RefLaTEA: a robust visualization and analysis framework leveraging background data for enhanced insight.

Area of Science:

• Data science
• Bioinformatics
• Environmental science

Background:

• Large-scale observational data often contain natural variation that can obscure subtle experimental signals.
• Experimental datasets are typically clean but sparse, making it challenging to interpret findings within a broader context.

Purpose of the Study:

• To introduce Reference-based Lattice Transfer Embedding Analysis (RefLaTEA), a novel framework for contextualizing small experimental datasets within large-scale observational data.
• To provide a robust method for interpreting experimental responses relative to natural variation in fields like environmental science and omics.

Main Methods:

• Constructs a reference embedding (lattice) using Uniform Manifold Approximation and Projection (UMAP) on background data.
• Projects experimental datasets into the reference lattice via transfer embedding for contextual analysis.
• Offers optional extensions for clustering, feature importance, and causal inference.

Main Results:

• RefLaTEA effectively contextualizes experimental data within large observational datasets.
• The framework demonstrates robustness to UMAP parameter changes and interpretability for subtle responses.
• Integrates transfer embedding with downstream analyses for deeper mechanistic insights.

Conclusions:

• RefLaTEA bridges the gap between observational and experimental data, facilitating exploratory analysis and hypothesis generation.
• Provides a flexible and robust approach for environmental and omics research.
• Enhances the interpretation of experimental results by situating them within the landscape of natural variation.