Development of multi-block statistical learning approaches for non-targeted screening of water samples

• Applicant: Professorin Maryam Vosough, Ph.D.
• Subject Area: Analytical Chemistry
• Term: since 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 520243139

Key Objectives: (a) Systematically and statistically compare multiple LC-HRMS feature detection algorithms, including conventional peak-picking and ROI-MCR methods, using real spatiotemporal river datasets. This objective aims to assess their representativeness, complementarity, and fragmentation across chemical space, and to construct an integrated feature set using multi-block fusion techniques. (b) Design a novel sparse-based feature selection strategy, suitable for unsupervised learning and multi-block factor analysis settings, enabling the prioritization of significant chemical and biological markers across combined datasets. (c) Developing factor analysis-based Self-Organizing Maps (Kohonen Maps) as a robust AI tool for exploring and visualizing multiple layers of shared or distinct spatial/temporal patterns across multiple data sources and explore complex patterns in a low-dimensional space. (d) Develop a comprehensive multi-omics framework that integrates LC-HRMS, GC-MS, and physicochemical data with shotgun metagenomic sequencing to jointly characterize pollutant signatures and microbial community dynamics in riverine ecosystems. This includes analysis at three metagenomic levels (read, assembly, and MAGs) and implementation of supervised multi-block modeling approaches (e.g., PLS-based models, support vector machines, and multiple kernel learning) to identify ecologically relevant cross-domain patterns.

• DFG Programme: Research Grants