Search

The fourth edition of the Computational Metabolomics seminars, Dagstuhl Seminar 22181, brought together a wide range of computational and experimental experts to share state-of-the art methodologies and push our collective understanding of how to interpret and maximise insight of metabolomic data. With increasing amounts of metabolomic data being generated, including large-scale epidemiological studies, and increasing sensitivity of instrumentation, development of sophisticated and robust computational solutions is required. Further, community agreement on which data standards should be used and which data sets are most apt for benchmarking computational tools is needed in the field. Building upon the previous successful formats of previous seminars (17491, 15492, and 20051) on this topic, attendees gathered each morning to collectively agree on the number of sessions and topics to discuss. A summary of the daily sessions were shared amongst all participants after dinner during each day’s final formal session. Further, informal evening sessions were spontaneously created to further dive into specific topics. As with past seminars, this format was very well received and enabled all participants to weigh in. Of particular note, this seminar was delayed and travel was complicated due to the pandemic. Despite these setbacks, this seminar brought together a balanced number of previous and new, seasoned and early career participants. All participants were active in these discussions, and a true sense of renewed energy ensued from the seminar. This report provides highlights of formal and informal evening sessions, including future anticipated research directions rooted from this seminar. Possible future workshops, such as a next phase of this Computational Metabolomics Dagstuhl seminar in late 2023 or 2024 were also discussed and will be applied for., DagRep, Volume 12, Issue 5, pages 1-16

Additional file 1: Table S1. Method implementation details. Fig. S1. Overview of novel methods ssClustPA and kPCA for a single pathway example using simulated data based on the COVID dataset. Fig. S2. Performance of ssPA methods using the full pathway set (including redundancy) based on 3 randomly enriched pathways, derived using semi-synthetic data based on the IBD dataset. Fig. S3. Performance of ssPA methods using the full KEGG human pathway set (including redundancy) based on 3 randomly enriched pathways, derived using semi-synthetic data based on the COVID dataset. Fig. S4. ssPA method ability to rank highly the 3 randomly enriched pathways using the full set of Reactome pathways. Fig. S5. ssPA method performance in response to varying levels of pathway coverage. Fig. S6. ssPA method performance in response to varying levels of pathway coverage. Fig. S7. PCA scatter plots and density plots of PC1 scores obtained using the IBD data at the metabolite (upper panels) and pathway level using kPCA (lower panels). Fig. S8. Clustered heatmap of IBD data transformed to pathway scores using the kPCA method. Fig. S9. Pathway clusters derived using hierarchical clustering on COVID dataset transformed to pathway scores using kPCA (top 30 pathways). Table S2. Runtimes of ssPA methods, alongside GSEA* for comparison to conventional PA methods (average across 10 iterations). Table S3. Over-representation analysis results from IBD dataset. Table S4. Top 50 features in random forest model based on IBD data.

As terabytes of multi-omics data are being generated, there is an ever-increasing need for methods facilitating the integration and interpretation of such data. Current multi-omics integration methods typically output lists, clusters, or subnetworks of molecules related to an outcome. Even with expert domain knowledge, discerning the biological processes involved is a time-consuming activity. Here we propose PathIntegrate, a method for integrating multi-omics datasets based on pathways, designed to exploit knowledge of biological systems and thus provide interpretable models for such studies. PathIntegrate employs single-sample pathway analysis to transform multi-omics datasets from the molecular to the pathway-level, and applies a predictive single-view or multi-view model to integrate the data. Model outputs include multi-omics pathways ranked by their contribution to the outcome prediction, the contribution of each omics layer, and the importance of each molecule in a pathway. Using semi-synthetic data we demonstrate the benefit of grouping molecules into pathways to detect signals in low signal-to-noise scenarios, as well as the ability of PathIntegrate to precisely identify important pathways at low effect sizes. Finally, using COPD and COVID-19 data we showcase how PathIntegrate enables convenient integration and interpretation of complex high-dimensional multi-omics datasets. The PathIntegrate Python package is available at https://github.com/cwieder/PathIntegrate.