Efficient Connectome Analyses for Identifying Bottleneck Neurons in Behaviorally-Relevant Pathways
Ishani Ganguly, Columbia University
The availability and scale of synapse-resolution connectomes containing the wiring of neural systems has grown rapidly in recent years. In order to exploit these rich datasets and extract neural circuit mechanisms underlying behavior, scalable connectome analysis tools capable of predicting information routing principles in circuit pathways are necessary. Current methods to extract neurons mediating information flow only consider a limited set of pathway lengths and do not scale to large connectome datasets. Here, we develop an efficient framework to identify such bottlenecks in pathways between arbitrary source and target neurons. We define an influence metric that quantifies the strength of pathways of arbitrary length from sources to targets and develop an algorithm that extracts the smallest group of neurons that account for a specified proportion of this total influence, which we term “bottleneck neurons." To validate our methods, we deploy them on the Drosophila hemibrain connectome to recover several well-studied neural pathways from sensory input to descending motor output. By varying a gain parameter that penalizes longer neural pathways when computing influence, we can prioritize neurons that are involved in longer or shorter pathways in these ground truth circuits. We then apply our framework to explore the subset of bottleneck neurons that mediate information flow from the sensory periphery to the functionally and anatomically distinct compartments of the mushroom body, the primary learning center in the insect brain. Extending our methods to other regions may elucidate which neurons are best suited for experimental manipulation and shed light on large-scale organizational principles of neural circuits.