Detecting signatures underlying the composition of biological data.

Area of Science:

• Microbiology
• Bioinformatics
• Systems Biology

Background:

• Biological compositional data is multidimensional, posing challenges for visualization and interpretation.
• Identifying co-occurring biological features, termed signatures, is crucial for understanding complex systems.
• Existing methods struggle with automatic decomposition and gradient capture in large datasets.

Purpose of the Study:

• To develop a software package, cvaNMF, for automatic decomposition of large compositional biological data.
• To capture gradients in co-occurring features (signatures) for enhanced data interpretation.
• To provide tools for identifying and visualizing biologically informative signatures.

Main Methods:

• Non-negative matrix factorization (NMF) for data decomposition.
• Cross-validation and a novel signature-similarity method for selecting optimal decomposition.
• Application of cvaNMF to diverse datasets including metagenomes and cellular composition data.

Main Results:

• cvaNMF effectively identifies informative signatures in synthetic and real-world biological data.
• 'Enterosignatures' from gut metagenomes differentiated human disease states.
• 'Terrasignatures' from rhizosphere metagenomes distinguished plant microbiomes and inferred geographic distances.
• Analysis of >13,000 metagenomes revealed environmental and host-associated microbiome signatures.
• Cell-type signatures separated cancerous from inflamed non-small cell lung cancer tissues.

Conclusions:

• The cvaNMF software package offers a robust solution for decomposing and interpreting complex biological compositional data.
• Identified signatures provide novel insights into host-microbe interactions, ecological relationships, and disease states.
• cvaNMF facilitates the discovery of biologically relevant patterns across various scales and sample types.