What Is Computational Neurobiology and How Does It Bridge Brain Levels?

Computational neurobiology is the use of mathematical models, simulations, and data-analysis methods to explain how the brain works across multiple scales — from the biophysical properties of synapses up to the behavior of whole neural systems. The Computational Neurobiology Laboratory (CNL), also known as the Sejnowski Lab, describes its long-range goal as building bridges between these brain levels by combining experimental and computational approaches. This approach is most useful when a question spans more than one level of description: for example, when you want to connect what happens at a synapse to how a circuit produces behavior, or how learning and sleep modify that circuit over time.

How computational neurobiology differs from experimental neuroscience

Experimental neuroscience typically measures or manipulates biological tissue and reports what is observed. Computational neurobiology takes those observations and asks what mechanism could produce them, then tests the mechanism in a model.

The two are not separate tracks at CNL — the lab states that its studies combine experimental and computational approaches. The computational side does three jobs:

• Formalizing a hypothesis so its consequences can be worked out precisely.
• Simulating processes that are hard to observe directly, such as synaptic signaling.
• Analyzing large or noisy recordings to find structure that isn't visible by eye.

The levels the lab tries to bridge

CNL frames its work as connecting levels rather than staying at one. The central issues it addresses are:

The "bridging" idea matters because a finding at one level often doesn't explain the next. Knowing that a synapse strengthens, for instance, doesn't by itself tell you how a circuit's output changes, or how that change shows up in behavior or during sleep.

Key methods and what they are for

CNL names two specific computational techniques in its own description:

Monte Carlo methods (MCell)

Monte Carlo simulation is used to explore synapses. Instead of solving every reaction analytically, the method simulates many individual molecular events and tracks the resulting behavior statistically. MCell is the specific tool named by the lab for this purpose. The practical value is that it can represent the messy, spatially detailed environment of a synapse where molecules diffuse and bind rather than behaving as a smooth average.

Independent component analysis (ICA)

ICA is used to analyze brain activity. It separates a mixed signal into components that are statistically independent of one another — useful when recordings contain overlapping contributions from many sources and you want to isolate them. This is the kind of method that turns a raw recording into something interpretable.

Example research questions

The lab's stated central issues give concrete examples of what "bridging levels" looks like in practice:

• Dendritic integration — how a single neuron combines the synaptic signals arriving on its dendrites.
• Neural circuits and behavior — how the activity of connected neurons produces an action or response.
• Learning and sleep — how these two processes adaptively modify circuits, linking a behavioral state to lasting changes in the network.

A specific thread running through the lab's work is fast-spiking parvalbumin-positive interneurons, studied both computationally and experimentally in two contexts: attention in the visual cortex and dysfunction in schizophrenia. This is a good illustration of the bridging idea — the same cell type is examined at the circuit level (attention) and connected to a clinical condition (schizophrenia), with computation and experiment used together rather than in isolation.

How computational models complement experimental work

Models don't replace experiments; they extend what experiments can conclude. In the attention and schizophrenia work, a computational model can propose how interneuron behavior affects circuit dynamics, and an experiment can test whether the predicted effect appears in tissue. Each side constrains the other.

The same pattern applies to the synapse work: Monte Carlo simulation of synaptic signaling generates predictions that experimental measurement can check, and the measurements feed back into the model.

Where to go next

CNL publishes its work and resources openly on its site, including a publications list with PDFs, a curriculum vitae, and links to events and external resources. If you want to follow the lab's output or see the methods applied in specific papers, the publications page is the direct entry point. The lab also points to broader resources, including its MOOC and book material on learning, which sit alongside the core research rather than replacing it.