Analysis of Evolving Cortical Neuronal Networks Using Visual Informatics

Understanding how neuronal population activity changes during development and after stimulation is essential for studying neuronal network dynamics.

This work examines how visual informatics can summarize high-dimensional spiking activity while retaining information that is biologically interpretable.

We develop a framework based on Minimum-Distortion Embedding (MDE), and compare it with Principal Component Analysis (PCA) and t-distributed Stochastic Neighbor Embedding (t-SNE).

In addition to evaluating the embeddings by visual separation, we quantify whether they preserve the cosine-shape radius within each condition and the pairwise distances between condition centroids.

Our \emph{in silico} experiments show that MDE with a cosine metric captures the trajectory of simulated network maturation and preserves the contraction of the activity cloud as connectivity increases.

Complementary \emph{in vitro} experiments on human cortical cultures show a coherent developmental trajectory from Day In VITRO 23 (DIV23) to DIV64.

We also study weak and strong stimulation in simulation, and long-term potentiation stimulation in primary cortical cultures.

In the stimulation experiments, MDE separates activity phases more clearly than PCA and preserves transient changes in within-phase variability that are missed by PCA.

These results show that metric selection is central to dimensionality reduction of neuronal data.

In particular, cosine distance between population activity vectors provides embeddings that better reflect changes in population activity patterns than Euclidean distance.

The proposed framework provides a quantitative way to visualize network development and stimulation-induced changes in neuronal activity.