Graph neural networks for investigating complex diseases: A case study on Parkinson's Disease

Poster (2023, July 25)

Omics data analysis is a critical component in the study of complex diseases, but the high dimension and heterogeneity of the data often pose challenges that are difficult to address by classical ... [more ▼]

Omics data analysis is a critical component in the study of complex diseases, but the high dimension and heterogeneity of the data often pose challenges that are difficult to address by classical statistical and machine learning methods. Recently, structured data analyses using graph neural networks (GNNs) have emerged as a promising complementary approach, particularly for investigating the relational information between samples. However, it is still unclear which strategies for designing and optimizing GNNs are most effective when working with real-world data from complex disorders, such as Parkinson's disease (PD). Our study addresses this gap by examining the application of various GNN models, including Graph Convolutional Network, ChebyNet, and Graph Attention Network, to identify and interpret discriminative patterns between PD patients and controls using omics data. The developed pipeline integrates Lasso penalty-based feature selection, similarity graph construction, and final modeling for sample classification. Through an end-to-end model building and evaluation process, we assess the practical utility of the pipeline on independent PD omics datasets. Overall, our analyses highlight some of the benefits and challenges of using graph structure data for machine learning analysis of disease-related omics data and provide directions for further research. [less ▲]

Ten Quick Tips for Biomarker Discovery and Validation Analyses Using Machine Learning

in PLoS Computational Biology (2022), 18(8), 1010357

High-throughput experimental methods for biosample profiling and growing collections of clinical and health record data provide ample opportunities for biomarker discovery and medical decision support ... [more ▼]

High-throughput experimental methods for biosample profiling and growing collections of clinical and health record data provide ample opportunities for biomarker discovery and medical decision support. However, many of the new data types, including single-cell omics and high-resolution cellular imaging data, also pose particular challenges for data analysis. A high dimensionality of the data in relation to small numbers of available samples, influences of additive and multiplicative noise, large numbers of uninformative or redundant data features, outliers, confounding factors and imbalanced sample group numbers are all common characteristics of current biomedical data collections. While first successes have been achieved in developing clinical decision support tools using multifactorial omics data, there is still an unmet need and great potential for earlier, more accurate and robust diagnostic and prognostic tools for many complex diseases. Here, we provide a set of broadly applicable tips to address some of the most common pitfalls and limitations for biomarker signature development, including supervised and unsupervised machine learning, feature selection and hypothesis testing approaches. In contrast to previous guidelines discussing detailed aspects of quality control, statistics or study reporting, we give a broader overview of the typical challenges and sort the quick tips to address them chronologically by the study phase (starting with study design, then covering consecutive phases of biomarker signature discovery and validation, see also the overview in Fig. 1). While these tips are not comprehensive, they are chosen to cover what we consider as the most frequent, significant, and practically relevant issues and risks in biomarker development. By pointing the reader to further relevant literature on the covered aspects of biomarker discovery and validation, we hope to provide an initial guideline and entry point into the more detailed technical and application-specific aspects of this field. [less ▲]