Insights into the Effects of Violating Statistical Assumptions for Dimensionality Reduction for Chemical "-omics"...

Area of Science:

• Metabolomics
• Cheminformatics
• Ecology

Background:

• Biological volatilome analysis presents challenges due to high dimensionality and wide variations in compound peak areas.
• Current dimensionality reduction methods often assume data normality and linearity, which are frequently violated in biological datasets.
• Log transformations can address normality deviations but require careful consideration of variable effects (additive vs. multiplicative).

Purpose of the Study:

• To assess the impact of statistical models (single vs. multivariable) and log transformation on volatilome dimensionality reduction.
• To evaluate how these factors influence downstream supervised and unsupervised classification analyses.
• To determine the most robust method for volatilome dimensionality reduction, particularly for complex biological samples.

Main Methods:

• Volatilome data from Shingleback lizards (Tiliqua rugosa) were collected across natural habitats and captivity.
• Analysis involved single and multivariable statistical models, with and without log transformation.
• Monte Carlo tests were employed for dimensionality reduction on both transformed and untransformed data.

Main Results:

• Excluding multiple explanatory variables led to overestimation of habitat (Bioregion) effects and inaccurate compound identification.
• Log transformation and normality assumption increased the number of significant compounds identified.
• Analyzing untransformed data with multivariable Monte Carlo tests yielded the most conservative dimensionality reduction.

Conclusions:

• Accurate volatilome analysis requires accounting for multiple explanatory variables and carefully considering data transformations.
• The chosen dimensionality reduction method significantly impacts the identification of biologically relevant compounds.
• Multivariable analysis of untransformed data offers a more reliable approach for complex biological volatilomes.