Ben D. Fulcher

Highly Comparative Feature-Based Time-Series Classification

A highly comparative, feature-based approach to time series classification is introduced that uses an extensive database of algorithms to extract thousands of interpretable features from time series. These features are derived from across the scientific time-series analysis literature, and include summaries of time series in terms of their correlation structure, distribution, entropy, stationarity, scaling properties, and fits to a range of time-series models. After computing thousands of features for each time series in a training set, those that are most informative of the class structure are selected using greedy forward feature selection with a linear classifier. The resulting feature-based classifiers automatically learn the differences between classes using a reduced number of time-series properties, and circumvent the need to calculate distances between time series. Representing time series in this way results in orders of magnitude of dimensionality reduction, allowing the method to perform well on very large data sets containing long time series or time series of different lengths. For many of the data sets studied, classification performance exceeded that of conventional instance-based classifiers, including one nearest neighbor classifiers using euclidean distances and dynamic time warping and, most importantly, the features selected provide an understanding of the properties of the data set, insight that can guide further scientific investigation.

A transcriptional signature of hub connectivity in the mouse connectome

Significance Some brain regions are highly connected with other areas, designating them as network hubs. These hubs are also heavily interconnected with each other, forming a dense core that integrates information across different neural systems. Here, we show that the functionally important projections linking hub areas of the mouse brain have a distinct genetic signature that is characterized by the tightly coupled expression of genes regulating the synthesis and metabolism of ATP, the primary energy source for neural activity. Our findings establish a direct link between molecular function and the large-scale organization of neuronal connectivity and suggest that coordinated gene expression between hub areas is closely related to the metabolic demands of these highly active and functionally important regions. Connectivity is not distributed evenly throughout the brain. Instead, it is concentrated on a small number of highly connected neural elements that act as network hubs. Across different species and measurement scales, these hubs show dense interconnectivity, forming a core or “rich club” that integrates information across anatomically distributed neural systems. Here, we show that projections between connectivity hubs of the mouse brain are both central (i.e., they play an important role in neural communication) and costly (i.e., they extend over long anatomical distances) aspects of network organization that carry a distinctive genetic signature. Analyzing the neuronal connectivity of 213 brain regions and the transcriptional coupling, across 17,642 genes, between each pair of regions, we find that coupling is highest for pairs of connected hubs, intermediate for links between hubs and nonhubs, and lowest for connected pairs of nonhubs. The high transcriptional coupling associated with hub connectivity is driven by genes regulating the oxidative synthesis and metabolism of ATP—the primary energetic currency of neuronal communication. This genetic signature contrasts that identified for neuronal connectivity in general, which is driven by genes regulating neuronal, synaptic, and axonal structure and function. Our findings establish a direct link between molecular function and the large-scale topology of neuronal connectivity, showing that brain hubs display a tight coordination of gene expression, often over long anatomical distances, that is intimately related to the metabolic requirements of these highly active network elements.

Highly comparative time-series analysis: the empirical structure of time series and their methods

The process of collecting and organizing sets of observations represents a common theme throughout the history of science. However, despite the ubiquity of scientists measuring, recording and analysing the dynamics of different processes, an extensive organization of scientific time-series data and analysis methods has never been performed. Addressing this, annotated collections of over 35 000 real-world and model-generated time series, and over 9000 time-series analysis algorithms are analysed in this work. We introduce reduced representations of both time series, in terms of their properties measured by diverse scientific methods, and of time-series analysis methods, in terms of their behaviour on empirical time series, and use them to organize these interdisciplinary resources. This new approach to comparing across diverse scientific data and methods allows us to organize time-series datasets automatically according to their properties, retrieve alternatives to particular analysis methods developed in other scientific disciplines and automate the selection of useful methods for time-series classification and regression tasks. The broad scientific utility of these tools is demonstrated on datasets of electroencephalograms, self-affine time series, heartbeat intervals, speech signals and others, in each case contributing novel analysis techniques to the existing literature. Highly comparative techniques that compare across an interdisciplinary literature can thus be used to guide more focused research in time-series analysis for applications across the scientific disciplines.

2015 Brainhack Proceedings

Table of contentsI1 Introduction to the 2015 Brainhack ProceedingsR. Cameron Craddock, Pierre Bellec, Daniel S. Margules, B. Nolan Nichols, Jörg P. PfannmöllerA1 Distributed collaboration: the case for the enhancement of Brainspell’s interfaceAmanPreet Badhwar, David Kennedy, Jean-Baptiste Poline, Roberto ToroA2 Advancing open science through NiDataBen Cipollini, Ariel RokemA3 Integrating the Brain Imaging Data Structure (BIDS) standard into C-PACDaniel Clark, Krzysztof J. Gorgolewski, R. Cameron CraddockA4 Optimized implementations of voxel-wise degree centrality and local functional connectivity density mapping in AFNIR. Cameron Craddock, Daniel J. ClarkA5 LORIS: DICOM anonymizerSamir Das, Cécile Madjar, Ayan Sengupta, Zia MohadesA6 Automatic extraction of academic collaborations in neuroimagingSebastien DeryA7 NiftyView: a zero-footprint web application for viewing DICOM and NIfTI filesWeiran DengA8 Human Connectome Project Minimal Preprocessing Pipelines to NipypeEric Earl, Damion V. Demeter, Kate Mills, Glad Mihai, Luka Ruzic, Nick Ketz, Andrew Reineberg, Marianne C. Reddan, Anne-Lise Goddings, Javier Gonzalez-Castillo, Krzysztof J. GorgolewskiA9 Generating music with resting-state fMRI dataCaroline Froehlich, Gil Dekel, Daniel S. Margulies, R. Cameron CraddockA10 Highly comparable time-series analysis in NitimeBen D. FulcherA11 Nipype interfaces in CBRAINTristan Glatard, Samir Das, Reza Adalat, Natacha Beck, Rémi Bernard, Najmeh Khalili-Mahani, Pierre Rioux, Marc-Étienne Rousseau, Alan C. EvansA12 DueCredit: automated collection of citations for software, methods, and dataYaroslav O. Halchenko, Matteo Visconti di Oleggio CastelloA13 Open source low-cost device to register dog’s heart rate and tail movementRaúl Hernández-Pérez, Edgar A. Morales, Laura V. CuayaA14 Calculating the Laterality Index Using FSL for Stroke Neuroimaging DataKaori L. Ito, Sook-Lei LiewA15 Wrapping FreeSurfer 6 for use in high-performance computing environmentsHans J. JohnsonA16 Facilitating big data meta-analyses for clinical neuroimaging through ENIGMA wrapper scriptsErik Kan, Julia Anglin, Michael Borich, Neda Jahanshad, Paul Thompson, Sook-Lei LiewA17 A cortical surface-based geodesic distance package for PythonDaniel S Margulies, Marcel Falkiewicz, Julia M HuntenburgA18 Sharing data in the cloudDavid O’Connor, Daniel J. Clark, Michael P. Milham, R. Cameron CraddockA19 Detecting task-based fMRI compliance using plan abandonment techniquesRamon Fraga Pereira, Anibal Sólon Heinsfeld, Alexandre Rosa Franco, Augusto Buchweitz, Felipe MeneguzziA20 Self-organization and brain functionJörg P. Pfannmöller, Rickson Mesquita, Luis C.T. Herrera, Daniela DenticoA21 The Neuroimaging Data Model (NIDM) APIVanessa Sochat, B Nolan NicholsA22 NeuroView: a customizable browser-base utilityAnibal Sólon Heinsfeld, Alexandre Rosa Franco, Augusto Buchweitz, Felipe MeneguzziA23 DIPY: Brain tissue classificationJulio E. Villalon-Reina, Eleftherios Garyfallidis

Quantitative modelling of sleep dynamics

Arousal is largely controlled by the ascending arousal system of the hypothalamus and brainstem, which projects to the corticothalamic system responsible for electroencephalographic (EEG) signatures of sleep. Quantitative physiologically based modelling of brainstem dynamics theory is described here, using realistic parameters, and links to EEG are outlined. Verification against a wide range of experimental data is described, including arousal dynamics under normal conditions, sleep deprivation, stimuli, stimulants and jetlag, plus key features of wake and sleep EEGs.

Quantitative physiologically based modeling of subjective fatigue during sleep deprivation

A quantitative physiologically based model of the sleep-wake switch is used to predict variations in subjective fatigue-related measures during total sleep deprivation. The model includes the mutual inhibition of the sleep-active neurons in the hypothalamic ventrolateral preoptic area (VLPO) and the wake-active monoaminergic brainstem populations (MA), as well as circadian and homeostatic drives. We simulate sleep deprivation by introducing a drive to the MA, which we call wake effort, to maintain the system in a wakeful state. Physiologically this drive is proposed to be afferent from the cortex or the orexin group of the lateral hypothalamus. It is hypothesized that the need to exert this effort to maintain wakefulness at high homeostatic sleep pressure correlates with subjective fatigue levels. The models output indeed exhibits good agreement with existing clinical time series of subjective fatigue-related measures, supporting this hypothesis. Subjective fatigue, adrenaline, and body temperature variations during two 72h sleep deprivation protocols are reproduced by the model. By distinguishing a motivation-dependent orexinergic contribution to the wake-effort drive, the model can be extended to interpret variation in performance levels during sleep deprivation in a way that is qualitatively consistent with existing, clinically derived results. The example of sleep deprivation thus demonstrates the ability of physiologically based sleep modeling to predict psychological measures from the underlying physiological interactions that produce them.

Incorporation of caffeine into a quantitative model of fatigue and sleep

A recent physiologically based model of human sleep is extended to incorporate the effects of caffeine on sleep-wake timing and fatigue. The model includes the sleep-active neurons of the hypothalamic ventrolateral preoptic area (VLPO), the wake-active monoaminergic brainstem populations (MA), their interactions with cholinergic/orexinergic (ACh/Orx) input to MA, and circadian and homeostatic drives. We model two effects of caffeine on the brain due to competitive antagonism of adenosine (Ad): (i) a reduction in the homeostatic drive and (ii) an increase in cholinergic activity. By comparing the model output to experimental data, constraints are determined on the parameters that describe the action of caffeine on the brain. In accord with experiment, the ranges of these parameters imply significant variability in caffeine sensitivity between individuals, with caffeines effectiveness in reducing fatigue being highly dependent on an individuals tolerance, and past caffeine and sleep history. Although there are wide individual differences in caffeine sensitivity and thus in parameter values, once the model is calibrated for an individual it can be used to make quantitative predictions for that individual. A number of applications of the model are examined, using exemplar parameter values, including: (i) quantitative estimation of the sleep loss and the delay to sleep onset after taking caffeine for various doses and times; (ii) an analysis of the systems stable states showing that the wake state during sleep deprivation is stabilized after taking caffeine; and (iii) comparing model output successfully to experimental values of subjective fatigue reported in a total sleep deprivation study examining the reduction of fatigue with caffeine. This model provides a framework for quantitatively assessing optimal strategies for using caffeine, on an individual basis, to maintain performance during sleep deprivation.

Mammalian rest/activity patterns explained by physiologically based modeling

Circadian rhythms are fundamental to life. In mammals, these rhythms are generated by pacemaker neurons in the suprachiasmatic nucleus (SCN) of the hypothalamus. The SCN is remarkably consistent in structure and function between species, yet mammalian rest/activity patterns are extremely diverse, including diurnal, nocturnal, and crepuscular behaviors. Two mechanisms have been proposed to account for this diversity: (i) modulation of SCN output by downstream nuclei, and (ii) direct effects of light on activity. These two mechanisms are difficult to disentangle experimentally and their respective roles remain unknown. To address this, we developed a computational model to simulate the two mechanisms and their influence on temporal niche. In our model, SCN output is relayed via the subparaventricular zone (SPZ) to the dorsomedial hypothalamus (DMH), and thence to ventrolateral preoptic nuclei (VLPO) and lateral hypothalamus (LHA). Using this model, we generated rich phenotypes that closely resemble experimental data. Modulation of SCN output at the SPZ was found to generate a full spectrum of diurnal-to-nocturnal phenotypes. Intriguingly, we also uncovered a novel mechanism for crepuscular behavior: if DMH/VLPO and DMH/LHA projections act cooperatively, daily activity is unimodal, but if they act competitively, activity can become bimodal. In addition, we successfully reproduced diurnal/nocturnal switching in the rodent Octodon degu using coordinated inversions in both masking and circadian modulation. Finally, the model correctly predicted the SCN lesion phenotype in squirrel monkeys: loss of circadian rhythmicity and emergence of ∼4-h sleep/wake cycles. In capturing these diverse phenotypes, the model provides a powerful new framework for understanding rest/activity patterns and relating them to underlying physiology. Given the ubiquitous effects of temporal organization on all aspects of animal behavior and physiology, this study sheds light on the physiological changes required to orchestrate adaptation to various temporal niches.

A Physiologically Based Model of Orexinergic Stabilization of Sleep and Wake

The orexinergic neurons of the lateral hypothalamus (Orx) are essential for regulating sleep-wake dynamics, and their loss causes narcolepsy, a disorder characterized by severe instability of sleep and wake states. However, the mechanisms through which Orx stabilize sleep and wake are not well understood. In this work, an explanation of the stabilizing effects of Orx is presented using a quantitative model of important physiological connections between Orx and the sleep-wake switch. In addition to Orx and the sleep-wake switch, which is composed of mutually inhibitory wake-active monoaminergic neurons in brainstem and hypothalamus (MA) and the sleep-active ventrolateral preoptic neurons of the hypothalamus (VLPO), the model also includes the circadian and homeostatic sleep drives. It is shown that Orx stabilizes prolonged waking episodes via its excitatory input to MA and by relaying a circadian input to MA, thus sustaining MA firing activity during the circadian day. During sleep, both Orx and MA are inhibited by the VLPO, and the subsequent reduction in Orx input to the MA indirectly stabilizes sustained sleep episodes. Simulating a loss of Orx, the model produces dynamics resembling narcolepsy, including frequent transitions between states, reduced waking arousal levels, and a normal daily amount of total sleep. The model predicts a change in sleep timing with differences in orexin levels, with higher orexin levels delaying the normal sleep episode, suggesting that individual differences in Orx signaling may contribute to chronotype. Dynamics resembling sleep inertia also emerge from the model as a gradual sleep-to-wake transition on a timescale that varies with that of Orx dynamics. The quantitative, physiologically based model developed in this work thus provides a new explanation of how Orx stabilizes prolonged episodes of sleep and wake, and makes a range of experimentally testable predictions, including a role for Orx in chronotype and sleep inertia.

An evaluation of the efficacy, reliability, and sensitivity of motion correction strategies for resting-state functional MRI

&NA; Estimates of functional connectivity derived from resting‐state functional magnetic resonance imaging (rs‐fMRI) are sensitive to artefacts caused by in‐scanner head motion. This susceptibility has motivated the development of numerous denoising methods designed to mitigate motion‐related artefacts. Here, we compare popular retrospective rs‐fMRI denoising methods, such as regression of head motion parameters and mean white matter (WM) and cerebrospinal fluid (CSF) (with and without expansion terms), aCompCor, volume censoring (e.g., scrubbing and spike regression), global signal regression and ICA‐AROMA, combined into 19 different pipelines. These pipelines were evaluated across five different quality control benchmarks in four independent datasets associated with varying levels of motion. Pipelines were benchmarked by examining the residual relationship between in‐scanner movement and functional connectivity after denoising; the effect of distance on this residual relationship; whole‐brain differences in functional connectivity between high‐ and low‐motion healthy controls (HC); the temporal degrees of freedom lost during denoising; and the test‐retest reliability of functional connectivity estimates. We also compared the sensitivity of each pipeline to clinical differences in functional connectivity in independent samples of people with schizophrenia and obsessive‐compulsive disorder. Our results indicate that (1) simple linear regression of regional fMRI time series against head motion parameters and WM/CSF signals (with or without expansion terms) is not sufficient to remove head motion artefacts; (2) aCompCor pipelines may only be viable in low‐motion data; (3) volume censoring performs well at minimising motion‐related artefact but a major benefit of this approach derives from the exclusion of high‐motion individuals; (4) while not as effective as volume censoring, ICA‐AROMA performed well across our benchmarks for relatively low cost in terms of data loss; (5) the addition of global signal regression improved the performance of nearly all pipelines on most benchmarks, but exacerbated the distance‐dependence of correlations between motion and functional connectivity; and (6) group comparisons in functional connectivity between healthy controls and schizophrenia patients are highly dependent on preprocessing strategy. We offer some recommendations for best practice and outline simple analyses to facilitate transparent reporting of the degree to which a given set of findings may be affected by motion‐related artefact. Graphical abstract Figure. No caption available. HighlightsWe examine 19 denoising pipelines for resting‐state fMRI across 4 datasets.No single method offers perfect motion control.Censoring and ICA‐AROMA pipelines perform well across most benchmarks.Pipeline choice impacts case‐control differences in functional connectivity.