Scientists have just published in Nature that they have completed the entire connectome of a fruit fly: Network statistics of the whole-brain connectome of Drosophila. The map includes 140,000 neurons and more than 50 million connections. This is an incredible achievement that marks a milestone in neuroscience and is likely to advance our research.

A “connectome” is a complete map of all the neurons and all the connections in a brain. The ultimate goal is to map the entire human brain, which has 86 billion neurons and about 100 trillion connections – that’s more than six orders of magnitude greater than the drosophila. The human genome project was started in 2009 through the NIH, and today there are several efforts contributing to this goal.

Right now we have what is called a mesoscale connectome of the human brain. This is more detailed than a macroscopic map of human brain anatomy, but not as detailed as a microscopic map at the neuronal and synapse level. It’s in between, so mesoscale. Essentially we have built a mesoscale map of the human brain from functional MRI and similar data, showing brain regions and types of neurons at the millimeter scale and their connections. We also have mesoscale connectomes of other mammalian brains. These are highly useful, but the more detail we have obviously the better for research.

We can mark progress on developing connectomes in a number of ways – how is the technology improving, how much detail do we have on the human brain, and how complex is the most complex brain we have fully mapped. That last one just got its first entry – the fruit fly or drosophila brain.

The Nature paper doesn’t just say – here’s the Drosophila brain. It does some interesting statistics on the connectome, showing the utility of having one. The ultimate goal is to fully understand how brains process information. Learning such principles (which we already have a pretty good idea of) can be applied to other brains, including humans. For example, the study finds that the Drosophila brain has hubs and networks, which vary in terms of their robustness. It also reflects what is known as rich-hub organization.

Rich-hub organization means that there are hubs of neurons that have lots of connections, and these hubs have lots of connections to other hubs. This structure allows brains to efficiently integrate and disseminate information. This follows the same principle as with any distribution system. Even Amazon follows a similar model, with distribution centers serving as hubs. Further, the researchers identified specific subsets of the hubs that serve as integrators of information and other subsets that serve as broadcasters.

The connectome also includes synapse and neurotransmitter level data, which is critical to asking any questions about function. A connectome is not just a map of wiring. Different neurons use different neurotransmitters, which have different functions. Some neurotransmitters, for example, are excitatory, which means they increase the firing rate of the neuron in which they synapse. Some neurotransmitters are inhibitory, which means they decrease firing rate. So at the very least we need to know if a connection is increasing or decreasing the activity of the neurons it connects to.

Now that the model is complete, they are just getting started examining the model. This is the kind of research that is primarily meant to facilitate other research, so expects lot of papers using the Drosophila connectome as its subject.

Meanwhile scientists are working on completing the connectome of the mouse, which will likely be the first mammalian brain connectome. We already have mesoscale connectomes, and detailed connectomes of small sections of mouse brain. A completed mouse brain connectome is likely 10-15 years off (but of course, could be longer). That would be a huge milestone, as all mammalian brains share a lot of anatomy in common. With the Drosophila brain we can learn a lot about network principles, but the anatomy evolved completely independently from mammals (beyond the very rudimentary brain of our common ancestor).

One type of research that I would love to see is not just mapping a connectome, but emulating it in a computer. This information may be out there somewhere, but I have not found it so far – do we have a computer powerful enough to emulate the functioning of a Drosophila brain in real time? That would be a good test of the completeness and accuracy of our connectome – does it behave like an actual fruit  fly?

Creating this would likely require more than just the connectome itself. We need, as I referenced above, some biological data as well. We need to know how the neurons are behaving biologically, not just as wires. We need to know how the neurotransmitters are behaving chemically. And we need to know how other cells in the brain, other than neurons, are affecting neuronal function. Then we need to give this virtual brain some input simulating a body and an environment, and simulate the environment’s response to the virtual fruit fly. That sounds like a lot of computing power, and I wonder how it compares to our current supercomputers. Likely we will be able to do this before we can do it in real time, meaning that a second of the life of our virtual Drosophila may take a day to compute (that is just a representative figure, I have no idea what the real current answer is). Then over time, our virtual Drosophila will go faster and faster until it catches up to real time.

Eventually the same will be true for a human. At some point we will have a full human connectome. Then we will be able to emulate in a computer, but very slowly. Eventually it will catch up to real time, but why would it stop there? We may eventually have a computer that can simulate a human’s thought processes 1000 times faster than a human.

There is another wrinkle to this whole story – the role of our current and likely short term future AI. We are already using AI as a tool to help us make sense of the mesoscale connectomes we have. Our predictions of how long it will take to have complete connectomes may be way off. What if someone figures out a way to use AI to predict neuron level connectomes from our current mesoscale connectomes? We are already seeing, in many contexts, AI being used to do literally years of research in days or weeks, or months of research in hours. This is especially true for information-heavy research questions involving highly complex systems – exactly like the connectome. It would therefore not surprise me at all if AI-boosted connectome research suddenly progresses orders of magnitude faster than previous predictions.

Another potential area of advance is using AI to figure out ways to emulate a mammalian or even human brain more efficiently. We don’t necessarily need to emulate every function of an entire brain. We can probably cheat our way to make simple approximations of the functions we are not interested in for any particular emulation or research project. Then dedicate the computing power to what we are interested in, such as higher level decision-making.

And of course I have to mention the ethical considerations of all of this? Would a high fidelity emulation of a human brain be a human? I think the answer is either yes, or very close to yes. This means we have to consider the rights of the emulated human. For this reason it actually may be more useful to emulate a mouse brain. We already have worked out ethical considerations for doing mouse research, and this would be an extension of that. I can imagine a future where we do lots of behavioral research on virtual mice in simulated environments. We could run millions of trials in a few minutes, without having to care for living creatures. We can then work our way evolutionarily toward humans. How far will we go? Would virtual primate research be OK? Can we guarantee our virtual models don’t “suffer”. Does it matter that they “exist” for just a fraction of a second? We’ll have to sort all this out eventually.