Selective pruning and neuronal death generate heavy-tail network connectivity

TL;DR

This paper introduces a selective pruning algorithm that models neuronal network development, showing how component loss combined with growth can produce stable, scale-invariant, and efficient network structures similar to biological brains.

Contribution

The study proposes a novel algorithm for network evolution through selective deletion, demonstrating its ability to generate scale-invariant degree distributions in neuronal networks.

Findings

Networks exhibit scale-invariance when predominantly feed-forward.

Selective pruning enhances network stability and efficiency.

Algorithm offers an alternative to preferential attachment for scale-free networks.

Abstract

From the proliferative mechanisms generating neurons from progenitor cells to neuron migration and synaptic connection formation, several vicissitudes culminate in the mature brain. Both component loss and gain remain ubiquitous during brain development. For example, rodent brains lose over half of their initial neurons and synapses during healthy development. The role of deleterious steps in network ontogeny remains unclear, yet it is unlikely these costly processes are random. Like neurogenesis and synaptogenesis, synaptic pruning and neuron death likely evolved to support complex, efficient computations. In order to incorporate both component loss and gain in describing neuronal networks, we propose an algorithm where a directed network evolves through the selective deletion of less-connected nodes (neurons) and edges (synapses). Resulting in networks that display scale-invariant degree distributions, provided the network is predominantly feed-forward. Scale-invariance offers several advantages in biological networks: scalability, resistance to random deletions, and strong connectivity with parsimonious wiring. Whilst our algorithm is not intended to be a realistic model of neuronal network formation, our results suggest selective deletion is an adaptive mechanism contributing to more stable and efficient networks. This process aligns with observed decreasing pruning rates in animal studies, resulting in higher synapse preservation. Our overall findings have broader implications for network science. Scale-invariance in degree distributions was demonstrated in growing preferential attachment networks and observed empirically. Our preferential detachment algorithm offers an alternative mechanism for generating such networks, suggesting that both mechanisms may be part of a broader class of algorithms resulting in scale-free networks.