LMdist

LMdist adjusts pairwise beta diversity measures to correct geometric distortions introduced by ordination methods and improve representation of ecological gradients and species-distribution variation.

Key Features:

• Unsupervised Algorithm: LMdist operates as an unsupervised algorithm that adjusts pairwise beta diversity measures without requiring labeled data.
• Graph Structure Projection: The tool employs a graph structure to project pairwise distances onto a manifold, enabling traversal across the manifold surface for adjustment.
• Dynamic Range Adjustment: By projecting and adjusting distances, LMdist extends beyond the original measure’s bounded dynamic range to more accurately represent ecological distances between samples.
• Selective Adjustment: LMdist selectively adjusts only oversaturated or undervalued pairwise distances in the presence of sampled gradients or manifold-like variation.
• Adjusted-Distance Output: LMdist produces adjusted pairwise distances as output for downstream ordination techniques and statistical testing.

Scientific Applications:

• Enhanced Ordination and Statistical Testing: The adjusted distances serve as input for ordination methods and statistical tests to improve detection of gradient effects.
• Recovery of Known Gradients: LMdist recovers distances along known ecological gradients and complex manifolds, such as the Swiss roll dataset, in real and simulated data.
• Revealing Orthogonal Variation: LMdist reveals variation orthogonal to environmental gradients that might otherwise remain obscured.
• Mitigation of Geometric Artifacts: LMdist mitigates geometric abnormalities such as the arch and horseshoe effects to provide a more accurate depiction of ecological relationships.

Methodology:

LMdist uses an unsupervised algorithm that projects pairwise beta diversity distances onto a graph-based manifold, traverses the manifold surface to adjust distances at the upper end of the dynamic range, extends values beyond the original bounded range, and selectively applies adjustments to oversaturated or undervalued pairwise distances.