Models of the Brain by Nikki Salla

Neuroscientists know an extraordinary amount about the brain. Yet it doesn’t feel enough, and we don’t fully understand how any of it adds up. Think of it like trying to understand a complex city and picture it as overlapping transparencies which represent different kinds of information: where roads are, how fast traffic moves, where the cross walks are, which intersections are one-way, which neighborhoods come alive at night. Understanding fully any single transparency feels like progress but it doesn’t really all come together until they’re all overlaid. For the human brain, this complete picture doesn’t exist yet… not even close. We have patches, snapshots, and fragments of wiring reconstructed from small chunks of post-mortem brain tissue. We have efforts to atlas the cell types across the entire human brain to catalog the hundreds of distinct neuronal and non-neuronal identities that exist. We have whole brain imaging that captures which brain regions are active and talking to each other, but they lack the resolution to see individual neurons. We have some sense of how brain regions differ from one another, but these still represent an infinitesimal fraction of what a full understanding would require. I study one protein on one type of connection in one region of the mouse brain, and most days even that feels impossibly complex. So when I heard that a company had simulated an entire nervous system (every neuron and every connection between them) and watched it behave through a simulated animal, I had to read it twice. It turned out the road map and some directionality was enough to tell us about the city. 

I had no idea that anatomy alone could give us such useful information as to replicate behavior. Eon Systems, a San Francisco based public benefit corporation, took two existing but never-before-combined datasets about the fruit fly brain and used them to simulate a fly that actually behaves. 

To understand why that is remarkable, you need to know what went into it. 

The fruit fly connectome is made up of 125,000 neurons and 50 million contact points between neurons called synapses. This map came from the FlyWire consortium, a Princeton-led collaboration of dozens of labs from around the world, that reconstructed the fly brain’s wiring using physically sliced brain sections into thousands of ultrathin sections and imaged using electron microscopy. This technique is powerful enough to resolve individual synapses. AI algorithms processed those millions of images into a preliminary wiring diagram, but it wasn’t perfect, so hundreds of scientists around the world logged into an online platform to proofread and correct the map, neuron by neuron. The result, published in 2024, was the first complete wiring diagram of an adult fruit fly brain. This is an extraordinary contribution to the field of neuroscience because understanding how the architecture of any brain works helps us understand others. 

The next important resource for Eon Systems was a machine learning model that looked at those electron microscopy images and predicted whether each connection between neurons was excitatory or inhibitory. In other words, whether it tells a connected neuron to fire or stay quiet. This distinction is critical because without knowing whether each connection excites or silences its synaptic partner, a wiring diagram is just a map with no traffic information. Philip Shiu, Eon Systems’ senior scientist, then built a computational brain model on top of both and showed that connectome structure along with the predictions about excitatory and inhibitory connections, could predict motor behavior. That model became the foundation for everything Eon Systems built next. 

A brain model without a body is still just a simulation in isolation. Eon Systems took the next decisive step. They connected the fly brain model to NeuroMechFly, a physics-simulated fly body with 87 joints built from measurements of a real fruit fly, and placed it inside a virtual environment. 

For the first time, a biologically constructed brain could receive sensory information from the (virtual) world and use it to move a body. The brain wasn’t programmed to do this, and it wasn’t trained through trial and error the way a machine learning agent would be. The behaviors like navigating toward food, feeding when it detected sugar, grooming its body, emerged purely from the wiring. 

Their longer term goal is to emulate the human brain. I have so many questions! What does this mean for understanding the human brain, and is anatomy really enough? 

Neuroscientists are trying to build these maps. The BRAIN initiative has funded large-scale efforts to catalog cell types and connectivity across the mouse and human brains. The Allen Brain Atlas has produced detailed maps of gene expression (which genes are switched on or off) across brain regions. The Human Cell Atlas is working to catalog every cell type in the body, including the brain. However, each of these is a snapshot drawn from a handful of donors at a single point in time. Think of it as having photographs of a few city blocks taken on a single afternoon while trying to understand a city that has been changing over time. Even if we have the complete road map, the biological complexity of the human brain makes this a different order of challenge entirely. The human brain contains 86 billion neurons, compared to roughly 130,000 in the fly brain, with an estimated 100 trillion to one quadrillion synapses between them. And the raw numbers still do not capture the real complexity and they make up just one transparency of the full map. Plus, not all neurons are alike. There are hundreds of distinct subtypes, each with different electrical properties, different connectivity rules, and different roles in brain circuit functions. Neurons aren’t even the whole story. Glial cells, roughly as numerous as neurons, regulate how neurons communicate, remodel synaptic connections, and shape how circuits develop and respond to injury. Scientists are building the infrastructure to understand each of the maps layers. The current resources and work are extraordinary, but even these represent only a fraction of what a complete picture of the human brain would require, at a single point in time, let alone across the full arc of a human life. 

Eon Systems showed that with just a few of the transparencies for the fly brain, the wiring diagram, the excitatory or inhibitory identity, and simplified firing rules, all drawn from anatomical information, were enough to recapitulate fly behavior. 

But before we get to the human brain, it’s worth asking what we learn from doing this for the fruit fly. A working brain simulation can be a testable model. You can ask  what happens when you silence specific neurons or entire classes of neurons without touching a living animal. The fly brain simulation revealed something really important: that the large-scale architecture of a nervous system (its wiring and directionality of its connections) encodes more information about behavior than anyone expected. 

Scale this to the human brain and the implications become harder to ignore. Many devastating neurological and psychiatric disorders are not necessarily diseases of single genes or single neurons. They are diseases of circuits: subtle shifts in connectivity and in the balance of excitation and inhibition, and how signals propagate across regions. We currently have no way to test circuit-level hypotheses in a living human brain. Animal models help but they don’t fully recapitulate human cognition or the disorders that affect it. A human brain, even an imperfect one, would offer something new: a testable model of how circuit level changes produce behavioral and cognitive symptoms and a platform for identifying which circuits to target for treatment. It will not replace current animal research or clinical trials. But it could help us figure out which questions we think to ask. 

The fly brain simulation is scientifically remarkable and philosophically neutral. A simulated fly navigating toward sugar raises no uncomfortable questions about what it means to be that fly. However a simulated human brain is a different matter. If Eon Systems achieves their goal, the initial questions are not just technical ones but more philosophical ones. Such as: whose brain? At what age? Are experiences encoded in a brain map? The human brain is not a static object. It is an accumulated product of every experience, every relationship, and every loss and decision across a lifetime. The wiring of a five year old’s brain is not the wiring of a sixty year old’s brain. A brain shaped by trauma is not wired like one that wasn’t. But what we still don’t know is if mapping a human brain means mapping a specific person’s brain or if that results in the template for human behavior. 

The fly simulation produces behavior that looks like a fly, but is anything happening inside it that resembles experience? Consciousness remains one of the most contested and least understood problems in science. We don’t know whether it emerges from connectivity alone or whether it requires something else like the biochemistry of living cells, the continuous dynamic of a brain embedded in a body, or something we haven’t named yet. “Uploading” the human brain might help us answer a lot about how cognition actually works and might hint at what consciousness is. The fly simulation offers exciting clues. Eon Systems didn’t need every detail of the fly nervous system to generate behavior. They needed wiring, directionality, and a simplified firing rule. The rest of the nuance was not necessary for general fly behavior. This raises another important question: what is the minimum information required to generate human behavior? Is it the connectome? The connectome plus cell type identity? Does it require information happening within each cell and also how they all communicate dynamically? We don’t know yet, but I’m curious to find out what we learn. 

I came into neuroscience because I wanted to understand how the brain builds itself and most days my small corner of it feels like too much to hold. Every generation of neuroscientists has had a tool that changed what questions were even askable. The microscope gave us the neuron. The electrode gave us the action potential. Sequencing gave us the molecular identity of individual cells. The connectome, and what Eon Systems has now done with it, might be this generation’s version of that shift. We are not close to understanding the human brain, but we are asking better questions. In science, that is not a small thing. 

It seems that a few transparencies of a brain can tell us more than any of us expected.