Identifying temporal and spatial patterns of variation from multimodal data using MEFISTO.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Factor analysis is crucial for dimensionality reduction in genome biology, aiding personalized health and single-cell studies.
• Current factor analysis models often assume sample independence, which is inadequate for spatio-temporal profiling.
• Spatio-temporal dependencies are common in biological data, necessitating advanced modeling techniques.

Purpose of the Study:

• Introduce MEFISTO, a versatile toolbox for modeling high-dimensional data with known spatial or temporal dependencies.
• Enable spatio-temporally informed dimensionality reduction, interpolation, and pattern separation.
• Facilitate integration of multiple related datasets by aligning underlying variation patterns.

Main Methods:

• MEFISTO employs a flexible factor analysis framework.
• The toolbox incorporates spatial or temporal dependencies into the modeling process.
• It allows for simultaneous identification and alignment of patterns across multiple datasets.

Main Results:

• MEFISTO successfully performs dimensionality reduction on spatio-temporally resolved data.
• The toolbox enables interpolation and separation of smooth from non-smooth variation patterns.
• MEFISTO effectively integrates multiple datasets, revealing underlying biological patterns.

Conclusions:

• MEFISTO offers a powerful solution for analyzing high-dimensional biological data with spatio-temporal structures.
• The toolbox enhances understanding of complex biological processes by accounting for sample dependencies.
• MEFISTO's applications span diverse fields, including developmental biology, microbiome studies, and single-cell multi-omics.