Fly brain connectome

Context

There have been tremendous improvements in our ability to describe the identity of neurons, its input neurons (presynaptic partners), and output neurons (postsynaptic partners). The whole of the fly central brain has recently been digitized,¹ and the connectivity between these neurons, also called the connectome, has been mapped out for a single female fly, with more on the way (very exciting for a fly neuroscientist like myself).

The limitation to digitizing brains is technological, so one would expect obtaining the whole brain connectivity to get (exponentially) cheaper and faster. Therefore, I expect the whole brain connectivity of other animals to come up soon (maybe in the next 5-10 years). I feel that we are now heading into a new era of neuroscience very similar to the genomic boom in the 2000s.

As I begin working excitedly on the connectome data,² I had quite a few implicit assumptions on how to use the dataset that need to be kept in mind while using it to understand neural computation. This is a growing note for myself to refer to as I continue working on the connectome.

Assumptions

1. The computation performed by the neurons is captured by the synaptic strength between them

Validity:

This is a reasonable assumption and a good starting point, especially considering how successful artificial neural networks have been.  

Caveats:  

• Electrical synapses are not captured in the connectome.
    Electrical synapses allow for rapid signal transfer between neurons as current flows directly between them. Chemical synapses are typically slower, but they make it up by their diversity and the dimensions they add to the discrete spike (this needs more fleshing out³).

• Weights are dependent on neurotransmitters.
    Chemical synapses release different types of neurotransmitters, each of which has specific properties. The common ones, such as Acetylcholine and GABA, are excitatory and inhibitory, respectively, which can be visualized as +ve or -ve weights. However, there are many other types of transmitters that are dependent on the receptors present on the postsynaptic neuron (output neuron) and/or context-dependent, and cannot be simplified as the valence of weights.

• Weights depend on the location of the synapse.⁴
    Unlike ANNs, the effect of a synapse depends upon the structure of the neuron and the physical location of the synapse. Synapses closer to the axon can shunt signals or modify outputs effectively with a lower number of synapses. Given that the fly brain (maybe animal brain in general) has high numbers of axon-axonal synapses, dendrite-dendritic synapses, and axon-dendritic synapses, the location of the synapse should be kept in mind (this can be done with the current connectome data).

• Neuropeptides influence neural computation in unclear ways.
    In addition to neurotransmitters, neurons release neuropeptides that modify the properties of the downstream neurons. I am not clear about this myself and need to flesh this out.³]

Conclusion:

Synaptic weight, at least with the resolution currently provided by the connectome, does not capture a lot of the other dimensions that might be important for information transfer between neurons. While this is a reasonable starting point, I think I need to keep in mind that this is just one of the dimensions of connectivity, and a coarse measure at that.

Footnotes

1. A set of papers related to the connectome ↩

2. Happy to talk more about it (and/or collaborate); reach out if you are interested! ↩

3. #incomplete_idea ↩ ↩²

4. This can be gotten around by using the location of the synapse on the skeleton of the neuron - this data is present in the connectome. ↩