Feature reduction for machine learning on molecular features: The GeneScore

TL;DR

The paper introduces GeneScore, a feature reduction method that combines various molecular data types into a single score, improving cancer classification accuracy in biomedical machine learning applications.

Contribution

GeneScore is a novel, knowledge-based feature reduction technique that outperforms binary matrices in cancer classification tasks using diverse molecular data.

Findings

GeneScore outperforms binary matrices in classification accuracy.

Integrates multiple molecular data types into a single score.

Utilizes expert knowledge for feature reduction.

Abstract

We present the GeneScore, a concept of feature reduction for Machine Learning analysis of biomedical data. Using expert knowledge, the GeneScore integrates different molecular data types into a single score. We show that the GeneScore is superior to a binary matrix in the classification of cancer entities from SNV, Indel, CNV, gene fusion and gene expression data. The GeneScore is a straightforward way to facilitate state-of-the-art analysis, while making use of the available scientific knowledge on the nature of molecular data features used.