Distance-based classifiers as potential diagnostic and prediction tools for human diseases.

Area of Science:

• Genomics
• Biostatistics
• Biomarker Discovery

Background:

• High-throughput gene expression profiling (e.g., RNAseq, microarrays) is common for biomarker discovery.
• Existing analytical pipelines face the "curse of dimensionality," leading to poor reproducibility of gene signatures.
• The number of genes often exceeds the number of patients, hindering robust algorithm training.

Purpose of the Study:

• To propose a novel approach for gene expression biomarker discovery that accounts for all assayed genes.
• To address the reproducibility issues inherent in current gene signature extraction methods.
• To develop a method that quantifies disease progression based on deviation from normal homeostasis.

Main Methods:

• Representing individual patient gene expression profiles as points in a multidimensional space.
• Clustering normal and affected samples within this multidimensional space.
• Calculating the degree of separation of a sample from the normal cluster to reflect disease-related drift.

Main Results:

• The novel approach was validated on a publicly available glioma dataset with survival data.
• Demonstrated applicability to classifying non-malignant conditions, using psoriasis as a model.
• The method showed potential for improved diagnostic signature reproducibility and disease state quantification.

Conclusions:

• The proposed cumulative gene expression pattern analysis offers a robust alternative to traditional signature extraction.
• This method enhances the understanding of pathophysiological processes by measuring deviation from homeostasis.
• The approach holds promise for more reliable biomarker discovery and disease classification across various medical conditions.