Evaluating sources of variability in pathway profiling

TL;DR

This paper presents an open-source bioinformatics pipeline for pathway profiling that assesses variability sources and integrates machine learning to identify disease-specific pathways, demonstrated on Parkinson's disease data.

Contribution

It introduces a reproducible, integrated computational pipeline that evaluates variability in pathway profiling and applies it to Parkinson's disease microarray data.

Findings

Different machine learning methods yield low overlap in pathway enrichment.

Methods identify distinct but meaningful biological aspects of Parkinson's disease.

Integration of multiple approaches enhances biological insight.

Abstract

A bioinformatics platform is introduced aimed at identifying models of disease-specific pathways, as well as a set of network measures that can quantify changes in terms of global structure or single link disruptions.The approach integrates a network comparison framework with machine learning molecular profiling. <CA>The platform includes different tools combined in one Open Source pipeline, supporting reproducibility of the analysis. We describe here the computational pipeline and explore the main sources of variability that can affect the results, namely the classifier, the feature ranking/selection algorithm, the enrichment procedure, the inference method and the networks comparison function. The proposed pipeline is tested on a microarray dataset of late stage Parkinsons' Disease patients together with healty controls. Choosing different machine learning approaches we get low pathway profiling overlapping in terms of common enriched elements. Nevertheless, they identify different but equally meaningful biological aspects of the same process, suggesting the integration of information across different methods as the best overall strategy. All the elements of the proposed pipeline are available as Open Source Software: availability details are provided in the main text.