Joint dynamic causal modeling and biophysical modeling to enable multiscale brain network functional modeling

Abstract

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for combined dynamic causal modeling and biophysical modeling of brain function are provided. In particular, the disclosed methods of brain function modeling can be used to integrate brain function measurements by two or more methods, such as functional neuroimaging and electrophysiological techniques. The present invention uses sequential model fitting to improve modeling accuracy to generate more comprehensive brain neuronal circuit models.

Description

[0001] Statement regarding federally sponsored research or development

[0002] This invention patent has received government support, including grants from the National Institutes of Health (NIH) under projects MH114227 and NS091461. The government owns certain rights to this invention. Technical Field

[0003] This invention relates to the combined dynamic causal modeling and biophysical modeling to achieve multi-scale brain network functional modeling. Background Technology

[0004] Brain function has been studied using a variety of experimental techniques at different spatiotemporal sampling rates and contrasts. Some of these techniques include functional magnetic resonance imaging (fMRI), electrophysiological techniques, and optical imaging. While each technique has its own unique advantages in terms of its ability to probe brain function at different scales, none are comprehensive enough. Therefore, it is helpful to find a means to use different measurements together to construct a comprehensive model of brain function. Summary of the Invention

[0005] This invention provides methods, systems, and apparatus for combined dynamic causal modeling and biophysical modeling of brain function, including computer programs encoded on a computer storage medium. In particular, the methods can be used to integrate brain function measurements through two or more methods, such as functional neuroimaging and electrophysiological techniques. This invention uses sequential model fitting to improve modeling accuracy, generating more comprehensive models of brain neuronal circuits.

[0006] In one aspect, a brain function modeling method is provided, the method comprising: a) performing functional neuroimaging of neural activity in one or more regions of interest in the brain of a subject; b) acquiring experimental electrophysiological data of one or more neurons in the one or more regions of interest in the subject's brain; c) fitting the functional neuroimaging data of the neural activity to a dynamic causal model (DCM); d) calculating estimates of inter-regional and intra-regional connectivity strength using the dynamic causal model; e) using the inter-regional and intra-regional connectivity strength estimates to generate a biophysical model simulating synthetic electrophysiological data; f) comparing the synthetic electrophysiological data with the experimental electrophysiological data; and g) iteratively adjusting the biophysical model until the model converges with the experimental electrophysiological data.

[0007] Exemplary functional neuroimaging techniques that can be used to implement the target method include, but are not limited to, functional magnetic resonance imaging (fMRI), positron emission tomography (PET), functional near-infrared spectroscopy (fNIRS), single-photon emission computed tomography (SPECT), and functional ultrasound imaging (fUS). In some embodiments, fMRI is used to detect blood oxygen level dependent (BOLD) signals.

[0008] Exemplary electrophysiological techniques that can be used to implement the target method include, but are not limited to, electroencephalography (EEG), magnetoencephalography (MEG), and patch-clamp technique.

[0009] In some embodiments, the method further includes using optogenetics to use light to excite or inhibit one or more selected target neurons. In some embodiments, the fMRI experimental data are optogenetic fMRI (ofMRI) experimental data.

[0010] Brain function modeling can be performed on any region of the brain using the method of the present invention. In some embodiments, the one or more regions of interest are located in the cerebrum, cerebellum, or brainstem regions of the brain. Regions of interest may include, but are not limited to, the basal ganglia, striatum, medulla oblongata, pons, midbrain, medulla oblongata, hypothalamus, thalamus, epithalamus, amygdala, superior colliculus, cerebral cortex, neocortex, xenogeneic cortex, hippocampus, claustrum, olfactory bulb, frontal lobe, temporal lobe, parietal lobe, occipital lobe, caudate putamen, lateral part of globus pallidus, medial part of globus pallidus, subthalamic nucleus, substantia nigra, thalamus, and motor cortex regions of the brain.

[0011] Functional neuroimaging and electrophysiological data of any type of neuron can be acquired, including but not limited to unipolar neurons, bipolar neurons, multipolar neurons, Golgi type I neurons, Golgi type II neurons, axonal neurons, pseudounipolar neurons, interneurons, motor neurons, sensory neurons, afferent neurons, efferent neurons, cholinergic neurons, GABAergic neurons, glutamatergic neurons, dopaminergic neurons, serotonergic neurons, histaminergic neurons, Purkinje cells, polyspinous projection neurons, Runshaw cells, and granule cells or combinations thereof.

[0012] In some embodiments, the method further includes optical imaging or multiphoton microscopy of one or more regions of interest in the subject's brain.

[0013] In some embodiments, the subject suffers from a neurological disease or condition.

[0014] In some embodiments, the method further includes detecting changes in the subject's brain function compared to that of a control subject, based on the brain function modeling.

[0015] In some embodiments, the method further includes identifying neural circuits involving pathological changes associated with the neurological disease or condition.

[0016] In some embodiments, the method further includes detecting changes in the subject's brain function after treatment compared to before treatment, based on brain function modeling of the subject before and after treatment for the neurological disease or condition.

[0017] In some embodiments, the functional neuroimaging and / or electrophysiological data are acquired while the subject performs a task or responds to a stimulus. In some embodiments, the method further includes identifying neural circuits involved in performing the task or responding to the stimulus. In some embodiments, the functional neuroimaging data and the experimental electrophysiological data are acquired while the subject is at rest (e.g., for comparison with data acquired while the subject performs the task or responds to a stimulus).

[0018] In some embodiments, the DCM is a typical DCM, a random DCM, a spectral DCM, or a dynamic DCM.

[0019] In another aspect, a computer-implemented method for modeling brain function is provided, the computer performing steps comprising: a) receiving functional neuroimaging data and experimental electrophysiological data of one or more brain regions of interest from a subject; b) fitting the functional neuroimaging data to a dynamic causal model; c) calculating inter-regional and intra-regional connectivity strength estimates using the dynamic causal model; d) generating a biophysical model simulating synthetic electrophysiological data using the inter-regional and intra-regional connectivity strength estimates; e) comparing the synthetic electrophysiological data with the experimental electrophysiological data; f) iteratively adjusting the biophysical model until the model converges with the experimental electrophysiological data; and g) displaying information related to the brain function modeling.

[0020] In some embodiments, the functional neuroimaging data includes functional magnetic resonance imaging (fMRI) data, positron emission tomography (PET) data, functional near-infrared spectroscopy (fNIRS) data, single-photon emission computed tomography (SPECT) data, or functional ultrasound imaging (fUS) data.

[0021] In some embodiments, the experimental electrophysiological data includes electroencephalography (EEG) data, magnetoencephalography (MEG) data, or patch-clamp data.

[0022] In some embodiments, the subject suffers from a neurological disease or condition, wherein the displayed information includes a list of one or more neural circuits involving pathological changes associated with the neurological disease or condition.

[0023] In some embodiments, the functional neuroimaging data and the experimental electrophysiological data are acquired while the subject performs a task or responds to a stimulus, wherein the displayed information includes a list of one or more neural circuits involved in performing the task or responding to the stimulus. In some embodiments, the functional neuroimaging data and the experimental electrophysiological data are acquired while the subject is at rest (e.g., for comparison with data acquired while the subject performs the task or responds to a stimulus).

[0024] On the other hand, a system is provided for performing brain function modeling using the computer-implemented method described herein, the system comprising: a) a storage component for storing data, wherein the storage component has instructions for performing brain function modeling based on analysis of the functional neuroimaging data and experimental electrophysiological data stored therein; b) a computer processor for processing the functional neuroimaging data and experimental electrophysiological data using one or more algorithms, wherein the computer processor is coupled to the storage component and configured to execute instructions stored in the storage component to receive the input functional neuroimaging data and experimental electrophysiological data and analyze the data according to the computer-implemented method described herein; and c) a display component for displaying the information relating to the brain function modeling.

[0025] In another aspect, a non-transitory computer-readable medium is provided, which contains program instructions that, when executed by a processor in a computer, cause the processor to perform the computer-implemented method described herein.

[0026] On the other hand, a kit is provided that includes the aforementioned non-transitory computer-readable medium and instructions for modeling brain function using functional neuroimaging data and experimental electrophysiological data. Attached Figure Description

[0027] Figure 1Modeling using fMRI BOLD data from seven regions of interest is presented. Experimental BOLD data from all subjects were averaged. Light stimulation was performed at 20 Hz, with each cycle consisting of 20 s on-20 s off-20 s on-20 s, repeated 6 times. The regions of interest were the caudate putamen (CPu), lateral globus pallidus (GPe), medial globus pallidus (GPi), subthalamic nucleus (STN), substantia nigra (SN), thalamus (THL), and motor cortex (MCX).

[0028] Figure 2 Modeling using electrophysiological data from four regions of interest is presented.

[0029] Figure 3 The prior generative network model is presented. We first performed a DCM analysis using the corresponding data for each subject in order to reveal the efficient connectivity estimates between and within each region using the prior generative network model.

[0030] Figure 4 The analysis results are presented. Connection strength estimates are in Hz. The magnitudes of the parameters in this figure reflect the influence of the rate of change of neuronal activity in one region on the neuronal activity in another region. These estimates are then used to influence neuronal population design in biophysical modeling.

[0031] Figures 5A-5D Electrophysiological data simulated using a biophysical model are presented. In constructing the biophysical model, we used the same mouse basal ganglia-thalamic cortex circuit structure model and modified neuronal model as used in our DCM analysis (Hill and Tononi, Journal of Neurophysiology, 93(3): 1671-1698, 2005; the contents of which are incorporated herein by reference). While in the DCM analysis, one node represents a region of interest, in the biophysical model, we used 100 neuronal units for a region of interest to better investigate neuronal-level mechanisms. The biophysical model was constructed without making any assumptions; that is, we derived the model directly from the connectivity parameters in the DCM results.

[0032] Figure 6 It demonstrates simulated activity in the thalamus.

[0033] Figure 7 This demonstrates how a thalamic neuron fires between GPi inhibitory inputs. Detailed Implementation

[0034] This invention provides methods, systems, and apparatus for combined dynamic causal modeling and biophysical modeling of brain function, including computer programs encoded on a computer storage medium. In particular, the disclosed brain function modeling methods can be used to integrate brain function measurements through two or more methods, such as functional magnetic resonance imaging (fMRI) and electrophysiological techniques. This invention uses sequential model fitting to improve modeling accuracy, generating more comprehensive models of brain neuron circuits.

[0035] Before describing the compositions, methods, and kits of the present invention, it should be understood that the invention is not limited to the specific methods or compositions described, as differences will certainly exist in actual implementation. It should also be understood that the terminology used herein is for describing specific embodiments only and is not intended to limit the inventive concept; the scope of the invention will be defined only by the appended claims.

[0036] When a numerical range is provided, it should be understood that each intermediate value between the upper and lower limits of the range is explicitly disclosed. Unless the context explicitly specifies otherwise, each intermediate value should be as low as one-tenth of the lower limit unit. This invention covers every smaller range between any stated value or intervention value within the range and any other stated value or intervention value within the range. The upper and lower limits of these smaller ranges may be independently included in or excluded from the range, and this invention also covers ranges with one limit, an unlimited value, or two limits included within the smaller ranges, while complying with any specifically excluded limits within the range. Where the range includes one or two limits, ranges excluding any one or both of the included limits are also included in this invention.

[0037] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. While similar or equivalent methods and materials may also be used in the implementation or testing of this invention, some potential and preferred methods and materials are described below. All publications mentioned herein are incorporated by reference to disclose and describe methods and / or materials relating to the cited publications. It should be understood that, in the event of conflict, the content of this invention shall supersede any disclosure in the cited publications.

[0038] Upon reading this invention, the following will be apparent to those skilled in the art: each individual embodiment described and listed herein has hierarchical components and features that can be quickly decomposed or combined with features of any of the other embodiments without departing from the scope and spirit of the invention. Any of the stated methods may be implemented in the order of the stated events or in any other logically possible order.

[0039] It is important to note that the singular forms “a,” “an,” and “the” used herein and in the appended claims include plural references unless the context clearly indicates otherwise. Thus, for example, “a neuron” refers to a plurality of such neurons, while “the measurement” refers to one or more measurements and equivalents known to those skilled in the art, and so on.

[0040] The publications discussed herein are only those disclosed prior to the filing date of this patent. Nothing herein is to be construed as an admission that this invention is not entitled to any publication prior to such publications due to prior inventions. Furthermore, the publication dates provided may differ from the actual publication dates and may require separate verification.

[0041] definition

[0042] The term “about,” especially when referring to a given quantity, is intended to cover a deviation of plus or minus five percent.

[0043] The term "connectivity" refers to causal, functional, or directional coupling between brain regions and / or groups of neurons. Neural processing may involve integrative networks of several regions within the brain as well as functional connectivity between spatially distant brain regions. Functional connectivity analysis may involve detecting neural interactions between and within regions, which relate to brain activity during specific cognitive or motor tasks or spontaneous brain activity during resting periods. Dynamic causal modeling of brain function can be used to estimate the "connectivity strength" between and within neural regions.

[0044] The terms “individual,” “subject,” “host,” and “patient” are used interchangeably in this document and refer to any subject with a brain, including invertebrates and vertebrates such as (but not limited to) arthropods (e.g., insects, crustaceans, arachnids), cephalopods (e.g., octopuses, squid), amphibians (e.g., frogs, salamanders, caecilians), fish, reptiles (e.g., turtles, crocodiles, snakes, caecilians, lizards, monitor lizards), mammals (including human and non-human mammals, such as non-human primates, including chimpanzees and other apes and monkeys); laboratory animals such as mice, rats, rabbits, hamsters, guinea pigs, and chinchillas; domesticated animals such as dogs and cats; farm animals such as sheep, goats, pigs, horses, and cattle; and birds such as poultry, wild birds, and game birds, including chickens, turkeys and other quails, ducks, and geese. In some cases, the methods of the present invention can be used for the development of laboratory animals, veterinary applications and animal disease models, including but not limited to rodents, including mice, rats and hamsters; primates and transgenic animals.

[0045] It will be apparent to those skilled in the art that various changes and modifications can be made without departing from the spirit or scope of the invention.

[0046] Brain function modeling using dynamic causal modeling combined with biophysical modeling

[0047] This invention provides a method for generating improved brain function models using dynamic causal modeling combined with biophysical modeling. Specifically, the proposed method can be used to integrate brain function measurements using two or more techniques, such as functional neuroimaging and electrophysiological techniques. For example, functional neuroimaging can be performed to image neural activity in one or more regions of interest within a subject's brain. Additionally, electrophysiological measurements can be performed on one or more neurons in the same brain region of interest. First, functional neuroimaging data of brain neural activity is fitted to a dynamic causal model (DCM) used to calculate estimates of inter-regional and intra-regional connectivity strengths. Next, a biophysical model is constructed using the inter-regional and intra-regional connectivity strength estimates calculated through the DCM analysis. This biophysical model is used to simulate synthetic electrophysiological data, which can be compared with experimental electrophysiological data of neurons in the acquired brain region of interest. The biophysical model can be iteratively adjusted until the model converges with the experimental electrophysiological data. In this method, sequential model fitting is used to improve modeling accuracy, resulting in a more accurate and comprehensive model of brain neuron circuitry.

[0048] The methods described herein are applicable to modeling brain function in any subject with a brain, including invertebrates and vertebrates such as (but not limited to) arthropods (e.g., insects, crustaceans, arachnids), cephalopods (e.g., octopuses, squid), amphibians (e.g., frogs, salamanders, caecilians), fish, reptiles (e.g., turtles, crocodiles, snakes, caecilians, lizards, monitor lizards), mammals (including humans and non-human mammals, such as non-human primates, including chimpanzees and other apes and monkeys); laboratory animals such as mice, rats, rabbits, hamsters, guinea pigs, and chinchillas; domesticated animals such as dogs and cats; farm animals such as sheep, goats, pigs, horses, and cattle; and birds such as poultry, wild birds, and game birds, including chickens, turkeys and other quails, ducks, and geese. In some cases, the methods of the present invention can be used for the development of laboratory animals, veterinary applications, and animal disease models, including but not limited to rodents, including mice, rats, and hamsters; primates, and transgenic animals.

[0049] Exemplary functional neuroimaging techniques that can be used to implement the target method include, but are not limited to, functional magnetic resonance imaging (fMRI), positron emission tomography (PET), functional near-infrared spectroscopy (fNIRS), single-photon emission computed tomography (SPECT), and functional ultrasound imaging (fUS). These functional neuroimaging techniques measure local changes in cerebral blood flow and changes in blood components associated with neural activity. Functional neuroimaging is used for non-invasive detection of brain activity patterns associated with specific stimuli or tasks.

[0050] In some embodiments, fMRI is used to monitor temporal changes in blood flow in relation to variations in brain activity levels. When a specific region of the brain is used, blood flow increases with neuronal activation. By detecting oxygen-dependent (BOLD) signals, fMRI can be used to image changes in brain activity. For descriptions of fMRI and methods of imaging brain activity using fMRI, see, for example, Ogawa et al. (1990), Magnetic Resonance in Medicine, 14(1): 68-78; Kim et al. (2002), New Insights in Neurobiology, 12(5): 607-15; Kim et al. (2012), Journal of Brain Blood Flow and Metabolism, 32(7): 1188-206; Bandettini (2012), Neuroimaging, 62(2): 575-88; Zarghami et al. (2020), Neuroimaging, 207: 116453; Logothetis, NK (June 12, 2008), Logothetis et al. (2008), Nature, 453(7197): 869-78; and Logothetis et al. (2001), Nature, 412 (6843): 150-157; its contents are incorporated herein by reference.

[0051] In some embodiments, PET is used to monitor changes in blood flow associated with changes in neural activity. PET uses a radioactive tracer that emits positrons for imaging. Brain activity can be imaged using PET by detecting changes in blood flow, which can be measured indirectly, such as using an oxygen-15 tracer. Areas with higher levels of radioactivity are associated with increased brain activity. For a description of PET and methods of imaging brain activity using PET, see, for example, Hiura et al. (2014), Journal of Brain Blood Flow and Metabolism, 34(3): 389-96; Baron et al. (2012), Neuroimaging, 61(2): 492-504; Law (2007), Danish Medical Bulletin, November 2007, 54(4): 289-305; Ramsey et al. (1996), Journal of Brain Blood Flow and Metabolism, 16(5): 755-64; the contents of which are incorporated herein by reference.

[0052] In some embodiments, SPECT imaging is used for brain imaging. Like PET, SPECT uses a radioactive tracer to detect changes in blood flow, but the tracer it uses emits gamma rays that can be detected by a gamma camera. Changes in blood flow can be detected (e.g., using technetium). 99m Tc-Essamexamic or 99m SPECT imaging of brain activity was performed using Tc-D,L-hexamethylene-propanethime. For descriptions of SPECT and methods of using SPECT to image brain activity, please see, for example, Cuocolo et al. (2018), International Review of Neurobiology, 141:77-96; Andersen (1989), Review of Cerebrovascular and Brain Metabolism, 1(4):288-318; Matsuda (2001), Annals of Nuclear Medicine, 15(2):85-92; Gonul et al. (2009), International Review of Psychiatry, 21(4):323-35; the contents of which are incorporated herein by reference.

[0053] In some embodiments, fNIRS is used to monitor changes in blood components near a neural event. In particular, fNIRS can be used to detect changes in oxyhemoglobin and deoxyhemoglobin levels using near-infrared light. Based on the difference in absorption spectra between oxyhemoglobin and deoxyhemoglobin, fNIRS can be used to measure relative changes in hemoglobin concentration. Cerebral hemodynamic responses are associated with brain activation or inactivation. For a description of fNIRS and methods for imaging brain activity using fNIRS, see, for example, Tachtsidis et al. (2020), Annual of the New York Academy of Sciences, 1464(1): 5-29; Ferrari et al. (2012), Neuroimaging, 63(2): 921-35; Scholkmann et al. (2014), Neuroimaging, 85 Pt 1: 6-27; Kim et al. (2017), Molecular & Cellular, 40(8): 523-532; the contents of which are incorporated herein by reference.

[0054] In some embodiments, fUS is used to monitor local changes in cerebral blood volume associated with changes in neural activity. For a description of fUS and methods of imaging brain activity using fUS, see, for example, Bleton et al. (2016), BMC Medical Imaging, 12; 16:22; Raaij et al. (2012), Neuroimaging, 63(3):1030-7; the contents of which are incorporated herein by reference.

[0055] Electrophysiological techniques are used to measure electrical properties in the brain, typically changes in voltage or current in neurons. Exemplary electrophysiological techniques that can be used to implement the target method include, but are not limited to, electroencephalography (EEG), magnetoencephalography (MEG), and patch-clamp techniques. These electrophysiological techniques can be used to identify specific types of neurons involved in a neural network and to measure neuron-specific activity changes associated with brain responses.

[0056] In some embodiments, EEG is used to record the electrical activity of neurons in the brain. EEG can be performed non-invasively using electrodes placed on the scalp. EEG measurements have the advantage of high temporal resolution and can detect changes in the electrical activity of the brain on a millisecond timescale. For a description of EEG and methods of recording the electrical activity of the brain using EEG, see, for example, Niedermeyer et al. (2004), Electroencephalography: Principles, Clinical Applications and Related Fields, Lippincott Williams & Wilkins; Jackson et al. (2014), Psychophysiology, 51(11): 1061-71; Khanna et al. (2015), Neuroscience and Biobehavioral Review, 49: 105-13; Feyissa et al. (2019), Handbook of Clinical Neurology, 160: 103-124; Beres et al. (2017), Applied Psychophysiology and Biofeedback, 42(4): 247-255; the contents of which are incorporated herein by reference.

[0057] In some embodiments, MEGs are used to record magnetic fields generated by electrical currents generated in the brain. MEGs detect weak magnetic fields generated by synchronized neuronal currents (i.e., ionic currents flowing in neuronal dendrites during synaptic transmission). These weak magnetic fields can be detected using magnetometers such as superconducting quantum unit interference devices (SQUIDs) or spin-free exchange relaxation (SERF) magnetometers. For a description of MEGs and methods of using MEGs to record magnetic fields associated with neuronal activity in the brain, see, for example, Hamalaren et al. (1993), Review of Modern Physics, 65(2): 413-497; Bailllet et al. (2017), Nature: Neuroscience, 20(3): 327-339; Gross et al. (2019), Neuron, 104(2): 189-204; Stapleton-Kotloski et al. (2018), Brain Science, 8(8): 157; the contents of which are incorporated herein by reference.

[0058] Dynamic causal modeling (DCM) is a Bayesian generative modeling paradigm that models causal relationships in the brain while taking into account neuronal dynamics, modulated inputs, and observational data (e.g., BOLD fMRI, EEG, or MEG). DCM can be used to estimate couplings between brain regions and quantify the effective connectivity strength between populations of neurons in one or more brain regions. Dynamic causal models of interacting neural populations can be used to estimate inter-regional and intra-regional connectivity strengths using functional neuroimaging data with Bayesian statistical methods. The models can include interactions between excitatory and inhibitory neural populations. Different types of DCM can be used to compute these connectivity strength estimates. Typical DCM assumes no state noise, while stochastic DCM (sDCM) utilizes state noise to account for uncertainties caused by neighboring regions. Spectral DCM is another paradigm that assumes state noise, but in this approach, the model is fitted over the cross spectrum of the system, making sDCM suitable for resting-state modeling. Dynamic effective connectivity (or dynamic DCM) is another approach that can be used to investigate temporal variations in connectivity strength. For descriptions of different types of DCM techniques, please refer to, for example, Friston et al. (2003), Neuroimaging, 19(4): 1273-1302; Friston et al. (2014), Neuroimaging, 94: 396-407; Park et al. (2018), Neuroimaging, 180: 594-608; Li et al. (2011), Neuroimaging, 58(2): 442-457; Stephan et al. (2010), Neuroimaging, 49(4): 3099-3109; Stephan et al. (2008), Neuroimaging, 42(2): 649-662; Marreiros et al., Neuroimaging, 39(1): 269-278; Razi et al. (2015), Neuroimaging, 106: 1-14; and Lee et al. (2006), Neuroimaging, 30 (4): 1243-1254; its contents are incorporated herein by reference.

[0059] Biophysical modeling is based on the biophysical properties of neurons and brain regions. Biophysical neuron models can include a variety of biophysical properties, including but not limited to the firing patterns of individual neurons, neuron type, receptors, inputs and outputs, topology, and self-connections. This invention uses estimates of inter-regional and intra-regional connectivity strength obtained through dynamic causal modeling to simulate synthetic electrophysiological data using a biophysical model. The synthetic electrophysiological data is then compared with experimental electrophysiological data collected from the subjects, and the model is improved to convergence by iteratively adjusting it. For a description of biophysical modeling, see, for example, Example 5, Hill and Tononi (2005), Journal of Neurophysiology, 93(3): 1671-1698; and Murray et al. (2018), Biopsychiatry: Cognitive Neuroscience & Neuroimaging, 3(9): 777-787, the contents of which are incorporated herein by reference.

[0060] In some embodiments, functional neuroimaging and electrophysiological techniques are used in combination to detect a subject's brain responses when exposed to stimuli or performing a task. Furthermore, functional neuroimaging and electrophysiological measurements of brain activity can be performed while the subject is at rest (e.g., without stimulation or task) to compare brain activity to the subject's "baseline" brain state, i.e., to identify brain regions exhibiting changes in neural activity associated with a specific stimulus or task.

[0061] In some embodiments, the methods described herein are used to evaluate changes in brain function in response to optogenetic perturbations of neural activity. In some embodiments, optogenetic techniques are used to induce cell-specific perturbations in the brain. For example, optogenetic techniques can be used to use light to excite or inhibit one or more selected target neurons. For a description of optogenetic techniques, see, for example, Abe et al., 2012; Desai et al., 2011; Duffy et al., 2015; Gerits et al., 2012; Kahn et al., 2013; Lee et al., 2010; Liu et al., 2015; Ohayon et al., 2013; Weitz et al., 2015; Weitz and Lee, 2013; which are incorporated herein by reference.

[0062] The methods described herein can also be used to evaluate changes in brain function in response to brain stimulation by an electric current or magnetic field applied to selected brain regions. For example, electrical brain stimulation (EBS) can be used to stimulate neurons or neural networks in the brain by directly or indirectly exciting their cell membranes with an electric current. For descriptions of EBS techniques, please see, for example, Aum et al. (2018), Frontiers in Bioscience (Landmark Ed), 23: 162-182; Tellez-Zenteno et al. (2011), North American Clinical Neurosurgery, 22(4): 465-75; Padberg et al. (2009), Experimental Neurology, 219: 2-13; Nahas et al. (2010), Biological Psychiatry, 67: 101-109; Lefaucheur et al. (2010), Experimental Neurology, 223: 609-614; Levy et al. (2008), Journal of Neurosurgery, 108: 707-714; Hanajima et al. (2002), Clinical Neurophysiology, 113: 635-641; Picillo et al. (2015), Brain Stimulation, 8: 840-842; Canavero (2014), Guide to Cerebral Cortical Stimulation, Berlin: De Gruyter Open; which is incorporated herein by reference. Alternatively, transcranial magnetic stimulation (TMS) can be used to electrically stimulate the brain via electromagnetic induction and can be used for non-invasive stimulation of specific brain regions. For descriptions of TMS techniques, see, for example, Klomjai et al. (2015), Annals of Physical and Rehabilitation Medicine, 58(4): 208-213; Lefaucheur (2019), Handbook of Clinical Neurology, 160: 559-580; Burke et al. (2019), Handbook of Clinical Neurology, 163: 73-92; the contents of which are incorporated herein by reference.

[0063] In some embodiments, the methods described herein are used to evaluate changes in brain activity that occur in response to a subject performing a cognitive or motor task. Such tasks may include, for example, but not limited to, tests of memory, intelligence, verbal / language, emotion, executive function (e.g., problem-solving, planning, organizational skills, selective attention, inhibitory control), visuospatial function, balance, or physical activity / exercise (e.g., walking, running, limb movement). The methods described herein can be used to analyze intra-regional and inter-regional neural interactions of brain activity involved in a specific cognitive or motor task and to compare said brain activity with the subject's brain activity at rest.

[0064] Brain function modeling can be performed on any region of the brain using the method of the present invention. In some embodiments, the one or more regions of interest are located in the cerebrum, cerebellum, or brainstem regions of the brain. Regions of interest may include, but are not limited to, the basal ganglia, striatum, medulla oblongata, pons, midbrain, medulla oblongata, hypothalamus, thalamus, epithalamus, amygdala, superior colliculus, cerebral cortex, neocortex, xenogeneic cortex, hippocampus, claustrum, olfactory bulb, frontal lobe, temporal lobe, parietal lobe, occipital lobe, caudate putamen, lateral part of globus pallidus, medial part of globus pallidus, subthalamic nucleus, substantia nigra, thalamus, and motor cortex regions of the brain.

[0065] Functional neuroimaging and electrophysiological data of any type of neuron can be acquired, including but not limited to unipolar neurons, bipolar neurons, multipolar neurons, Golgi type I neurons, Golgi type II neurons, axonal neurons, pseudounipolar neurons, interneurons, motor neurons, sensory neurons, afferent neurons, efferent neurons, cholinergic neurons, GABAergic neurons, glutamatergic neurons, dopaminergic neurons, serotonergic neurons, histaminergic neurons, Purkinje cells, polyspinous projection neurons, Runshaw cells, and granule cells or combinations thereof.

[0066] Systems and computer-based methods for brain function modeling

[0067] In another aspect, the present invention includes a computer-implemented method for modeling brain function using a combination of dynamic causal modeling and biophysical modeling as described herein. The computer performs the steps comprising: a) receiving functional neuroimaging data and experimental electrophysiological data for one or more brain regions of interest from a subject; b) fitting the functional neuroimaging data to a dynamic causal model; c) calculating estimates of inter-regional and intra-regional connectivity strengths using the dynamic causal model; d) generating a biophysical model simulating synthetic electrophysiological data using the inter-regional and intra-regional connectivity strength estimates; e) comparing the synthetic electrophysiological data with the experimental electrophysiological data; f) iteratively adjusting the biophysical model until the model converges with the experimental electrophysiological data; and g) displaying information related to the brain function modeling.

[0068] In some embodiments, the functional neuroimaging data includes functional magnetic resonance imaging (fMRI) data, positron emission tomography (PET) data, functional near-infrared spectroscopy (fNIRS) data, single-photon emission computed tomography (SPECT) data, or functional ultrasound imaging (fUS) data.

[0069] In some embodiments, the experimental electrophysiological data includes electroencephalography (EEG) data, magnetoencephalography (MEG) data, or patch-clamp data.

[0070] The method can be implemented in digital electronic circuits or computer software, firmware, or hardware. The disclosed and other embodiments can be implemented in the form of one or more computer program products (i.e., one or more computer program instruction modules encoded on a computer-readable medium for execution by a data processing device or to control the operation of a data processing device). The computer-readable medium can be a machine-readable storage device, a machine-readable storage substrate, a memory device, a material composition that influences machine-readable propagation signals, or any combination thereof.

[0071] Computer programs (also known as programs, software, software applications, scripts, or code) can be written in any programming language, including compiled or interpreted languages, and can be deployed in any form, including as standalone programs or as modules, components, subroutines, or other units suitable for use in a computing environment. A computer program does not necessarily correspond to a file in a file system. Programs can be stored as part of a file containing other programs or data (e.g., in one or more scripts within a markup language document); in a single file dedicated to the program in question; or in multiple coordinating files (e.g., in a file storing one or more modules, subroutines, or portions of code). Computer programs can be deployed to execute on a single computer or on multiple computers located at a site or distributed across multiple sites and interconnected via a communication network.

[0072] In another aspect, systems are provided for performing computer-implemented methods as described herein. Such systems include computers containing processors, storage components (i.e., memory), display components, and other components typically found in general-purpose computers. The storage components store information accessible to the processor, including instructions executable by the processor and data that can be retrieved, manipulated, or stored by the processor.

[0073] The storage component includes instructions. For example, the storage component includes instructions for modeling brain function based on analysis of functional neuroimaging data and experimental electrophysiological data stored therein. The computer processor is coupled to the storage component and configured to execute the instructions stored in the storage component to receive the functional neuroimaging data and experimental electrophysiological data, and analyze the data using dynamic causal modeling and biophysical modeling as described herein, according to one or more algorithms. The display component displays information related to the brain function modeling. In some embodiments, the functional neuroimaging data and experimental electrophysiological data are acquired from a subject with a neurological disease or condition, wherein the information displayed regarding the brain function modeling includes listing one or more neural circuits involved in pathological changes associated with the neurological disease or condition. In some embodiments, the functional neuroimaging data and experimental electrophysiological data are acquired while the subject performs a task or responds to a stimulus, wherein the information displayed regarding the brain function modeling includes listing one or more neural circuits involved in performing the task or responding to the stimulus. In some embodiments, the functional neuroimaging data and the experimental electrophysiological data are acquired while the subject is at rest (e.g., for comparison with data acquired while the subject is performing a task or responding to stimuli).

[0074] The storage component can be any type of storage component capable of storing information accessible to the processor, such as a hard disk drive, memory card, ROM, RAM, DVD, CD-ROM, USB flash drive, writable memory, and read-only memory. The processor can be any well-known processor, such as a processor from Intel Corporation. Alternatively, the processor can be a dedicated controller, such as an ASIC.

[0075] The instructions can be any set of instructions that are executed directly (e.g., machine code) or indirectly (e.g., scripts) by the processor. Therefore, the terms "instruction," "step," and "program" are used interchangeably herein. The instructions can be stored as object code for direct processing by the processor, or in any other computer language, including scripts or sets of independent source code modules that are interpreted on demand or pre-compiled.

[0076] The processor can retrieve, store, or modify data according to the instructions. For example, although the system is not limited to any particular data structure, the data can be stored in: computer registers; in a relational database as a table with multiple distinct fields and records; in an XML document; or in a flat file. The data can also be formatted in any computer-readable format, such as (but not limited to) binary values, ASCII, or Unicode. Furthermore, the data can contain any information sufficient to identify the relevant information, such as numbers, descriptive text, proprietary code, pointers, references to data stored in other memory (including other network locations), or information about functions used to compute the relevant data.

[0077] In some embodiments, the processor and storage components may comprise multiple processors and storage components, which may or may not be stored within the same physical housing. For example, some of the instructions and data may be stored on a removable CD-ROM, while others may be stored within a read-only computer chip. Some or all of the instructions and data may be stored at a location physically remote from the processor but still accessible to the processor. Similarly, the processor may comprise a collection of processors, which may or may not operate in parallel.

[0078] kit

[0079] The present invention also provides a kit for implementing the brain function modeling method, which employs a combination of dynamic causal modeling and biophysical modeling as described herein, utilizing functional neuroimaging data and experimental electrophysiological data. In some embodiments, the kit includes software for implementing the computer-implemented method for brain function modeling using a combination of dynamic causal modeling and biophysical modeling. In some embodiments, the kit includes a system for brain function modeling as described herein.

[0080] Additionally, the kit may further include (in some embodiments) instructions for implementing the target method. These instructions may exist in the target kit in various forms, one or more of which may be present in the kit. For example, the instructions may be printed information on a suitable medium or substrate (e.g., a sheet or several sheets of paper on which information is printed), kit packaging, packaging instructions, etc. Another form of these instructions is a computer-readable medium on which information is already recorded, such as a floppy disk, optical disc (CD), portable flash drive, etc. Yet another form of these instructions may be a URL, thereby allowing access to information on a remote website via the Internet.

[0081] utility

[0082] The revealed methods will have multiple applications in research and development in the field of neuroscience, particularly in the study of brain function and dysfunction, and in the development of therapies for treating neurological diseases and conditions. The modeling approach described herein, combining DCM and biophysical modeling, can correlate different types of data with varying spatial and temporal resolutions. In particular, the inventors have shown that combining DCM and biophysical modeling can generate more comprehensive models of brain function that can reproduce fMRI and electrophysiological data (see examples). Therefore, this modeling approach can be used to interpret experimental data obtained through different techniques and used to test neuroscience hypotheses. Applications of this modeling approach include basic research on neural processes involved in cognitive and motor functions; identifying neural circuits and determining their function in various brain regions; identifying pathologically affected neural circuits and brain regions in patients with neurological diseases and conditions; and monitoring the brain's response to therapies for treating neurological diseases and conditions.

[0083] The target method is applicable to brain function modeling of any brain region, including but not limited to the cerebrum or regions thereof, including but not limited to the frontal lobe, parietal lobe, temporal lobe or occipital lobe, gyri (e.g., frontal gyrus, temporal gyrus, precentral gyrus, postcentral gyrus or cingulate gyrus), sulci (e.g., central sulcus, cingulate sulcus, calcarine sulcus, corpus callosum sulcus or hippocampal sulcus), cerebral cortex (e.g., neocortex or xenogeneic cortex) and corpus callosum; cerebellum or regions thereof, including but not limited to the anterior lobe, posterior lobe and flocculonodular lobe; brainstem or regions thereof, including but not limited to the midbrain, pons and medulla; or basal ganglia or regions thereof, including but not limited to the striatum, including the dorsal striatum (caudate nucleus and putamen) and ventral striatum (nucleus accumbens and olfactory tubercle), globus pallidus, ventral globus pallidus, substantia nigra and subthalamic nuclei.

[0084] The method is also applicable to modeling the function of any type of neuron in any region of the brain or nervous system, including but not limited to unipolar neurons, bipolar neurons, multipolar neurons, Golgi type I neurons, Golgi type II neurons, axonal neurons, pseudounipolar neurons, interneurons (e.g., basket cells, Lugaro cells, unipolar brush cells, spindle cells), pyramidal neurons (e.g., RSad neurons, RSna neurons, IB neurons, and Betz cells), motor neurons (upper motor neurons and lower motor neurons), sensory neurons, afferent neurons, efferent neurons, cholinergic neurons, GABAergic neurons, glutamatergic neurons, dopaminergic neurons, serotonergic neurons, histaminergic neurons, cerebellar Purkinje cells, striatal neurons (e.g., polyspinous projection neurons, cholinergic interneurons, and GABAergic interneurons), leucosan cells, and granule cells. Furthermore, the method is applicable to modeling the function of interconnected groups of neurons forming neural circuits between neurons in one or more selected brain regions, as well as interregional and intraregional connections.

[0085] The target method is also applicable to modeling changes in brain function associated with neurological diseases and conditions, including but not limited to brain injuries such as (but not limited to) frontal lobe injuries, parietal lobe injuries, temporal lobe injuries, and occipital lobe injuries; brain dysfunctions such as (but not limited to) aphasia, writing disorders, dysarthria, apraxia, agnosia, and amnesia; cranial nerve disorders such as (but not limited to) trigeminal neuralgia and hemifacial spasm; autonomic nervous system disorders such as (but not limited to) autonomic dysfunction and multiple system atrophy; epileptic disorders such as epilepsy; central and... Peripheral nervous system motor disorders, such as (but not limited to) Parkinson's disease, essential tremor, amyotrophic lateral sclerosis, Tourette syndrome, and multiple sclerosis, as well as various types of peripheral neuropathy; sleep disorders, such as narcolepsy; and other neurological diseases and conditions associated with neurological damage, developmental disorders, biochemical, anatomical, or electrical malfunctions and / or disease pathology, such as (but not limited to) attention deficit hyperactivity disorder, autism, Asperger's syndrome, obsessive-compulsive disorder, Huntington's disease, Alzheimer's disease, multiple sclerosis, and organic mental disorders. Furthermore, the target method can be used to screen therapies for these neurological diseases and conditions and to monitor the efficacy of selected treatment regimens.

[0086] experiment

[0087] The following examples are provided to fully disclose and describe the methods for constructing and using the present invention to those skilled in the art, and are not intended to limit the scope of the inventors' invention as they see it, nor to represent that the following experiments are all or only the experiments performed. Efforts have been made to ensure the accuracy of the figures used (e.g., quantities, temperatures, etc.), but some experimental errors and biases should be taken into account. Unless otherwise stated, parts are by weight, molecular weight is weight-average molecular weight, temperature is in degrees Celsius, and pressure is at or near atmospheric pressure.

[0088] All publications and patent applications cited in this specification are incorporated herein by reference as if each individual publication or patent application were specifically and individually indicated to be incorporated by reference.

[0089] To include preferred embodiments of the invention, the invention has been described based on specific embodiments discovered or proposed by the inventors. Those skilled in the art will understand that numerous modifications and variations can be made to the exemplified embodiments without departing from the intended scope of the invention. All such modifications are intended to be included within the scope of the appended claims.

[0090] Example 1

[0091] Combining dynamic causal modeling and biophysical modeling to achieve multi-scale brain network functional modeling

[0092] We devised a novel modeling approach to integrate brain function measurements from different perspectives. Brain function was investigated using a variety of experimental techniques at varying spatiotemporal sampling rates and contrasts. Some of the techniques employed included functional magnetic resonance imaging (fMRI), electrophysiological techniques, and optical imaging. While each technique has its own unique advantages in its ability to probe brain function at different scales, none is comprehensive enough. Therefore, it is crucial to have a means to understand how to utilize different measurements to construct a comprehensive model of brain function.

[0093] To construct comprehensive models of brain function, a strategy is needed to model data obtained from multimodal imaging / recording. In this study, we offer a novel approach for comprehensive modeling of whole-brain-scale function, capable of simulating peak-level activity. We first fit the model to data from one imaging / recording modality (e.g., functional magnetic resonance imaging). For example, the result of this process could be an estimate of connectivity between selected regions. Then, we begin modeling neural activity at a single unit scale using the results of the model fitting procedure. This modeling utilizes and is fitted with results from a first dataset as a foundation. By leveraging the influence of another paradigm to attempt fitting experimental data, we obtain a robust model.

[0094] More specifically, we provide an example that can be generalized to other measurement methods and computational modeling techniques. In the following steps, we use a dynamic causal model (DCM) for fMRI data and a biophysical model for electrophysiological data. The DCM constitutes the first model influencing the second model (i.e., the biophysical model). The steps of our method are as follows:

[0095] 1. Fit fMRI data to the DCM model and extract effective inter-regional and intra-regional connectivity strength estimates.

[0096] 2. Using these DCM estimates, we constructed a biophysical model and simulated synthetic electrophysiological data. A model can include other biophysical properties as hypotheses for testing, such as neuron type, receptors, input and output topology, etc.

[0097] 3. Compare the synthetic electrophysiological data with the experimental electrophysiological data. If they match, we terminate our pipeline. If they do not match, we update the biophysical model and repeat the process until convergence.

[0098] DCM is more of a data-driven approach, while biophysical modeling relies heavily on assumptions. This joint modeling approach integrates the advantages of both DCM and biophysical modeling, compensating for their respective major drawbacks. Using DCM estimation in biophysical modeling can also reduce parameter search time. This biophysical modeling approach can, in principle, reproduce and interpret fMRI and electrophysiological data, but the parameter space grows exponentially with the number of regions. On the other hand, biophysical modeling extends DCM with many physiological details. For example, instead of modeling a brain region as a single node, biophysical modeling describes a region as a group of neurons with input and output mechanisms and topological structures.

[0099] In summary, by combining these two modeling methods, we can directly base our analysis on fMRI and electrophysiological data and test hypotheses about potential mechanisms of target brain function.

[0100] Example 2

[0101] Modeling the basal ganglia in the mouse brain

[0102] Using the basal ganglia of the mouse brain as an example, we demonstrate that the modeling techniques described in Example 1 are powerful tools for constructing comprehensive brain function models. In this example, we utilize real fMRI and electrophysiological data. Bernal-Casas et al. (Neuron, 93(3): 522-532, 2017; their full text is incorporated herein by reference) describe in detail the use of optogenetic fMRI (ofMRI) data for dynamic causal modeling (DCM), and for illustrative purposes, we provide a brief description of that description here.

[0103] The basal ganglia are a crucial region for motor control and many other important brain functions. This region has long been challenging to study due to its anatomically complex mix of different cell types and intricate circuitry. Bernal-Casas et al. (ibid.) used optogenetic fMRI to separately stimulate D1 and D2 medium-sized polyspinous neurons in the striatum, obtaining high-resolution data on pathways associated with these two cell types in the basal ganglia. We used fMRI BOLD data from seven regions of interest and electrophysiological data from four regions of interest for modeling (Figs. 1 and 2). Fig. 1 shows the averaged experimental BOLD data from all subjects. Light stimulation was performed at 20 Hz, with each cycle repeated 20 s on-20 s off-20 s on 6 times. The regions of interest were the caudate putamen (CPu), lateral globus pallidus (GPe), medial globus pallidus (GPi), subthalamic nucleus (STN), substantia nigra (SN), thalamus (THL), and motor cortex (MCX).

[0104] We first performed a DCM analysis using data from each subject to reveal the effective connectivity estimates between and within each region. For this analysis, we employed the prior generative network model shown in Figure 3. The results are shown in Figure 4, where connectivity strength estimates are in Hz. The magnitudes of the parameters in this figure reflect the influence of the rate of change of neuronal activity in one region on the neuronal activity in another region. These estimates were then used to influence the design of neuronal populations in biophysical modeling.

[0105] In constructing the biophysical model, we used the same mouse basal ganglia-thalamic cortex circuit structure model as used in the DCM analysis and the modified neuron model shown in

[10] . The Hill and Tononi neuron model is a Hodgkin-Huxley model that depicts biologically accurate details. While in the DCM analysis, one node represents a region of interest, in the biophysical model we used 100 neuronal units for a region of interest to better study neuronal-level mechanisms. The number of units was arbitrarily set.

[0106] We first constructed a biophysical model without making any assumptions; that is, we derived the model directly from the connectivity parameters in the DCM results. Figure 5 shows the electrophysiological data simulated using this model, which, along with… Figure 2 The experimental data shown are consistent. However, DCM is a region-level analysis and may not reflect true neuronal-level mechanisms; therefore, testing hypotheses about neuronal-level mechanisms is a key advantage of our biophysical modeling. For example, anatomically, the projection from GPi to the thalamus is GABAergic, an inhibitory connection. In principle, this connection should result in opposite polarities of GPi and thalamic activity, but our experiments show that they both exhibit increased activity during D1-MSN stimulation.

[0107] We hypothesize that the thalamus can be activated between GPi inhibitory signals, and that synchronization between GPi neurons is key to this seemingly paradoxical phenomenon. This hypothesis is based on the observation that in our D1-MSN stimulation experiments, the level of synchronization between neurons in multiple regions of interest increased. A thalamic neuron can receive input from multiple GPi neurons. Before D1 stimulation, the level of synchronization between GPi neurons is relatively low, so a thalamic neuron can receive asynchronous inhibitory input from intrinsically activated GPi firing. During stimulation, GPi firing and inhibition of the thalamus become more synchronized; therefore, intrinsically active thalamic neuronal firing can occur between synchronized GPi inhibitory signals.

[0108] To test this hypothesis, we generated a new model by forcibly changing the GPi-thalamic connection to an inhibitory connection. The rest of this model remained the same as the first model. Despite GPi inhibition, we successfully increased thalamic activity during stimulation. Figure 6 illustrates simulated activity in the thalamus, and Figure 7 shows how a thalamic neuron fires between GPi inhibitory inputs. These results demonstrate that our combined DCM and biophysical modeling approach serves as a powerful tool for interpreting fMRI and electrophysiological data and testing potential mechanistic hypotheses.

[0109] Example 3

[0110] Sequential model fitting

[0111] Our method can be applied to perform sequential model fitting using information from one model to improve another model when data is collected from more than two different sources at different scales. Therefore, all existing applications of the modeling paradigm we use can benefit from our method. This makes the number of potential applications quite large, including but not limited to the following:

[0112] 1. Optogenetics experiments, in which selective cell firing can be performed during fMRI scans and electrophysiological data acquisition. Our method will enable comprehensive brain circuit modeling.

[0113] 2. Resting-state experiments, in which the brain receives no direct external input. Resting-state fMRI (rsfMRI) and electrophysiological data collection are common, usually followed by connectivity modeling. Our method will enable comprehensive brain circuit modeling.

[0114] The experimental methods described above can be used to study diseases such as Parkinson's disease and epilepsy. For example, in [2], MRI and DCM were used to study neural circuits that play an important role in Parkinson's disease. Similarly, in [4], fMRI was used to analyze epileptic networks. Our modeling approach can play a key role in advances in these fields. Our combined DCM and biophysical modeling can generate brain function models that can reproduce fMRI and electrophysiological data in a unique way. Importantly, this approach can correlate two types of data with different spatial and temporal resolutions. This modeling approach can be a powerful tool for interpreting experimental data from different technologies and for testing neuroscience hypotheses.

[0115] Example 4

[0116] BOLD fMRI and electrophysiological data

[0117] Functional magnetic resonance imaging (fMRI) is a neuroimaging technique that allows for the non-invasive, large-scale monitoring of temporal changes in brain activity. It has achieved great success as a functional neuroimaging technique over the past three decades and has been applied in clinical and cognitive neuroscience [9]. In a 1992 paper [1], the authors proposed an MRI technique that allows us to image temporal changes in brain activity in MRI by detecting blood oxygen levels. This paper marked the birth of fMRI by detecting blood oxygen level-dependent (BOLD) signals. The phenomenon behind these signals is the oxygen demand of neurons.

[0118] By capturing images periodically, the activity of each voxel can be extracted using the scanning process. After the fMRI scan is complete, we obtain a 4-D signal, where the fourth dimension is time. We then anatomically mask this tensor to select regions of interest and perform dimensionality reduction on the selected regions. We compute the average time series for each region to represent them as a single time series. In our approach, the BOLD fMRI signal is one of the data sources we use for modeling.

[0119] Electrophysiological techniques study the electrical properties of neurons, typically voltage or current changes. Commonly used non-invasive electrophysiological techniques include electroencephalography (EEG) and magnetoencephalography (MEG), as well as invasive techniques such as intracellular, patch-clamp, and extracellular recordings, which have proven successful in acquiring high temporal resolution data [5]

[18]

[11] . In our current study, we used peak potential sequences collected from extracellular recordings. One advantage of electrophysiological techniques is their recent integration with cutting-edge techniques such as multiphoton microscopy and optogenetics. In our work, we used data from optogenetic fMRI, which allows researchers to use light to excite or inhibit specific neurons, providing researchers with unprecedented high spatial and temporal resolution

[13] . By recording the electrical activity around stimulated or inhibited neurons, a direct link can be established between neuronal excitability and the response of downstream / related cells.

[0120] In summary, our approach uses non-invasive fMRI to monitor temporal changes in brain regions of interest via an indicator called the BOLD signal, and uses electrophysiological data to investigate the behavior and related mechanisms of specific neurons.

[0121] Example 5

[0122] Effective connectivity and dynamic causal modeling (DCM)

[0123] DCM is a Bayesian generative modeling paradigm that can comprehensively model causal relationships in the brain [6], [7]. It takes into account neuronal dynamics, regulation and / or excitatory input, and observation methods (BOLD fMRI, EEG / MEG, etc.). Therefore, it is considered one of the most biologically sound and technologically advanced modeling methods

[17] . The purpose of DCM is to estimate the coupling between brain regions and their dependence on the experimental environment, and to find model evidence to compare different hypotheses. The paper mentioned in the results section also used this technique, see [2].

[0124] DCM uses a state-space system to simulate brain dynamics. There are two sets of equations governing this system: the evolution equation and the observation equation, as given in (1).

[0125]

[0126] The time series x(t) represents neuronal activity, u(t) represents the input stimulus, and y(t) represents the observed signal. w(t) represents state noise, and ∈(t) represents observation noise. The function g(.) is the observation model, which varies with the imaging method, while the function f(.) is as follows:

[0127]

[0128] In this function, matrix A represents the effective connection strength because it defines the variation in neuronal activity in a region based on the current activity of all regions. This matrix is ​​typically the primary target for the model fitting task. Matrices B and C in the model reflect the influence of the input on the dynamics. Matrix B manages the connection modulating effect of the input, while matrix C simulates the direct contribution of the input to the dynamics.

[0129] There are several different types of DCMs in the literature. Typical DCM [6] assumes no state noise, while in stochastic DCM (sDCM)

[14] , state noise is used to explain the uncertainty caused by neighboring regions. Spectral DCM [7] is another paradigm that assumes state noise, but in this approach, the model fits on the cross spectrum of the system, which makes it suitable for resting state modeling. Dynamically effective connectivity (or dynamic DCM)

[16] is also a approach we use in our method, in which the temporal variation of connectivity strength is investigated. In our approach, we are able to use any type of DCM as one of the components of the joint modeling.

[0130] DCM inversion

[0131] DCM is a Bayesian modeling approach, which means that its parameters are not considered constants to be found, but rather distributions to be estimated, i.e., random variables. Therefore, we are concerned with p(θ|y, m), i.e., the posterior, where θ is the union of all parameters to be estimated and m is the model hypothesis we use. The model hypothesis in the DCM context can be viewed as a specific choice of f, g, and E and their respective priors. For example, there is DCM using a nonlinear evolution function

[19] and some other models with two neuron states

[15] . Therefore, in this context, the choice is called the model.

[0132] During the inversion process, due to the Bayesian nature of this approach, we need to use priors. The choice of priors is actually a crucial consideration for correct modeling. Some parameters allow the use of information from previous neuroscience research (e.g., hemodynamic priors), while for others, we can only assume supports (e.g., A). In the latter type, we only make a binary decision about the existence of the parameter (i.e., whether its posterior is non-zero). If it is set to zero, the posterior will have zero mean and zero variance.

[0133] Inversion of the DCM model is based on calculating the model evidence p(y|m), which is the probability of observed data given the model. Analyzing and calculating model evidence is often complex; therefore, the variational Bayesian method is commonly used for inversion.

[0134] In this variational approach, our task is to iteratively update the distribution known as the variational distribution (denoted by q(θ)). Evidence for the logarithmic model is as follows:

[0135]

[0136] Since we do not know p(θ|y, m), we cannot compute the first term. However, due to the properties of the KL divergence, we know it is non-negative, so the second term constitutes the lower bound of the model evidence. Therefore, the best estimate (rather than maximizing the logarithmic model evidence) is to maximize the so-called negative free energy.

[0137]

[0138] In this equation, q(θ) is the variational distribution, which is also an approximation of the posterior distribution, and KL(.) is the Kulbeck-Leibler divergence operation. The first term represents accuracy, while the second term has a complexity interpretation (because it measures the deviation from the prior distribution). If q*(θ) = p(θ|y, m), then the above lower bound makes the equation hold. In DCM, we also make an assumption about the distribution of the parameters; we assume they are normally distributed. This assumption is called the Laplace approximation, which we reflect in the prior distribution p(θ|m).

[0139] In summary, our problem can be summarized as follows:

[0140]

[0141] This problem is solved using an iterative method called Expectation Maximization (EM). EM is a steepest descent algorithm based on Jacobi calculations and is a common tool in variational analysis. Once the algorithm converges, it returns the model evidence for the outcome along with the variational distribution that produced the value. As mentioned earlier, this distribution is an approximation of the posterior distribution of the parameters. Maximum a posteriori (MAP) estimates of these distributions will yield the parameter estimates we are exploring.

[0142] Biophysical modeling methods

[0143] Modeling based on the biophysical properties of neurons and brain regions is another major approach to neural modeling. A key advantage of this approach is that the models can reflect brain behavior at different scales and resolutions.

[0144] At least a priori model should be constructed based on experimental results and the hypotheses to be tested. The priori model usually consists of two parts: a structural model and a neuronal model. The structural model identifies the brain regions involved in the hypotheses regarding the target processes and regional connections; the neuronal model expands each region into a set of neuronal units to reflect detailed biophysical mechanisms. In our joint DCM and biophysical modeling process, the structural model is also the priori model for DCM analysis. The neuronal model first describes the firing patterns of individual neurons, usually through differential equations

[22]

[12] . Sometimes other biophysical properties are also described, such as self-connectivity, neuron type, receptors, input and output topology, etc.

[10]

[21] .

[0145] In our example study, we used a neuronal model, see Hill and Toni (Sleep and wakefulness modeling in the thalamic cortex system, Journal of Neurophysiology, 93(3): 1671-1698, 2005; the full text of which is incorporated herein by reference). Other types of biophysical models may be chosen. If the simulated neuron is not unresponsive and the membrane potential exceeds a threshold, it fires a single spike. A neuron enters a refractory period after firing the spike, during which no new spikes are produced. The membrane potential V is managed based on the following formula.

[0146]

[0147] g Na and g K It is a constant leakage conductance for Na and K. E Na and E K It is the reversal potential, assumed to be constant. I syn These are synaptic currents originating from other neurons. int It is the intrinsic current. I re It is the repolarization current, effective only during the refractory period. τ m It is a time constant.

[0148] The output of the biophysical model is electrophysiological data, which can be in the form of time series in EEG / MEG style or in the form of spike potential sequences in the style of single neuron recordings, depending on the type of neuron model selected. In our example study, the output is a spike potential sequence from a single neuron. The simulated electrophysiological data can then be converted into fMRI BOLD time series using hemodynamic functions. Buxton et al. [3] provided a comprehensive model to explain this process, which is called the “balloon model”. In this model, the complex relationship between blood oxygen levels and neuronal activity is presented as a set of differential equations. The exact set of equations is shown in (7), where x is neuronal activity, s is the vasodilation signal, f is the blood inflow, κ is the signal attenuation constant, γ is the feedback regulation constant, v is the blood volume of the brain region, and f out τ is the blood outflow, α is the average blood transit time, α is the vascular stiffness, q is the deoxygenated hemoglobin content, and E0 is the oxygen extraction fraction at rest.

[0149] (7)

[0150] Under these underlying dynamics, the observed signal y(t), i.e. the time process of the voxel intensity, is proposed as the following equation:

[0151] (8)

[0152] The coefficients k1, k2, and k3 were tested in

[20] and are combinations of scan and biological parameters. In this model, neuronal activity leads to vasodilation, which in turn increases blood inflow. Net flow directly affects blood volume, and the last two relationships demonstrate the close relationship between blood flow, volume, and deoxyhemoglobin content. This model forward models neuronal activity as changes in fMRI BOLD signal. Therefore, the simulation results can be directly compared with the experimental results by comparing the simulated fMRI time series and DCM results with the corresponding experimental results, and the prior model can be modified accordingly or another prior model can be selected.

[0153] software

[0154] Professor Karl Friston and his colleagues at the Wellcome Centre for Human Neuroimaging, University College London, have released a MATLAB toolbox called SPM (Statistical Parametric Mapping), which can be used for many functional neuroimaging applications [8]. DCM inversion and analysis are among the features of this toolbox and are frequently used in DCM literature. Our approach is based on their toolbox.

[0155] References

[0156] [1] Peter A. Bandettini, Eric C. Wong, R. Scott Hinks, Ronald S. Tikofsky and James S. Hyde. Temporal epigraph of human brain function during task activation. Magnetic Resonance Imaging in Medicine, 25(2): 390–397, 1992.

[0157] [2] David Bernal-Casas, Hyun Joo Lee, Andrew J Weitz and Jin Hyung Lee. Study of brain circuit function using dynamic causal modeling applicable to optogenetic fmri. Neuron, 93(3): 522–532, 2017.

[0158] [3] Richard B Buxton, Eric C Wong and Lawrence R Frank. Hemodynamic and oxygenation kinetic changes during brain activation: a balloon model. Magnetic Resonance in Medicine, 39(6): 855–864, 1998.

[0159] [4] ManKin Choy, Ben A Duffy and Jin Hyung Lee. Optogenetic study of networks in epilepsy. Journal of Neuroscience Research, 95(12): 2325–2335, 2017.

[0160] [5] Fernando Lopes da Silva. EEG and MEG: Relevance to Neuroscience. Neuron, 80(5): 1112-1128, 2013.

[0161] [6] Karl J Friston, Lee Harrison and Will Penny. Dynamic causal modeling. Neuroimaging, 19(4): 1273–1302, 2003.

[0162] [7] Karl J. Friston, Joshua Kahan, Bharat Biswal and Adeel Razi. DCM for resting-state fMRI. Neuroimaging, 94:396–407, 2014.

[0163] [8] KJ Friston, J. Ashburner, SJ Kiebel, TE Nichols and WDPenny, eds. Statistical parameter mapping: Image analysis of brain function. Academic Press, 2007.

[0164] [9] Gary H Glover. Overview of functional magnetic resonance imaging. Clinical Neurosurgery, 22(2): 133–139, 2011.

[0165]

[10] Sean Hill and Giulio Tononi. Sleep and wakefulness modeling in the thalamic cortex. Journal of Neurophysiology, 93(3): 1671–1698, 2005.

[0166]

[11] Donald R Humphrey and Edward M Schmidt. Extracellular single-unit recording methods. Neurophysiological Techniques, pp. 1–64. Springer, 1990.

[0167]

[12] Eugene M Izhikevich. A simple model of spike neurons. IEEE Transactions on Neural Networks, 14(6): 1569–1572, 2003.

[0168]

[13] Hyun Joo Lee, Andrew J Weitz, David Bernal-Casas, Ben A Duffy, ManKin Choy, Alexxai V Kravitz, Anatol C Kreitzer and Jin Hyung Lee. Activation of medium-sized polyspinous neurons in direct and indirect pathways drives different whole-brain responses. Neuron, 91(2): 412-424, 2016.

[0169]

[14] Baojuan Li, Jean Daunizeau, Klaas E. Stephan, Will Penny, Dewen Hu and Karl Friston. Generalized filtering and randomized DCM for fMRI. Neuroimaging, 58(2): 442-457, 2011.

[0170]

[15] Andr'e C Marreiros, Stefan J Kiebel and Karl J Friston. Dynamic causal modeling for fmri: a two-state model. Neuroimaging, 39(1): 269-278, 2008.

[0171]

[16] Hae-Jeong Park, Karl J Friston, Chongwon Pae, Bumhee Park and Adeel Razi. Dynamic effective connectivity in resting-state fMRI. Neuroimaging, 180:594-608, 2018.

[0172]

[17] Stephen M. Smith. The future of fmri connectivity. Neuroimaging, 62(2): 1257-1266, 2012.

[0173]

[18] Harald Sontheimer and Christopher B Ransom. Whole-cell patch-clamp recording. Patch-clamp analysis, pp. 35–67. Springer, 2002.

[0174]

[19] Klaas Enno Stephan, Lars Kasper, Lee M Harrison, Jean Daunizeau, Hanneke EM den Ouden, Michael Breakspear and Karl J Friston. Nonlinear dynamic causal models for fMRI. Neuroimaging, 42(2): 649-662, 2008.

[0175]

[20] Klaas Enno Stephan, Nikolaus Weiskopf, Peter M Drysdale, Peter A Robinson and Karl J Friston. Comparison of hemodynamic models with dcm. Neuroimaging, 38(3): 387-401, 2007.

[0176]

[21] MA Tagamets and B Horwitz. Integrating electrophysiological and anatomical experimental data to create large-scale models of human brain imaging studies simulating delayed-matched samples. Cerebral Cortex (NY, NY: 1991), 8(4): 310–320, 1998.

[0177]

[22] Hugh R Wilson and Jack D Cowan. Excitatory and inhibitory interactions in local populations of model neurons. Journal of Biophysics, 12(1): 1-24, 1972.

Claims

1. A computer-based method for brain function modeling, the method comprising the following steps: a) Receive functional neuroimaging data of neural activity in one or more regions of interest in the subject's brain; b) Receive experimental electrophysiological data of one or more neurons in one or more regions of interest in the subject's brain; c) Fit the functional neuroimaging data of the neural activity to a dynamic causal model; d) Calculate the estimated connection strength between and within neural regions using the dynamic causal model described above; e) Use the inter-regional and intra-regional connectivity strength estimates to generate biophysical models that simulate synthetic electrophysiological data; f) Compare the synthetic electrophysiological data with the experimental electrophysiological data; and g) Iteratively adjust the biophysical model until the model converges with the experimental electrophysiological data.

2. The computer-implemented method according to claim 1, wherein the functional neuroimaging data includes functional magnetic resonance imaging data, positron emission tomography (PET) data, functional near-infrared spectroscopy data, single-photon emission computed tomography (SPECT) data, or functional ultrasound imaging data.

3. The computer-implemented method according to claim 2, wherein the functional magnetic resonance imaging data is a blood oxygen level-dependent signal.

4. The computer-implemented method according to claim 1, wherein the experimental electrophysiological data includes electroencephalography (EEG) data, magnetoencephalography (MEG) data, or patch-clamp data.

5. The computer-implemented method of claim 1, further comprising receiving data acquired after using optogenetics to excite or inhibit one or more selected target neurons with light.

6. The computer-implemented method according to claim 2, wherein the functional magnetic resonance imaging data is optogenetic functional magnetic resonance imaging data.

7. The computer-implemented method of claim 1, wherein one or more regions of interest are located in the cerebrum, cerebellum, or brainstem region of the brain.

8. The computer-implemented method of claim 7, wherein one or more regions of interest are located in the basal ganglia, striatum, medulla, pons, hypothalamus, thalamus, superior colliculus, cerebral cortex, neocortex, xenogeneic cortex, hippocampus, or olfactory bulb region of the brain.

9. The computer-implemented method of claim 8, wherein one or more regions of interest are located in the frontal lobe, temporal lobe, parietal lobe, or occipital lobe of the brain.

10. The computer-implemented method of claim 8, wherein one or more regions of interest are located in the caudate putamen, lateral part of the globus pallidus, medial part of the globus pallidus, subthalamic nucleus, substantia nigra, thalamus, or motor cortex of the brain.

11. The computer-implemented method according to claim 1, wherein the one or more neurons are unipolar neurons, bipolar neurons, multipolar neurons, Golgi type I neurons, Golgi type II neurons, axonal neurons, pseudounipolar neurons, interneurons, motor neurons, sensory neurons, afferent neurons, efferent neurons, cholinergic neurons, GABAergic neurons, glutamatergic neurons, dopaminergic neurons, serotonergic neurons, histaminergic neurons, Purkinje cells, polyspinous projection neurons, Runshaw cells, or granule cells or combinations thereof.

12. The computer-implemented method of claim 1, further comprising receiving optical imaging or multiphoton microscopy data of the one or more regions of interest in the brain of the subject.

13. The computer-implemented method of claim 1, wherein the subject suffers from a neurological disease or neurological disorder.

14. The computer-implemented method of claim 13, further comprising, based on the brain function modeling, detecting changes in the brain function of the subject compared to the brain function of a control subject.

15. The computer-implemented method of claim 13, further comprising identifying neural circuits involving pathological changes associated with the neurological disease or neurological condition based on the brain function modeling.

16. The computer-implemented method of claim 13, further comprising: Receive functional neuroimaging data and experimental electrophysiological data collected from the subject following treatment for the neurological disease or neurological condition; and Based on brain function modeling of the subject before and after treatment for the neurological disease or neurological condition, changes in the subject's brain function after treatment compared to before treatment are detected.

17. The computer-implemented method according to claim 1, wherein the dynamic causal model is a typical dynamic causal model, a stochastic dynamic causal model, a spectral dynamic causal model, or a dynamic dynamic causal model.

18. The computer-implemented method of claim 1, wherein the functional neuroimaging data or experimental electrophysiological data is acquired while the subject is at rest.

19. The computer-implemented method of claim 1, wherein the functional neuroimaging data or experimental electrophysiological data is acquired while the subject performs a task or responds to a stimulus.

20. The computer-implemented method of claim 19, further comprising identifying neural circuits involved in performing the task or responding to the stimulus.

21. A computer-implemented method for brain function modeling, wherein the computer performs steps, the steps comprising: a) Receive functional neuroimaging data and experimental electrophysiological data of one or more brain regions of interest from the subject; b) Fit the functional neuroimaging data to a dynamic causal model; c) Calculate the estimated connection strength between and within neural regions using the dynamic causal model; d) Use the inter-regional and intra-regional connection strength estimates to generate biophysical models that simulate synthetic electrophysiological data; e) Compare the synthetic electrophysiological data with the experimental electrophysiological data; f) Iteratively adjust the biophysical model until the model converges using the experimental electrophysiological data; and g) Display information about the brain function modeling.

22. The computer-implemented method of claim 21, wherein the functional neuroimaging data comprises functional magnetic resonance imaging data, positron emission tomography (PET) data, functional near-infrared spectroscopy (FIR) data, single-photon emission computed tomography (SPECT) data, or functional ultrasound imaging data.

23. The computer-implemented method according to claim 21, wherein the experimental electrophysiological data includes electroencephalography (EEG) data, magnetoencephalography (MEG) data, or patch-clamp data.

24. The computer-implemented method of claim 21, wherein the subject suffers from a neurological disease or neurological disorder.

25. The computer-implemented method of claim 24, wherein step g) comprises listing one or more neural circuits involving pathological changes associated with the neurological disease or the neurological condition.

26. The computer-implemented method of claim 21, wherein the functional neuroimaging data and the experimental electrophysiological data are acquired while the subject performs a task or responds to a stimulus, wherein step g) comprises listing one or more neural circuits involved in performing the task or responding to the stimulus.

27. A system for performing brain function modeling using a computer-implemented method according to any one of claims 1 to 26, the system comprising: a) A storage component for storing data, wherein the storage component has instructions for performing brain function modeling based on analysis of the functional neuroimaging data and experimental electrophysiological data stored therein; b) A computer processor for processing the functional neuroimaging data and experimental electrophysiological data using one or more algorithms, wherein the computer processor is coupled to the storage component and configured to execute instructions stored in the storage component to receive the input functional neuroimaging data and experimental electrophysiological data, and to analyze the data using the computer-implemented method according to any one of claims 1 to 26; and c) A display component for displaying information related to the brain function modeling.

28. A non-transitory computer-readable medium comprising program instructions that, when executed by a processor in a computer, cause the processor to perform the method according to any one of claims 1 to 26.

Images