Figure 6– Gradient Stability Across Estimation Methods:

A. Scatterplot between gradient 1 estimated with a Pearson correlation similarity matrix and diffusion mapping dimensionality reduction (the original approach used in this study), and gradient 1 estimated with a Pearson correlation similarity matrix and principal component analysis dimensionality (PCA) dimensionality reduction. B. Same as A. but with gradient 2. C-D. Absolute value of the Pearson correlation between gradient 1 (C.) and gradient 2 (D.) for different methods of estimating the similarity matrix and the subsequent dimensionality reduction. Arrows point to the rows and columns corresponding to the Pearson-DM approach used in the main findings of this study. DM – Diffusion mapping, LE – Laplacian embedding, PCA – Principal component analysis.