PHASOR: An Open-Source Pipeline for Hidden Markov Model Analysis of Stimulus-Locked Retinal Multielectrode-Array Population Activity
The paper introduces PHASOR, an open-source pipeline that integrates stimulus alignment, Fourier-based classification, and Hidden Markov Modeling to analyze and visualize stimulus-locked population activity in retinal multielectrode-array recordings, demonstrating its effectiveness through synthetic data and diverse ex vivo retinal experiments.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
The retina is often thought of as a simple camera lens for the brain, but it is actually a complex neural circuit that processes visual information before sending it on. Inside this thin layer of tissue, thousands of individual neurons, called retinal ganglion cells, fire electrical signals in response to light. While scientists have long studied how single cells react to light, a growing understanding is that the true story of vision lies in how these cells work together as a group. Just as a single instrument in an orchestra cannot convey a symphony, a single neuron cannot fully explain how the brain sees the world. The challenge for researchers has been to listen to the chatter of hundreds of these cells at once and find the recurring patterns in their collective behavior, especially when those patterns are tied to specific moments in time, like the flicker of a light.
To solve this, a team of researchers has developed a new software tool called PHASOR, designed to listen to the retina's group conversation and translate it into a clear story. The researchers recorded the electrical activity of retinal cells from mice, both healthy ones and those with a genetic condition that damages their light-sensing cells. They exposed these retinas to a rhythmic, full-field flicker of light, varying the brightness from dim to very bright. The goal was to see if the group of cells would settle into distinct, repeating modes of activity that corresponded to the timing and intensity of the light, rather than just firing randomly.
The process begins by lining up every electrical spike from the cells with the exact moment the light changed. The software then sorts the cells based on how they react to the rhythm: some fire when the light turns on, some when it turns off, and others have more complex responses. Once the cells are categorized, the software uses a statistical method to group the entire population into hidden states. Imagine the population of cells as a large crowd; at any given second, the crowd might be in a state where mostly one type of person is talking, or another state where a different mix is active. The software tries to figure out how many of these distinct crowd states exist and which one the retina is in at every moment.
When tested on computer-generated data where the researchers knew the exact number of states beforehand, the tool proved remarkably accurate. Even when the software chose to describe the data with slightly more states than actually existed, it still managed to reconstruct the sequence of events with near-perfect fidelity. In these simulations, the tool correctly identified the timing and order of the hidden states almost every time, showing that it could recover the underlying structure of the data even if the exact count of states was slightly off. This distinction is crucial: the tool is designed to find the pattern of activity, not necessarily to count the states with mathematical precision.
When applied to real recordings from healthy mouse retinas, the tool revealed a rich tapestry of activity. The retinas did not just fire randomly; they shifted between distinct states that lasted for tens to hundreds of milliseconds. These states were not uniform; some were dominated by cells that responded to light turning on, while others were dominated by cells that responded to light turning off. The tool showed that as the contrast of the light increased, the population shifted into new states, and the specific mix of active cells changed in a predictable way. For instance, at higher contrast levels, the population spent more time in states where the "on" cells were highly active, and the transitions between these states were tightly locked to the rhythm of the flickering light.
To ensure these patterns were truly caused by the light and not just random noise, the researchers ran two control experiments. In one, they ran the same timing protocol but with no light projected onto the retina. In this case, the cells fired at a steady, low rate, and the software found no organized states that aligned with the timing of the protocol. In the second control, they used a retina from a mouse with a genetic condition that destroys the light-sensing cells. Even though the light was projected, the cells did not organize into the rhythmic patterns seen in healthy eyes. These controls confirmed that the complex, repeating states found in the healthy retinas were a direct response to the visual stimulus.
The tool also allowed the researchers to test how the retina reacts when a specific chemical pathway is blocked. By applying a drug that silences the "on" pathway of the retina, they observed a dramatic change in the population's behavior. The sequence of states shifted visibly, and the timing of the responses changed, confirming that the tool could detect how specific biological manipulations alter the collective activity of the neural network. This sensitivity suggests that the method can be used to understand how different parts of the visual system contribute to the overall picture the brain receives.
The significance of this work lies in its ability to turn a chaotic stream of electrical signals into a manageable, descriptive model of how the retina works as a team. The researchers did not discover a new type of cell or a new biological law; instead, they built a reusable pipeline that connects the timing of a stimulus to the collective behavior of the population. The states identified by the tool provide a descriptive map of the population's activity, showing which groups of cells are active together and for how long. While the exact number of states found in the simulations was sometimes slightly higher than the true number, the sequence of events was recovered with high precision, proving that the method is robust for finding the underlying structure of neural activity. This approach offers a new way for scientists to inspect how the brain's first visual station processes information, moving beyond the study of individual cells to understand the group dynamics that define vision.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.