Biology subjects

Mitelut, C.

Publications and source records attributed to Mitelut, C..

10 recordsLinked to original sources

bioRxiv2026

Nethobench: Forecasting design and conditioning shape the structural realism of generated neural dynamics

Forecasting models are increasingly used to model neural population activity, yet existing benchmarks typically evaluate them by pointwise predictive fidelity to held-out observations. In stochastic, partially observed systems, agreement with one sampled realization is distinct from preserving the statistical and dynamical structure of the process. We introduce Nethobench, a benchmark that evaluates structural agreement across five metric families: distributional, temporal--spectral, relational, geometric, and state-dynamical. We validate Nethobench using targeted perturbations and stochastic systems with known dynamics, and apply it to mouse calcium imaging, mouse Neuropixels recordings, and human intracranial recordings, using empirical references to calibrate interpretation. Nethobench detects targeted structural failures and assigns high scores to independent samples from the correct synthetic process despite trajectory mismatch. It tracks structural changes during training and under domain shift. Model comparisons reveal how forecasting formulation, architecture, training regime, and loss shape structural profiles, while OmniMouse input ablations improve all five family means when behavior is added to video in one held-out session. In Brain Treebank, analog retrieval passes more structural checks while a multistep GRU better predicts the observed realization. These results motivate reporting calibrated family-level structural metrics alongside pointwise accuracy, proper probabilistic scores, and task-specific endpoints when evaluating neural forecasts across species and recording modalities.

neuroscience↗

bioRxiv2026

Associative memory formation reshapes the learning trajectory of a novel cognitive map

Learning never occurs on a blank slate, as each experience an animal encounters in its daily life is encoded by circuits already shaped by previous ones. During such experiences, the hippocampus builds place-cell-based cognitive maps of the environments explored, and encodes ensemble-based memories for associations among events that occur within them. These spatial and mnemonic representations draw on neurons embedded within the same network. However, whether forming an associative memory of an experience unfolding in one context changes how the same hippocampal network maps a different one is unknown. Here, we longitudinally imaged CA3 neurons as mice familiarized with one environment, encoded either an associative fear memory or a context-only memory in a different environment, and then explored a novel one. Across separate experiences, neuronal recruitment into cognitive maps and memory ensembles followed structured, partially orthogonal allocation rules that were consistent across fear and neutral memories, with ensemble membership and place-cell identity showing no reciprocal enrichment. Fear memory encoding left the properties of the familiar map unchanged, while altering those of the novel one. From first exposure, the novel map already recruited a familiar-like fraction of place cells and supported spatial decoding with near-familiar precision, and its population activity occupied a latent subspace that aligned with the familiar map while remaining context-specific. Yet this novel map failed to undergo the experience-dependent refinement of decoding precision and across-day latent alignment that normally accompanies familiarization, even though place cells continued to gain spatial information and to remap. Associative memory encoding therefore shifts CA3 into a distinct representational regime, decoupling the immediate precision of a new cognitive map from the experience-dependent refinement of its context-specific population embedding.

neuroscience↗

bioRxiv2026

Equivalent volitional learning emerges through circuit-specific population dynamics in motor cortex and hippocampus

Learning operates across different brain circuits to associate population activity patterns with desired outcomes, and to enable volitional reactivation of those patterns to control behavior. These circuits differ profoundly in their architecture and dynamical regimes, yet which features of learning are shared across them and which arise from circuit-specific implementations remains unknown. Here, we use a brain-computer interface (BCI) to train mice to modulate the activity of selected neuronal ensembles toward configurations that trigger reward delivery. By making reward delivery contingent directly on population activity, we impose an identical associative learning problem on two circuits with distinct dynamical regimes: the primary motor cortex (M1) and the hippocampal area CA3. Mice acquired robust volitional control in both regions, and learning produced a set of shared signatures across circuits, including modulation of reward-controlling neurons, network-level sparsification, and greater exploration of reward-related activity patterns. These signatures were underpinned by distinct population dynamics: M1 activity flowed continuously through reward-associated states, whereas CA3 activity traced approach-and-return dynamics around them. Recurrent network models endowed with distinct minimal connectivity constraints chosen to reflect the dominant dynamical regime associated with each region captured key features of these shared signatures and region-specific dynamics, indicating that local architectural constraints are sufficient to account for the distinct implementations of learning. These findings indicate that equivalent learning outcomes arise from divergent dynamical implementations across architecturally distinct circuits. This principled degeneracy reveals that learning is not a single canonical solution, but is implemented through multiple circuit-specific mechanisms shaped by local network architecture.

neuroscience↗

bioRxiv2023

Distinct developmental trajectories of autonomous behaviors and agency in rodents

Behavioral interactions within the nuclear family play a pivotal role in the emergence of agency: the capacity to regulate physiological, psychological and social needs. While behaviors may develop over days or weeks in line with nervous system maturation, individual behaviors can occur on sub-second time scales making it challenging to track development in lab studies with brief observation periods, or in field studies with limited temporal precision and animal identification. Here we study development in families of gerbils, a highly social rodent, collecting tens of millions of behavior time points and implementing machine learning methods to track individual subjects. We provided maturing gerbils with a large, undisturbed environment between postnatal day 15 and the age at which they would typically disperse from the family unit (day 30). We identified complex and distinct developmental trajectories for food and water acquisition, solitary exploration, and social behaviors, some of which displayed sex differences and diurnal patterns. Our work supports the emergence of well-delineated autonomous and social behavior phenotypes that correlate with specific periods and loci of neural maturation.

animal behavior and cognition↗

bioRxiv2023

Behavioral Classification of Sequential Neural Activity Using Time Varying Recurrent Neural Networks

Shifts in data distribution across time can strongly affect early classification of time-series data. When decoding behavior from neural activity, early detection of behavior may help in devising corrective neural stimulation before the onset of behavior. Recurrent Neural Networks (RNNs) are common models for sequence data. However, standard RNNs are not able to handle data with temporal distributional shifts to guarantee robust classification across time. To enable the network to utilize all temporal features of the neural input data, and to enhance the memory of an RNN, we propose a novel approach: RNNs with time-varying weights, here termed Time-Varying RNNs (TV-RNNs). These models are able to not only predict the class of the time-sequence correctly but also lead to accurate classification earlier in the sequence than standard RNNs. In this work, we focus on early sequential classification of brain-wide neural activity across time using TV-RNNs applied to a variety of neural data from mice and humans, as subjects perform motor tasks. Finally, we explore the contribution of different brain regions on behavior classification using SHapley Additive exPlanation (SHAP) value, and find that the somatosensory and premotor regions play a large role in behavioral classification.

neuroscience↗

bioRxiv2021

Mesoscale cortex-wide neural dynamics predict self-initiated actions in mice several seconds prior to movement

Volition - the sense of control or agency over ones voluntary actions - is widely recognized as the basis of both human subjective experience and natural behavior in non-human animals. Several human studies have found peaks in neural activity preceding voluntary actions, e.g. the readiness potential (RP), and some have shown upcoming actions could be decoded even before awareness. Others propose that random processes underlie and explain pre-movement neural activity. Here we seek to address these issues by evaluating whether pre-movement neural activity in mice contains structure beyond that present in random neural activity. Implementing a self-initiated water-rewarded lever pull paradigm in mice while recording widefield [Ca++] neural activity we find that cortical activity changes in variance seconds prior to movement and that upcoming lever pulls could be predicted between 3 to 5 seconds (or more in some cases) prior to movement. We found inhibition of motor cortex starting at approximately - 5sec prior to lever pulls and activation of motor cortex starting at approximately 2sec prior to a random unrewarded left limb movement. We show that mice, like humans, are biased towards commencing self-initiated actions during specific phases of neural activity but that the pre-movement neural code changes over time in some mice and is widely distributed as behavior prediction improved when using all vs single cortical areas. These findings support the presence of structured multi-second neural dynamics preceding self-initiated action beyond that expected from random processes. Our results also suggest that neural mechanisms underlying self-initiated action could be preserved between mice and humans.

neuroscience↗

bioRxiv2020

Nonlinear decoding of natural images from large-scale primate retinal ganglion recordings

Decoding sensory stimuli from neural activity can provide insight into how the nervous system might interpret the physical environment, and facilitates the development of brain-machine interfaces. Nevertheless, the neural decoding problem remains a significant open challenge. Here, we present an efficient nonlinear decoding approach for inferring natural scene stimuli from the spiking activities of retinal ganglion cells (RGCs). Our approach uses neural networks to improve upon existing decoders in both accuracy and scalability. Trained and validated on real retinal spike data from > 1000 simultaneously recorded macaque RGC units, the decoder demonstrates the necessity of nonlinear computations for accurate decoding of the fine structures of visual stimuli. Specifically, high-pass spatial features of natural images can only be decoded using nonlinear techniques, while low-pass features can be extracted equally well by linear and nonlinear methods. Together, these results advance the state of the art in decoding natural stimuli from large populations of neurons. Author summaryNeural decoding is a fundamental problem in computational and statistical neuroscience. There is an enormous literature on this problem, applied to a wide variety of brain areas and nervous systems. Here we focus on the problem of decoding visual information from the retina. The bulk of previous work here has focused on simple linear decoders, applied to modest numbers of simultaneously recorded cells, to decode artificial stimuli. In contrast, here we develop a scalable nonlinear decoding method to decode natural images from the responses of over a thousand simultaneously recorded units, and show that this decoder significantly improves on the state of the art.

neuroscience↗

bioRxiv2020

YASS: Yet Another Spike Sorter applied to large-scale multi-electrode array recordings in primate retina

Spike sorting is a critical first step in extracting neural signals from large-scale multi-electrode array (MEA) data. This manuscript presents several new techniques that make MEA spike sorting more robust and accurate. Our pipeline is based on an efficient multi-stage "triage-then-cluster-then-pursuit" approach that initially extracts only clean, high-quality waveforms from the electrophysiological time series by temporarily skipping noisy or "collided" events (representing two neurons firing synchronously). This is accomplished by developing a neural network detection and denoising method followed by efficient outlier triaging. The denoised spike waveforms are then used to infer the set of spike templates through nonparametric Bayesian clustering. We use a divide-and-conquer strategy to parallelize this clustering step. Finally, we recover collided waveforms with matching-pursuit deconvolution techniques, and perform further split-and-merge steps to estimate additional templates from the pool of recovered waveforms. We apply the new pipeline to data recorded in the primate retina, where high firing rates and highly-overlapping axonal units provide a challenging testbed for the deconvolution approach; in addition, the well-defined mosaic structure of receptive fields in this preparation provides a useful quality check on any spike sorting pipeline. We show that our pipeline improves on the state-of-the-art in spike sorting (and outperforms manual sorting) on both real and semi-simulated MEA data with > 500 electrodes; open source code can be found at https://github.com/paninski-lab/yass.

neuroscience↗

bioRxiv2019

Stereotyped large amplitude cortical LFP events can be clustered and reveal precisely ordered phase-locking in neuronal populations

During quiet wakefulness, slow-wave sleep and anesthesia, mammalian cortex exhibits a synchronised state during which transient changes in the local field potential (LFP) accompany periods of increased single neuron firing, known as UP-states. While UP-state genesis is still debated (Crunelli and Hughes, 2010) such transitions may constitute the default activity pattern of the entire cortex (Neske, 2016). Recent findings of preserved firing order between UP-state transitions and stimulus processing in high-firing rate (>2Hz) rat auditory and barrel cortex neurons (Luczak et al., 2015) support this hypothesis. Yet it is unknown whether UP-states are homogeneous and whether neurons with firing rates <2Hz in visual cortex or other species exhibit spiking order. Using extracellular recordings during anesthetized states in cat visual cortex and mouse visual, auditory and barrel cortex, we show that UP-states can be tracked and clustered based on the shape of the LFP waveform. We show that LFP event clusters (LECs) have current-source-density profiles that are common across different recordings or animals and using simultaneous electrophysiology and widefield voltage and calcium imaging in mouse we confirm that LEC transitions are cortex-wide phenomena. Individual LEC events can be resolved in time to within 1 - 4 ms and they elicit synchronous firing of over 75% of recorded neurons with most neurons synchronizing their firing to within {+/-}5 - 15 ms relative LECs. Firing order of different neurons during LEC events was preserved over periods of ~30 minutes enabling future studies of UP-state transitions and firing order with near millisecond precision.\n\nSignificant StatementDuring sleep and anesthetic states mammalian cortex undergoes substantial changes from awake active states. Recent studies show that single neurons in some cortical areas in rats undergo increased spiking during sleep and anesthetic states (called UP-state transitions) with some neurons firing in an order similar to awake states. This suggests that sensory processing may be similar across all states and that firing order is important for stimulus processing. Yet UP-state transitions remain poorly understood and it is unclear whether firing order is present in other cortical areas or species. Here we describe multiple classes of UP-state transitions and show most neurons in visual cortex in cats and visual, barrel and auditory cortex in mice exhibit firing order during such transitions.

neuroscience↗

bioRxiv2017

Real-time spike sorting platform for high-density extracellular probes with ground-truth validation and drift correction

Electrical recordings from a large array of electrodes give us access to neural population activity with single-cell, single-spike resolution. These recordings contain extracellular spikes which must be correctly detected and assigned to individual neurons. Despite numerous spike-sorting techniques developed in the past, a lack of high-quality ground-truth datasets hinders the validation of spike-sorting approaches. Furthermore, existing approaches requiring manual corrections are not scalable for hours of recordings exceeding 100 channels. To address these issues, we built a comprehensive spike-sorting pipeline that performs reliably under noise and probe drift by incorporating covariance-based features and unsupervised clustering based on fast density-peak finding. We validated performance of our workflow using multiple ground-truth datasets that recently became available. Our software scales linearly and processes up to 1000-channel recording in real-time using a single workstation. Accurate, real-time spike sorting from large recording arrays will enable more precise control of closed-loop feedback experiments and brain-computer interfaces.

neuroscience↗