably, mitochondrial gene expression exhibited significant negative correlations with metastasis, providing potential biological insights into the underlying mechanisms of cancer progression.