minimum-distortion-embedding-neuronal

name	minimum-distortion-embedding-neuronal
description	Minimum-Distortion Embedding (MDE) framework for analyzing evolving neuronal network dynamics. Use when dimensionality-reducing high-dimensional spiking activity, analyzing network development trajectories, or comparing stimulation effects in neuronal cultures.

Minimum-Distortion Embedding for Neuronal Analysis

Context

MDE with cosine metric captures neuronal network maturation trajectories and preserves activity cloud contraction as connectivity increases — outperforming PCA and t-SNE for neuronal population analysis.

Key Findings (arXiv:2502.20862)

• Cosine > Euclidean: Cosine distance between population activity vectors better reflects changes in population activity patterns
• Developmental tracking: MDE captures trajectory from DIV23 to DIV64 in human cortical cultures
• Stimulation detection: MDE separates activity phases more clearly than PCA
• Transient preservation: Preserves transient within-phase variability changes missed by PCA

Framework Steps

1. Extract population activity vectors from spiking data
2. Apply MDE with cosine metric (not Euclidean)
3. Quantify cosine-shape radius preservation within conditions
4. Measure pairwise distances between condition centroids
5. Compare with PCA/t-SNE baselines
6. Validate on both in silico and in vitro data

Metric Selection Guide

• Cosine distance: Best for population activity pattern changes
• Euclidean distance: May miss important within-phase variability
• MDE: Superior to PCA for stimulation experiments
• t-SNE: Good for visualization but less quantitative

Pitfalls

• Metric mismatch: Using Euclidean distance may obscure biologically relevant patterns
• Developmental scale: DIV23-DIV64 represents significant network maturation window
• Stimulation types: Weak vs strong stimulation show different embedding responses

Activation: neuronal embedding, MDE, dimensionality reduction, cortical cultures, stimulation analysis, network development