Biology subjects

Vidaurre, D.

Publications and source records attributed to Vidaurre, D..

At least 37 records · Page 2Linked to original sources

bioRxiv2023

Impaired activation of the prefrontal executive network during working memory processing in multiple sclerosis

AbstractIn multiple sclerosis (MS), working memory (WM) impairment occurs soon after disease onset and significantly affects the patients quality of life. Functional imaging research in MS aims to investigate the neurophysiological underpinnings of WM impairment. In this context, we utilized a data-driven technique, the time delay embedded- hidden Markov model (TDE-HMM), to extract spectrally defined functional networks in magnetoencephalographic (MEG) data acquired during a WM visual-verbal n-back task. We observed that two networks show an altered activation in RR-MS patients. First, the activation of an early theta prefrontal network linked to stimulus encoding and attentional control significantly decreased in RR-MS compared to HC. This diminished activation correlated with reduced accuracy in task performance in the MS group, suggesting an impaired encoding and learning process. Secondly, a frontoparietal network characterized by beta coupling is activated between 300 and 600 ms after stimulus onset; this resembles the characteristic event-related P300, a cognitive marker extensively explored in EEG studies. The activation of this network is amplified in patients treated with benzodiazepine, in line with the well-known benzodiazepine-induced beta enhancement. Altogether, the TDE-HMM technique extracted task-relevant functional networks showing disease-specific and treatment- related alterations, revealing potential new markers to assess and track WM impairment in MS. HighlightsO_LIWe decomposed the brain dynamics underlying a WM n-back task in data-driven, spectrally defined whole-brain networks in both healthy controls and people with relapsing-remitting-MS (pwMS). C_LIO_LIPwMS showed a significantly decreased activation of an early theta prefrontal network linked to stimulus encoding and attentional control. C_LIO_LIThe weaker activation of this prefrontal theta network is correlated with worse task performance. C_LIO_LIA beta frontoparietal network with a P300-like temporal evolution was significantly modulated by the use of benzodiazepines. C_LIO_LIThe model distinguished disease-induced and treatment-induced dynamic network alterations. C_LI

neuroscience↗

bioRxiv2023

Stacking models of brain dynamics improves prediction of subject traits in fMRI

Beyond structural and time-averaged functional connectivity brain measures, the way brain activity dynamically unfolds can add important information when investigating individual cognitive traits. One approach to leveraging this information is to extract features from models of brain network dynamics to predict individual traits. However, there are two potential sources of variation in the models estimation which will in turn affect the predictions: first, in certain cases, the estimation variability due to different initialisations or choice of inference method; and second, the variability induced by the choice of the model hyperparameters that determine the complexity of the model. Rather than merely being statistical noise, this variability may be useful in providing complementary information that can be leveraged to improve prediction accuracy. We propose stacking, a prediction-driven approach for model selection, to leverage this variability. Specifically, we combine predictions from multiple models of brain dynamics to generate predictions that are accurate and robust across multiple cognitive traits. We demonstrate the approach using the Hidden Markov Model, a probabilistic generative model of brain network dynamics. We show that stacking can significantly improve the prediction of subject-specific phenotypes, which is crucial for the clinical translation of findings.

neuroscience↗

bioRxiv2023

Predicting subject traits from brain spectral signatures: an application to brain ageing

The prediction of subject traits using brain data is an important goal in neuroscience, with relevant applications in clinical research, as well as in the study of differential psychology and cognition. While previous prediction work has predominantly been done on neuroimaging data, our focus is on electroencephalography (EEG), a relatively inexpensive, widely available and non-invasive data modality. However, EEG data is complex and needs some form of feature extraction for subsequent prediction. This process is sometimes done manually, risking biases and suboptimal decisions. Here we investigate the use of data-driven kernel methods for prediction from single-channels using the the EEG spectrogram, which reflects macro-scale neural oscillations in the brain. Specifically, we introduce the idea of reinterpreting the the spectrogram of each channel as a probability distribution, so that we can leverage advanced machine learning techniques that can handle probability distributions with mathematical rigour and without the need for manual feature extraction. We explore how the resulting technique, Kernel Mean Embedding Regression, compares to a standard application of Kernel Ridge Regression as well as to a non-kernelised approach. Overall, we found that the kernel methods exhibit improved performance thanks to their capacity to handle nonlinearities in the relation between the EEG spectrogram and the trait of interest. We leveraged this method to predict biological age in a multinational EEG data set, HarMNqEEG, showing the methods capacity to generalise across experiments and acquisition setups.

neuroscience↗

bioRxiv2023

Spontaneous network transitions predict somatosensory perception

Sensory perception is essential for transforming incoming information in the brain into targeted behavior. Our brains are everlastingly active, and variations in perception are ubiquitously associated with human behavioral performance. Previous studies indicate that changes in spontaneous neural activity within local sensory areas correlate with the perception of ambiguous stimuli. However, the contribution of whole brain spontaneous networks to perception is not well understood. Using an ambiguous tactile temporal discrimination task, we demonstrate that the interaction between wholebrain networks in the seconds of the spontaneous pre-stimulus period also contributes to perception during the task. Transitions to a frontal and a multi-frequency network across the brain are essential for the correct percept. Conversely, incorrect percepts are mainly preceded by transitions to an alphaparietal network. Brain transitions occur faster during the period before stimulus presentation for correct stimuli detection, suggesting the need for enhanced network flexibility during this phase. Significance statementOur brain is constantly engaged in processing sensory input and translating it into sensory perceptions. When confronted with ambiguous sensory information, individuals do not always have the same perceptual experience. We demonstrate that brain network transitions to frontal areas are essential for the correct percept. Conversely, incorrect percepts are mainly preceded by transitions to an alpha-parietal network. Correct stimuli detections are characterized by faster transitions, suggesting the need for enhanced network flexibility. These results extend our knowledge of perception by pointing to the relevance of whole-brain spontaneous networks and their dynamic properties.

neuroscience↗

bioRxiv2023

osl-dynamics: A toolbox for modelling fast dynamic brain activity

Neural activity contains rich spatio-temporal structure that corresponds to cognition. This includes oscillatory bursting and dynamic activity that span across networks of brain regions, all of which can occur on timescales of a tens of milliseconds. While these processes can be accessed through brain recordings and imaging, modelling them presents methodological challenges due to their fast and transient nature. Furthermore, the exact timing and duration of interesting cognitive events is often a priori unknown. Here we present the OHBA Software Library Dynamics Toolbox (osl-dynamics), a Python-based package that can identify and describe recurrent dynamics in functional neuroimaging data on timescales as fast as tens of milliseconds. At its core are machine learning generative models that are able to adapt to the data and learn the timing, as well as the spatial and spectral characteristics, of brain activity with few assumptions. osl-dynamics incorporates state-of-the-art approaches that can be, and have been, used to elucidate brain dynamics in a wide range of data types, including magneto/electroencephalography, functional magnetic resonance imaging, invasive local field potential recordings and electrocorticography. It also provides novel summary measures of brain dynamics that can be used to inform our understanding of cognition, behaviour and disease. We hope osl-dynamics will further our understanding of brain function, through its ability to enhance the modelling of fast dynamic processes. HighlightsO_LIAn open-source toolbox for identifying and describing brain dynamics in neuroimaging data on fast timescales. C_LIO_LIIncludes visualisation and quantification of oscillatory bursting and network dynamics. C_LIO_LIProvides novel summary measures and group analysis tools of brain dynamics that can be used to inform our understanding of cognition, behaviour and disease. C_LIO_LIImplemented in Python and makes use of TensorFlow. C_LIO_LIIncludes comprehensive documentation and tutorials. C_LI

neuroscience↗

bioRxiv2023

Large-scale cortical networks are organized in structured cycles

The brain needs to perform a diverse set of cognitive functions essential for survival, but it is unknown how it is organized to ensure that each of these functions is fulfilled within a reasonable period. One way in which this requirement can be met is if each of these cognitive functions occur as part of a repeated cycle. Here, we studied the temporal evolution of canonical large-scale cortical networks, and show that while network dynamics are stochastic, the overall ordering of their activity forms a robust cyclical pattern. This cyclical structure groups states with similar function and spectral content at specific phases of the cycle and occurs at timescales of (300-1000 ms). These results are reproduced in five large magnetoencephalography (MEG) datasets. Moreover, we show that metrics that characterize the cycle strength and speed are heritable, and relate to age, cognition, and behavioural performance. These results reveal for the first time that the activations of a canonical set of large-scale cortical networks are organised in an inherently cyclical manner, ensuring periodic activation of essential cognitive functions.

neuroscience↗

bioRxiv2023

Differential activity patterns in the upper and lower blade of the dentate gyrus

The dentate gyrus is considered the first stage in the trisynaptic circuit of the hippocampus. Granule cells of the dentate gyrus fire very sparsely with a low probability of overlapping patterns of active neurons. This characteristic supports the most prevalent view of the dentate gyrus function as that of a pattern separator: to generate dissimilar neuronal representations from overlapping input states that represent similar but not identical environments. However, there are two distinct granule cell blades that have been shown to have different activity patterns: the upper blade and the lower blade. These may support different purposes, but their differential function is not well understood. Here we have recorded calcium imaging data from both the upper and the lower blade of the dentate gyrus from two mice (the upper blade from one, the lower blade from the other), while they perform a simple decision-making task. We found that only the lower blade encodes subjective emotional response while the upper blade preferentially encodes the actual response. Importantly, we found that correlations between cells carried more information about these two behavioural variables than individual firings, suggesting that neuron correlations encode here important information above and beyond individual unit activity.

neuroscience↗

bioRxiv2023

Predicting individual traits from models of brain dynamics accurately and reliably using the Fisher kernel

Predicting an individuals cognitive traits or clinical condition using brain signals is a central goal in modern neuroscience. This is commonly done using either structural aspects, such as structural connectivity or cortical thickness, or aggregated measures of brain activity that average over time. But these approaches are missing a central aspect of brain function: the unique ways in which an individuals brain activity unfolds over time. One reason why these dynamic patterns are not usually considered is that they have to be described by complex, high-dimensional models; and it is unclear how best to use these models for prediction. We here propose an approach that describes dynamic functional connectivity and amplitude patterns using a Hidden Markov model (HMM) and combines it with the Fisher kernel, which can be used to predict individual traits. The Fisher kernel is constructed from the HMM in a mathematically principled manner, thereby preserving the structure of the underlying model. We show here, in fMRI data, that the HMM-Fisher kernel approach is accurate and reliable. We compare the Fisher kernel to other prediction methods, both time-varying and time-averaged functional connectivity-based models. Our approach leverages information about an individuals time-varying amplitude and functional connectivity for prediction and has broad applications in cognitive neuroscience and personalised medicine.

neuroscience↗

bioRxiv2023

Towards stability of dynamic FC estimates in neuroimaging and electrophysiology: solutions and limits

Time-varying functional connectivity methods are used to map the spatiotemporal organization of brain activity. However, their estimation can be unstable, in the sense that different runs of the inference may yield different solutions. But to draw meaningful relations to behaviour, estimates must be robust and reproducible. Here, we propose two solutions using the Hidden Markov Model (HMM) as a descriptive model of time-varying FC. The first, best-ranked HMM, involves running the inference multiple times and selecting the best model based on a quantitative measure combining fitness and model complexity. The second, hierarchical clustered HMM, generates stable aggregated state timeseries by applying hierarchical clustering to the state timeseries obtained from multiple runs. Experimental results on fMRI and MEG data demonstrate that these approaches substantially improve the stability of time-varying FC estimations. Overall, hierarchical clustered HMM is preferred when the inference variability is high, while the best-ranked HMM performs better otherwise.

neuroscience↗

bioRxiv2023

Dissecting unsupervised learning through hidden Markov modelling in electrophysiological data

Unsupervised, data-driven methods are commonly used in neuroscience to automatically decompose data into interpretable patterns. These patterns differ from one another depending on the assumptions of the models. How these assumptions affect specific data decompositions in practice, however, is often unclear, which hinders model applicability and interpretability. For instance, the hidden Markov model (HMM) automatically detects characteristic, recurring activity patterns (so-called states) from time series data. States are defined by a certain probability distribution, whose state-specific parameters are estimated from the data. But what specific features, from all of those that the data contain, do the states capture? That depends on the choice of probability distribution and on other model hyperparameters. Using both synthetic and real data, we aim at better characterizing the behavior of two HMM types that can be applied to electrophysiological data. Specifically, we study which differences in data features (such as frequency, amplitude or signal-to-noise ratio) are more salient to the models and therefore more likely to drive the state decomposition. Overall, we aim at providing guidance for an appropriate use of this type of analysis on one or two-channel neural electrophysiological data, and an informed interpretation of its results given the characteristics of the data and the purpose of the analysis. NEW & NOTEWORTHYCompared to classical supervised methods, unsupervised methods of analysis have the advantage to be freer of subjective biases. However, it is not always clear what aspects of the data these methods are most sensitive to, which complicates interpretation. Focusing on the Hidden Markov Model, commonly used to describe electrophysiological data, we explore in detail the nature of its estimates through simulations and real data examples, providing important insights about what to expect from these models.

neuroscience↗

bioRxiv2023

A novel description of the network dynamics underpinning working memory

Working memory (WM) plays a central role in cognition, prompting neuroscientists to investigate its functional and structural substrates. The WM dynamic recruits large-scale frequency-specific brain networks that unfold over a few milliseconds - this complexity challenges traditional neuroimaging analyses. In this study, we unravel the WM network dynamics in an unsupervised, data-driven way, applying the time delay embedded-hidden Markov model (TDE-HMM). We acquired MEG data from 38 healthy subjects performing an n-back working memory task. The TDE-HMM model inferred four task-specific states with each unique temporal (activation), spectral (phase-coherence connections), and spatial (power spectral density distribution) profiles. A theta frontoparietal state performs executive functions, an alpha temporo-occipital state maintains the information, and a broad-band and spatially complex state with an M300 temporal profile leads the retrieval process and motor response. The HMM states can be straightforwardly interpreted within the neuropsychological multi-component model of WM, significantly improving the comprehensive description of WM. HighlightsO_LIWorking memory recruits different frequency-specific brain networks that wax and wane at a millisecond scale. C_LIO_LIThrough the time-delay embedded hidden (TDE-HMM) we are able to extract data-driven functional networks with unique spatial, spectral, and temporal profiles. C_LIO_LIWe demonstrate the existence of four task-specific brain networks that can be interpreted within the well-known Baddeleys multicomponent model of working memory. C_LIO_LIThis novel WM description unveils new features that will lead to a more in-depth characterization of cognitive processes in MEG data. C_LI

neuroscience↗

bioRxiv2023

Modelling the stable subspace of distributed brain responses in stimulus processing

Each brain response to a stimulus is, to a large extent, unique. However this variability, our perceptual experience feels stable. Standard decoding models, which utilise information across several areas to tap into stimuli representation and processing, are fundamentally based on averages. Therefore, they can focus precisely on the features that are most stable across stimulus presentations. But which are these features exactly is difficult to address in the absence of a generative model of the signal. Here, I introduce genephys, a generative model of brain responses to stimulation publicly available as a Python package that, when confronted with a decoding algorithm, can reproduce the structured patterns of decoding accuracy that we observe in real data. Using this approach, I characterise how these patterns may be brought about by the different aspects of the signal, which in turn may translate into distinct putative neural mechanisms. In particular, the model shows that the features in the data that support successful decoding --and, therefore, likely reflect stable mechanisms of stimulus representation-- have an oscillatory component that spans multiple channels, frequencies and latencies of response; and an additive, slower response with a specific (cross-frequency) relation to the phase of the oscillatory component. At the individual trial level, still, responses are found to be highly variable, which can be due to various factors including phase noise and probabilistic activations.

neuroscience↗

bioRxiv2022

The relationship between frequency content and representational dynamics in the decoding of neurophysiological data

Decoding of high temporal resolution, stimulus-evoked neurophysiological data is increasingly used to test theories about how the brain processes information. However, a fundamental relationship between the frequency spectra of the neural signal and the subsequent decoding accuracy timecourse is not widely recognised. We show that, in commonly used instantaneous signal decoding paradigms, each sinusoidal component of the evoked response is translated to double its original frequency in the subsequent decoding accuracy timecourses. We therefore recommend, where researchers use instantaneous signal decoding paradigms, that more aggressive low pass filtering is applied with a cut-off at one quarter of the sampling rate, to eliminate representational alias artefacts. However, this does not negate the accompanying interpretational challenges. We show that these can be resolved by decoding paradigms that utilise both a signals instantaneous magnitude and its local gradient information as features for decoding. On a publicly available MEG dataset, this results in decoding accuracy metrics that are higher, more stable over time, and free of the technical and interpretational challenges previously characterised. We anticipate that a broader awareness of these fundamental relationships will enable stronger interpretations of decoding results by linking them more clearly to the underlying signal characteristics that drive them. HighlightsO_LIWe investigate different decoding paradigms applied to epoched data and characterise the information content available to each over time. C_LIO_LIUnder commonly used instantaneous signal decoding paradigms, sinusoidal components of the evoked response are translated to double their original frequency in decoding accuracy metrics, presenting technical and interpretational challenges. C_LIO_LIWhen instantaneous signal decoding is used, we recommend using low pass filters with a cut-off less than one quarter of the sampling rate to eliminate spurious representational alias artefacts. C_LIO_LIThe interpretational issues associated with instantaneous signal decoding can be resolved with alternative paradigms such as complex spectrum decoding. C_LIO_LIWe show that complex spectrum decoding results in decoding accuracy metrics that are higher, more stable over time, and free of representational aliasing. C_LI

neuroscience↗

bioRxiv2021

Brain stimulation boosts perceptual learning by altering sensory GABAergic plasticity and functional connectivity

Interpreting cluttered scenes --a key skill for successfully interacting with our environment-- relies on our ability to select relevant sensory signals while filtering out noise. Training is known to improve our ability to make these perceptual judgements by altering local processing in sensory brain areas. Yet, the brain-wide network mechanisms that mediate our ability for perceptual learning remain largely unknown. Here, we combine transcranial direct current stimulation (tDCS) with multi-modal brain measures to modulate cortical excitability during training on a signal-in-noise task (i.e. detection of visual patterns in noise) and test directly the link between processing in visual cortex and its interactions with decision-related areas (i.e. posterior parietal cortex). We test whether brain stimulation alters inhibitory processing in visual cortex, as measured by magnetic resonance spectroscopy (MRS) of GABA and functional connectivity between visual and posterior parietal cortex, as measured by resting state functional magnetic resonance imaging (rs-fMRI). We show that anodal tDCS during training results in faster learning and decreased GABA+ during training, before these changes occur for training without stimulation (i.e. sham). Further, anodal tDCS decreases occipito-parietal interactions and time-varying connectivity across the visual cortex. Our findings demonstrate that tDCS boosts learning by accelerating visual GABAergic plasticity and altering interactions between visual and decision-related areas, suggesting that training optimises gain control mechanisms (i.e. GABAergic inhibition) and functional inter-areal interactions to support perceptual learning.

neuroscience↗

bioRxiv2021

Spatiotemporally Resolved Multivariate Pattern Analysis for M/EEG

An emerging goal in neuroscience is tracking what information is represented in brain activity over time as a participant completes some task. Whilst EEG and MEG offer millisecond temporal resolution of how activity patterns emerge and evolve, standard decoding methods present significant barriers to interpretability as they obscure the underlying spatial and temporal activity patterns. We instead propose the use of a generative encoding model framework that simultaneously infers the multivariate spatial patterns of activity and the variable timing at which these patterns emerge on individual trials. An encoding model inversion allows predictions to be made about unseen test data in the same way as in standard decoding methodology. These SpatioTemporally Resolved MVPA (STRM) models can be flexibly applied to a wide variety of experimental paradigms, including classification and regression tasks. We show that these models provide insightful maps of the activity driving predictive accuracy metrics; demonstrate behaviourally meaningful variation in the timing of pattern emergence on individual trials; and achieve predictive accuracies that are either equivalent or surpass those achieved by more widely used methods. This provides a new avenue for investigating the brains representational dynamics and could ultimately support more flexible experimental designs in future. HIGHLIGHTSO_LIWe introduce SpatioTemporally Resolved MVPA (STRM), an approach that explicitly models how successive stages of stimulus processing are distributed in both space and time in M/EEG data. C_LIO_LIWe show that STRM is broadly applicable to diverse types of M/EEG data and outputs meaningful and interpretable maps of how neural representations evolve in space and time at millisecond resolution. C_LIO_LIThe trial-specific deviations in activity pattern timings identified by STRM are not random, but vary systematically with inter-trial differences in behavioural, cognitive and physiological variables. C_LIO_LIThese methods result in predictive accuracy metrics that are mostly equivalent to, or a modest improvement on, conventional methods. C_LI

neuroscience↗

bioRxiv2021

Data and model considerations for estimating time-varying functional connectivity in fMRI

Functional connectivity (FC) in the brain has been shown to exhibit subtle but reliable modulations within a session. One way of estimating time-varying FC is by using state-based models that describe fMRI time series as temporal sequences of states, each with an associated, characteristic pattern of FC. However, the estimation of these models from data sometimes fails to capture changes in a meaningful way, such that the model estimation assigns entire sessions (or the largest part of them) to a single state, therefore failing to capture within-session state modulations effectively; we refer to this phenomenon as the model becoming static, or model stasis. Here, we aim to quantify how the nature of the data and the choice of model parameters affect the models ability to detect temporal changes in FC using both simulated fMRI time courses and resting state fMRI data. We show that large between-subject FC differences can overwhelm subtler within-session modulations, causing the model to become static. Further, the choice of parcellation can also affect the models ability to detect temporal changes. We finally show that the model often becomes static when the number of free parameters that need to be estimated is high and the number of observations available for this estimation is low in comparison. Based on these findings, we derive a set of practical recommendations for time-varying FC studies, in terms of preprocessing, parcellation and complexity of the model. HighlightsO_LITime-varying FC models sometimes fail to detect temporal changes in fMRI data C_LIO_LIBetween- and within-subject FC variability affect model stasis C_LIO_LIThe choice of parcellation affects model stasis in real fMRI data C_LIO_LIThe number of observations and free parameters critically affect model stasis Keywords: fMRI; time-varying FC; Hidden Markov Model (HMM); resting state C_LI

neuroscience↗

bioRxiv2021

Can fMRI functional connectivity index dynamic neural communication?

In order to continuously respond to a changing environment and support self-generating cognition and behaviour, neural communication must be highly flexible and dynamic at the same time than hierarchically organized. While whole-brain fMRI measures have revealed robust yet changing patterns of statistical dependencies between regions, it is not clear whether these statistical patterns --referred to as functional connectivity-- can reflect dynamic large-scale communication in a way that is relevant to human cognition. For functional connectivity to reflect cognition, and therefore actual communication, we propose three necessary conditions: it must span sufficient temporal complexity to support the needs of cognition while still being highly organized so that the system behaves reliably; it must be able to adapt to the current behavioural context; it must exhibit fluctuations at timescales that are compatible with the timescales of cognition. To obtain reliable estimations of time-varying functional connectivity, we developed principal components of connectivity analysis (PCCA), an approach based on applying principal component analysis on multiple runs of a time-varying functional connectivity model. We use PCCA to show that functional connectivity follows low-yet multi-dimensional trajectories that can be reliably measured, and that these trajectories meet the aforementioned criteria. These analyses suggest that these trajectories might index certain aspects of communication between neural populations and support moment-to-moment cognition. Significance StatementfMRI functional connectivity is one of the most widely used metrics in neuroimaging research in both theoretical research and clinical applications. However, this work suffers from a lack of context because we still do not fully understand what fMRI functional connectivity can or cannot reflect biologically and behaviourally. In particular, can it reflect between-region neuronal communication? We develop methods to reliably quantify temporal trajectories of functional connectivity and investigate the nature of these trajectories across different experimental conditions. Using these methods, we demonstrate that functional connectivity exhibits reliable changes that are context-dependent, reflect cognitive complexity, and bear a relationship with cognitive abilities. These conditions show that fMRI functional connectivity could reflect changes in between-region communication above and beyond non-neural factors.

neuroscience↗

bioRxiv2021

Multi-dynamic Modelling Reveals Strongly Time-varying Resting fMRI Correlations

The activity of functional brain networks is responsible for the emergence of time-varying cognition and behaviour. Accordingly, time-varying correlations (Functional Connectivity) in resting fMRI have been shown to be predictive of behavioural traits, and psychiatric and neurological conditions. Typically, methods that measure time varying Functional Connectivity (FC), such as sliding windows approaches, do not separately model when changes occur in the mean activity levels from when changes occur in the FC, therefore conflating these two distinct types of modulation. We show that this can bias the estimation of time-varying FC to appear more stable over time than it actually is. Here, we propose an alternative approach that models changes in the mean brain activity and in the FC as being able to occur at different times to each other. We refer to this method as the Multi-dynamic Adversarial Generator Encoder (MAGE) model, which includes a model of the network dynamics that captures long-range time dependencies, and is estimated on fMRI data using principles of Generative Adversarial Networks. We evaluated the approach across several simulation studies and resting fMRI data from the Human Connectome Project (1003 subjects), as well as from UK Biobank (13301 subjects). Importantly, we find that separating fluctuations in the mean activity levels from those in the FC reveals much stronger changes in FC over time, and is a better predictor of individual behavioural variability Statement of SignificanceO_LIMAGE is multi-dynamic in that it models temporal fluctuations in FC independently from fluctuations in the mean of the activity. C_LIO_LIMAGE reveals stronger changes in FC over time than single-dynamic approaches, such as sliding window correlations. C_LIO_LIMulti-dynamic modelling provides an explanation and a solution as to why resting fMRI FC has previously looked so stable. C_LIO_LIMAGE models fMRI data as a set of reoccurring brain states, and importantly, these states do not have to be binary and mutually exclusive (e.g., multiple states can be active at one time-point). C_LIO_LIMAGE estimated time-varying FC is a better predictor of behavioural variability in the resting-state fMRI data than established methods. C_LI

neuroscience↗