Goals

Goals

General Goals

• Bring the community together and learn the state-of-the-art in the analysis of connectomes

• Get a better understanding of how connectomes will be used and the types of analyses and infrastructure needed

• Feedback on DVID API improvements (our hope is to better define API and infrastructure for reconstruction and analysis between teams, so that research effort is better focused)

Specific Problems

The goal is to get some discussion and preferably code to help address some of the following challenges. We hope to provide links to (preliminary) open-source solutions over the course of the next several months.

• Determine interesting and reoccurring connectivity motifs. E.g, can the columnar neurons be automatically and correctly clustered into the seven columns by analyzing the connectome?.

• Determine outliers in connectivity. This could be useful both for understanding how a circuit works and for determining whether there are proofreading errors in the reconstruction. For instance, in our dataset, we observe that some neurons have autapses (they have synapses that connect to themselves). Algorithms that can detect whether certain neurons have similar outlier features could be useful.

• We are fortunate in this reconstruction in that many of the neuron types have been determined by light data and Golgi studies and we can map our neuron shapes to this pre-existing library. It should be possible to determine different classes of neurons by their connectivity pattern. This would be useful in reconstructions where the neuron type is not easily identifiable morphologically or where there is no pre-existing library. An interesting problem would be whether neuron types like Mi1 could be automatically determined by examining its connectivity alone.

• Visualization. How to best visualize a connectome? For instance, showing a graph where node locations are given by the center of mass might be inferior to showing the node plotted where most of its connections are, depending on the application. There needs to be flexible ways to manipulate visualizations to inspire analysis.

• Biological insight. How can we use a connectome to constrain the possible set of functions of the underlying circuits? This problem is exacerbated by the absence of knowledge about the signs of the connections (see following point). Perhaps the use of simulations of information flow through a connectome might be able to provide a space of possible functions for the circuitry.

• In the fly, the structural connectome does not provide any insight onto the transmitter of the synapse, i.e., is a synapse inhibitory or excitatory. (To people familiar with the vertebrate cortex, this is different from that system where excitatory and inhibitory synapses are structurally different). Nevertheless, it might still be possible to determine by information flow or alternative metrics whether the sign of a given connection is mostly likely to be positive or negative. Alternatively, could the determination of just a few of these signs (by separately sequencing the RNA of those cell types) help determine the signs within the rest of the network? Even crude hypotheses could be used to guide hypotheses of the function of the underlying circuitry.

• Given a set of hypotheses for the function for the circuits underlying the connectome, is it possible to identify experiments that would be particularly informative about the function of specific components. Given that there are only a limited set of manipulations possible, within a circuit (e.g. through the selective stimulation or ablation of a given neuron), identifying which manipulations to perform (and what input regime the manipulations should show the largest effect) would be extremely helpful in limiting the number of experiments that might need to be performed.

• Creation of standard libraries for the simulation of neuronal activity within specific circuit motifs. Through connectomics efforts, such motifs might be automatically identified in other neuropils. In that case, these simulation models might be useful in deriving functional hypotheses for small subcircuits within the connectome.