Publications

Description: If you were headed to a game show but couldn't know the rules, topics, content, competition, or even language, what could you do to prepare? This is what every brain faces in terms of development, evolution, and daily life: the world is unpredictable. All it can do is pick some assortment of general principle that maximize its chances of success faced with an unpredictable challenge of arbitrary difficulty. Mathematically, your best bet in this situation is to choose criticality. However, there's a BIG problem with this; for a century, the backbone of theoretical neuroscience has been low dimensional dynamical models. But criticality requires a system of infinite size: it simply doesn't exist, formally speaking, in these models. How then, could criticality be a universal, necessary explanation of the brain if it doesn't even exist in our most important models? Here, Leandro drops a heavy theory paper, defining criticality in low dimensional systems by considering time as equivalent to space. He shows that there are two mechanistically distinct sources of criticality and that some but not all canonical models exhibit criticality (and only a subset of these are biologically plausible). Consistent with the core hypothesis, information processing peaks at criticality. This ties disparate modeling approaches together within a single framework and suggests that optimal information processing emerges not from model-specific mechanisms but from fundamental properties of critical dynamics themselves.

Description: Study of early neural network development using human cortical organoids, revealing fundamental patterns of spontaneous activity that model intrinsic brain states.

Description: Gemechu and Aidan asked a fundamental question: is information about a neuron's anatomical location embedded in its spike train? The null must be no: there is no prior evidence that we can look at a timeseries (a list of times when a neuron fires) and say anything about that neuron's identity. Further, we've long modeled neurons as rate varying Poisson (random) processes whose outputs (spike trains) transmit information about their inputs (stimuli). Against this backdrop, G & A worked with Allen Institute datasets to train machine learning models on single neuron time series. Lo and behold, they were able to identify the anatomical home of individual cells based on spike timing. Further, these models (trained on Allen Institute recordings) tested above chance on data from an independent group. Combined with the LOLCAT paper (see Scheider, Azabou 2023, below), this suggests that, in addition to stimulus information, spike trains carry robust latent information about identity.

Description: At two weeks old, mice abruptly begin to actively use their whiskers to explore the world. Molly and Lucy discovered that this behavioral switch comes alongside a transient increase in the excitability of excitatory neurons in layer 4 barrel cortex. These are the neurons that process whisker input --- this brief increase in excitability is poised to drive plasticity and requires prior experience.

Description: Invited review of Ruggiero... Slutsky (2024, Neuron). Ravi makes a strong argument that homeostasis is an emergent process that is concerned with emergent end-points. Since reliability of behavior and cognition are of utmost importance, it must be the case that there are cellular mechanisms that result in its (the reliability's) maintenance.

Description: David and Aidan developed a revolutionary approach to understanding brain states that doesn't rely on traditional oscillations or rhythms. Instead, they rigorously and exhautively demonstrated that the brain's sleep/wake state is most accurately captured by millisecond-scale patterns in neural activity. Mathematically, this precludes the canonical waves that have been used to describe/identify sleep and wake for a century. As a result, it raises the possibility of local control of states. Aidan and David elegantly showed that local states of all flavors (sleep in a waking brain, wake in sleeping brain, etc.) occur regularly throughout the brain as brief "flickers". Flickers predict macroscopic features of animal behavior. A plausible interpretation is that the minimal unit of a state (e.g., sleep) is not a slow, global wave, but a millisecond-scale structuring of neural activity that plays out in micrometers of the brain. A global state, as we experience them, then requires a massive coordinating signal. Perhaps this is the role of the slower processes?

Description: Investigation of how sensory experience shapes neural connectivity during critical developmental periods using advanced imaging techniques.

Description: Why all animals sleep is one of the great questions in biology. Because sleep improves everything a brain does (and conversely, sleep deprivation undermines everything from reasoning to breathing), it is reasonable to hypothesize that the core, restorative function of sleep is to resurrect a universally necessary feature of computation. In this paper, Yifan demonstrates that sleep restores criticality, an optimal computational regime, in cortical networks. More than any other tested measure, distance from criticality predicts future sleep/wake. This provides a direct explanation of how sleep supports brain function.

Description: Aidan and Mehdi used Allen Institute data (dense microelectrode and calcium imaging) to as if information about a neuron's transcriptomic cell is embedded in its spike timeseries. While there's some prior indication that certain cell types tend to fire at higher rates than others, this isn't sufficient to identify a single cell --- it's a population statistic. Put simply, if you were given a raster plot of one neuron's activity (recorded in an awake, behaving animal) with no other information, could you reliably identify the cell type? With this as the challenge, A & M show that cell type can be robustly recovered from the activity timeseries. Taken with Aidan and Gemechu's 2025 eLife paper (anatomy is embedded in the spike timeseries), it seems clear that neuronal spiking carries much more information than stimuli alone. Such models are fundamentally data-limited; future work will increase the accuracy and universalizability of these insights.

Description: Development of Internet of Things infrastructure for high-throughput cellular biology research, enabling large-scale automated experiments.

Description: Novel machine learning approach for generating synthetic neural activity data using self-supervised learning methods.

Description: Development and characterization of a mouse model of MYT1L syndrome, revealing mechanisms of altered neuronal maturation in neurodevelopmental disorders.

Description: Investigation of human microglia in brain organoid models, revealing conserved states and regulatory functions in neural development.

Description: Methodological paper detailing construction and use of carbon fiber microelectrodes for neural recording applications.

Description: Theoretical and experimental investigation of how homeostatic mechanisms regulate different aspects of cortical circuit function.

Description: Investigation of how Shank3, an autism-associated protein, regulates homeostatic plasticity in visual cortex.

Description: Invited review discussing daily fluctuations in excitatory-inhibitory balance and their functional significance.

Description: Investigation of how visual cortex rapidly adapts firing rates during light-dark transitions through homeostatic mechanisms.

Description: For brain activity to be stable/reliable (on a long timescale), there is likely to be homeostatic control over neuronal firing rates. This is fundamental, but quite hard to test. We developed one of the first methods for following ensembles of single neurons continuously for 9 days. In response to sustained perturbation, neuronal activity slowly recovered --- but the homeostatic recovery only occurred when animals were awake. This paper established three important contributions: 1) Individual neurons have a thermostat-like set-point for their own firing rate that is independent of stimuli, 2) homeostatic plasticity interacts with sleep/wake states, and 3) methods of chronic, continuous recording from single neurons in freely behaving animals.

Description: Investigation of anesthetic tolerance mechanisms involving GABA receptor subunit changes in respiratory control circuits.

Description: Investigation of respiratory adaptations during pregnancy involving changes in GABA receptor expression. Selected by Faculty of 1000 as top 2% of published articles.

Description: Study of GABA receptor plasticity during hibernation and its effects on drug sensitivity in respiratory control circuits.

Description: Cognitive psychology study examining reading processes using experimental paradigms from decision-making research.

Description: Investigation of regional differences in drug sensitivity that develop during hibernation, focusing on brainstem versus cortical responses.

Description: Conference presentation exploring how neuronal location information is encoded in spike train patterns at COSYNE 2024.

Description: Machine learning approach for detecting changes in neural population dynamics using contrastive learning methods.

Description: Oral presentation at NeurIPS (top 1% of submitted papers) on self-supervised neural activity generation.

Description: Workshop oral presentation on self-supervised learning methods for neural data analysis.

Description: Workshop paper exploring self-supervised learning applications to understanding neural coding principles.

Description: How can we precisely define and measure how close neural systems are to the critical point? Sam Sooter (next gen genius...) developed a rigorous mathematical framework based on renormalization group theory (1982 Nobel Prize to Kenneth Wilson) to precise measure distance to temporal criticality. At criticality, a signal will carry information at all time scales --- the millisecond and the hour both have meaning. As you move away from criticality, this range decreases. To directly measure that range, Sam created a computational tool, d₂, providing what will be a field standard method. Sam's work is essential for understanding criticality in neural systems. One of the big deliverables is that distance to criticality can be measured in seconds, rather than hours.

Description: Interdisciplinary work examining how temperature affects predation dynamics and its implications for future ecosystem structure and climate change impacts.

Description: Investigation of selective vulnerability patterns in the default mode network in the context of Alzheimer's disease pathology, revealing how certain brain circuits remain vulnerable despite widespread amyloid accumulation.

Description: Machine learning methodology for detecting changes in time series data using optimal transport theory and Sinkhorn divergences.