Nervous System Emulation with Synaptomics

 

The seven things that matter—and the traps that wreck most simulations

The big idea: Connectomics tells you which neurons are connected. Synaptomics tells you what each connection actually does. A nervous-system emulator needs both, but synapses should be treated as active computational components—not lines drawn between neurons.

1. The synapse is the real computational unit

A neuron is not simply “on” or “off,” and a connection is not merely a weight. Each synapse should be modeled as a stateful object:

Synapse = source + target + type + strength + delay + kinetics + plasticity + current state

Synapses differ enormously in molecular composition and functional behavior—even synapses belonging to the same neuron.

Rule: Don’t build a brain-shaped spreadsheet of neurons and weights.

---

2. Wiring is necessary—but nowhere near sufficient

A connectome can show hundreds of thousands of synapses, yet it cannot automatically tell you how the circuit will behave. The complete larval Drosophila connectome contains 3,016 neurons and roughly 548,000 synapses, but function still depends on synapse properties, cell dynamics and operating conditions.

Two identical-looking connections may transmit signals differently because of receptor type, release probability, dendritic location or previous activity.

Connectome = road map. Synaptome = vehicles, traffic lights and road conditions.

---

3. Every edge needs a biological identity

“Attractive” and “inhibitory” are not enough. At minimum, record:

• Neurotransmitter and receptor family
• Excitatory, inhibitory or modulatory action
• Conductance or effective strength
• Rise and decay time
• Axonal transmission delay
• Release probability
• Short- and long-term plasticity rules
• Location on the target neuron

Synaptic dynamics vary across cortical layers, cell classes and species, meaning one generic synapse model will erase important circuit behavior.

---

4. Time is part of the architecture

Brains compute using when, not just how much.

A useful synaptic update resembles:

$$I_{ij}(t)=g_{ij}(t)\,s_{ij}(t)\,[V_i(t)-E_{ij}]$$

Here, the current depends on a changing conductance, receptor state, membrane voltage and reversal potential. Add transmission delays, refractory periods, adaptation, facilitation and depression.

Rule: If every update happens simultaneously in fixed frames, you are emulating a spreadsheet—not a nervous system.

---

5. Learning is synaptic state change, not weight optimization

Biological learning can alter:

• Synaptic strength
• Release probability
• Receptor population
• Spine structure
• Connection existence
• Plasticity thresholds
• How future plasticity occurs

Plasticity is often controlled by local calcium activity plus wider neuromodulatory signals. This creates a practical three-factor rule:

$$\Delta w = \text{presynaptic activity} \times \text{postsynaptic response} \times \text{modulatory signal}$$

That third factor can represent reward, novelty, pain, attention or internal need.

---

6. Neuromodulators are operating modes

Dopamine, serotonin, acetylcholine and other modulators should not be treated as ordinary messages traveling from A to B. They can change how entire circuits interpret messages by adjusting:

• Neuronal excitability
• Synaptic gain
• Plasticity eligibility
• Sensory priority
• Action selection
• Exploration versus exploitation

The same wiring can therefore produce completely different behavior under different modulatory states.

Think software modes: seek food, escape, rest, learn and protect injury.

---

7. A nervous system must close the loop

A convincing emulator is not judged by how many neurons it contains. It is judged by whether it produces stable, adaptive behavior:

$$\text{Environment} \rightarrow \text{Sensors} \rightarrow \text{Synaptic network} \rightarrow \text{Motor output} \rightarrow \text{Environment}$$

Test it at four levels:

Synapse: Does transmission match expected timing and plasticity?
Circuit: Do oscillations, competition and routing emerge?
Behavior: Does the agent reach goals and recover from disturbances?
Adaptation: Does experience improve later performance?

A smaller closed-loop emulation with realistic synapses is usually more scientifically useful than a gigantic static network that cannot sense, act or learn.

---

The one sentence to remember

Nervous-system emulation is not about reproducing every neuron—it is about reproducing the changing rules by which synapses transform experience into action.