Method for analyzing on-chip brain neuron population, simulation platform and related device

Abstract

The present application relates to the field of on-chip brain technology, and provides an analysis method of an on-chip brain neuron group, a simulation platform and related devices. The neural dynamics model is analyzed in structure to obtain potential neural structure relationship; according to a preset initial state, the neural dynamics model is dynamically predicted and analyzed to obtain dynamic prediction information corresponding to each initial state; according to a preset stimulation parameter range, the neural dynamics model is analyzed in feasible region to obtain corresponding feasible region information; and the neural structure relationship, all dynamic prediction information and feasible region information are taken as analysis results of the target neuron group. The neural dynamics model is used to simulate and analyze the neuron group, thereby simplifying the analysis operation of the neuron group and improving the analysis efficiency of the neuron group.

Description

Technical Field

[0001] The present invention relates to the field of brain-on-a-chip technology, and in particular to an analysis method, simulation platform and related devices for a neuron population in a brain-on-a-chip. Background Art

[0002] In vitro brain-on-chips are three-dimensional cell aggregates developed from human embryonic stem cells or induced pluripotent stem cells. They possess certain structural and functional properties of brain tissue in vitro and can mimic certain aspects of the human brain. The development of these brain-on-chips provides new tools for neuroscience research and disease studies.

[0003] Currently, for in vitro cultured brain-on-chips, although some platforms can observe and record the status of neuronal populations based on technologies such as microelectrode arrays and calcium ion imaging, they cannot analyze the activity process of neuronal populations. However, the activity analysis of neuronal populations is very necessary and has important significance for research directions such as drug screening, disease modeling, and disease exploration of brain-on-chips. Summary of the Invention

[0004] In view of this, an object of the present invention is to provide an analysis method, simulation platform and related devices for on-chip brain neuron populations.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] In a first aspect, the present invention provides a method for analyzing a brain-on-chip neuron population, which is applied to a simulation platform having a pre-stored neural dynamics model for simulating a target neuron population. The method for analyzing a brain-on-chip neuron population comprises:

[0007] Performing structural analysis on the neural dynamics model to obtain potential neural structural relationships; the neural structural relationships represent the connection relationships between neurons in the target neuron population;

[0008] According to each preset initial state, the neural dynamics model is dynamically predicted and analyzed to obtain dynamic prediction information corresponding to each initial state; the dynamic prediction information represents the relationship between the discharge activity of the target neuron population starting from the initial state and time;

[0009] Performing feasible domain analysis on the neural dynamics model according to a preset stimulation parameter range to obtain corresponding feasible domain information; the feasible domain information represents the state change range of the target neuron population under the stimulation parameter range;

[0010] The neural structure relationship, all dynamic prediction information and the feasible domain information are used as the analysis results of the target neuron group.

[0011] In an optional embodiment, each neuron in the target neuron population has a corresponding number;

[0012] The step of performing structural analysis on the neural dynamics model to obtain potential neural structural relationships includes:

[0013] The connection matrix is ​​obtained from the model parameters of the neural dynamics model, and the connection matrix is ​​expressed as follows:

[0014]

[0015] Among them, w ij represents the connection strength between the i-th neuron and the j-th neuron in the target neuron population;

[0016] The eigenvalues ​​of the connection matrix are calculated, and the distribution of the eigenvalues ​​is analyzed to obtain a spectral distribution map, and the spectral distribution map is used as the neural structure relationship.

[0017] In an optional embodiment, the step of dynamically predicting and analyzing the neural dynamics model according to each preset initial state to obtain dynamic prediction information corresponding to each initial state includes:

[0018] For any of the initial states, using the neural dynamics model to make predictions based on the initial state to obtain multiple neural state sequences;

[0019] Performing signal analysis and dimensionality reduction operations on the multiple neural state sequences to obtain a dynamic behavior matrix, and using the dynamic behavior matrix as dynamic prediction information corresponding to the initial state;

[0020] Traverse each of the initial states to obtain dynamic prediction information corresponding to each of the initial states.

[0021] In an optional embodiment, the step of using the neural dynamics model to predict based on the initial state to obtain multiple neural state sequences includes:

[0022] Using the neural dynamics model to make predictions based on the initial state to obtain a first neural state sequence, and using the first neural state sequence as a current neural state sequence;

[0023] Using the neural dynamics model to predict based on the current neural state sequence, to obtain a next neural state sequence;

[0024] Counting the total number of neural state sequences obtained, and comparing the total number with a preset value;

[0025] if the total number does not reach the preset value, repeating the step of predicting a next neural state sequence based on the current neural state sequence by using the neural dynamics model, and taking the next neural state sequence as a new current neural state sequence;

[0026] if the total number reaches the preset value, obtaining all neural state sequences to obtain the plurality of neural state sequences.

[0027] In an optional implementation, the step of performing a feasible region analysis on the neural dynamics model according to a preset stimulation parameter range to obtain corresponding feasible region information comprises:

[0028] generating a stimulation parameter sequence comprising a plurality of stimulation parameters based on the stimulation parameter range;

[0029] performing a stimulation operation on the neural dynamics model according to each of the stimulation parameters to obtain a neural response sequence corresponding to each of the stimulation parameters;

[0030] performing a state analysis based on the neural response sequence corresponding to each of the stimulation parameters to obtain a neural state vector corresponding to each of the stimulation parameters, and taking a set comprising all neural state vectors as the feasible region information.

[0031] In an optional implementation, the step of performing a stimulation operation on the neural dynamics model according to each of the stimulation parameters to obtain a neural response sequence corresponding to each of the stimulation parameters comprises:

[0032] taking a first stimulation parameter in the stimulation parameter sequence as a target stimulation parameter;

[0033] predicting an original neural state sequence based on a preset original state by using the neural dynamics model, and taking the original neural state sequence as a current neural response sequence;

[0034] generating a target stimulation sequence based on the target stimulation parameter, and predicting a next neural response sequence based on the target stimulation sequence and the current neural response sequence by using the neural dynamics model;

[0035] taking the next neural response sequence as the neural response sequence corresponding to the target stimulation parameter;

[0036] If the target stimulation parameter is not the last stimulation parameter of the stimulation parameter sequence, then after using the next stimulation parameter of the target stimulation parameter as a new target stimulation parameter and the next neural response sequence as a new current neural response sequence, repeatedly performing the steps of generating a target stimulation sequence based on the target stimulation parameter, and performing prediction based on the target stimulation sequence and the current neural response sequence using the neural dynamics model to obtain a next neural response sequence;

[0037] If the target stimulation parameter is the last stimulation parameter of the stimulation parameter sequence, a neural response sequence corresponding to each stimulation parameter is obtained.

[0038] In an optional embodiment, the neural dynamics model is obtained in the following manner:

[0039] Acquire multiple neural state sequence samples, and split each of the neural state sequence samples into a first neural state sequence and a second neural state sequence of equal length;

[0040] Acquire multiple neural response sample groups; each of the neural response sample groups includes a stimulation sequence, an original neural state, and a true neural response sequence;

[0041] For each of the first neural state sequences, using the to-be-trained model to predict based on the first neural state sequence to obtain a third neural state sequence, thereby obtaining each third neural state sequence;

[0042] For each of the neural response sample groups, using the to-be-trained model to make predictions based on the stimulation sequence and the original neural state in the neural response sample group to obtain an estimated neural response sequence, thereby obtaining each estimated neural response sequence;

[0043] The model to be trained is trained based on all second neural state sequences, all third neural state sequences, all true neural response sequences, and all estimated neural response sequences to obtain the neural dynamics model.

[0044] In a second aspect, the present invention provides an on-chip brain neuron population analysis device, which is applied to a simulation platform having a pre-stored neural dynamics model for simulating a target neuron population. The on-chip brain neuron population analysis device comprises:

[0045] A neural analysis module, configured to perform structural analysis on the neural dynamics model to obtain potential neural structural relationships; the neural structural relationships represent the connection relationships between neurons in the target neuron population;

[0046] According to each preset initial state, the neural dynamics model is dynamically predicted and analyzed to obtain dynamic prediction information corresponding to each initial state; the dynamic prediction information represents the relationship between the discharge activity of the target neuron population starting from the initial state and time;

[0047] Performing feasible domain analysis on the neural dynamics model according to a preset stimulation parameter range to obtain corresponding feasible domain information; the feasible domain information represents the state change range of the target neuron population under the stimulation parameter range;

[0048] The result acquisition module is used to use the neural structure relationship, all dynamic prediction information and the feasible domain information as the analysis result of the target neuron group.

[0049] In a third aspect, the present invention provides a simulation platform comprising a processor and a memory, wherein the memory stores a computer program, and when the processor executes the computer program, the method for analyzing the on-chip brain neuron population described in any one of the aforementioned embodiments is implemented.

[0050] In a fourth aspect, the present invention provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for analyzing a neuron population of a brain-on-a-chip as described in any one of the aforementioned embodiments.

[0051] The present invention provides an analysis method, simulation platform, and related devices for a brain-on-a-chip neuron population. The method performs structural analysis on a neural dynamics model to obtain potential neural structural relationships. The neural structural relationships represent the connection relationships between neurons in a target neuron population. Based on each preset initial state, the neural dynamics model is dynamically predicted and analyzed to obtain dynamic prediction information corresponding to each initial state. The dynamic prediction information represents the relationship between the discharge activity of the target neuron population starting from the initial state and changes over time. Based on a preset stimulation parameter range, the neural dynamics model is subjected to feasible domain analysis to obtain corresponding feasible domain information. The feasible domain information represents the range of state changes of the target neuron population under the stimulation parameter range. The neural structural relationships, all dynamic prediction information, and feasible domain information are used as the analysis results of the target neuron population. By simulating and analyzing the neuron population using the neural dynamics model, the analysis operation of the neuron population is simplified and the analysis efficiency of the neuron population is improved.

[0052] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some of the embodiments of the present application, and therefore should not be regarded as a limitation to the scope. Other related drawings can also be obtained by those of ordinary skill in the art without creative labor.

[0054] Figure 1 Fig. 1 shows one of the flow diagrams of the analysis method of the on-chip brain neuron population provided by the embodiments of the present application;

[0055] Figure 2 Fig. 2 shows one of the example diagrams of the analysis method of the on-chip brain neuron population provided by the embodiments of the present application;

[0056] Figure 3 Fig. 3 shows another of the example diagrams of the analysis method of the on-chip brain neuron population provided by the embodiments of the present application;

[0057] Figure 4 Fig. 4 shows another of the example diagrams of the analysis method of the on-chip brain neuron population provided by the embodiments of the present application;

[0058] Figure 5 Fig. 5 shows another of the flow diagrams of the analysis method of the on-chip brain neuron population provided by the embodiments of the present application;

[0059] Figure 6 Fig. 6 shows the functional module diagram of the analysis device of the on-chip brain neuron population provided by the embodiments of the present application;

[0060] Figure 7 Fig. 7 shows the block diagram of the simulation platform provided by the embodiments of the present application.

[0061] Fig. 7 shows the block diagram of the simulation platform provided by the embodiments of the present application. DETAILED DESCRIPTION

[0062] The technical solutions of the embodiments of the present application will be described clearly and completely in the following with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0063] The following detailed description of embodiments of the application in the drawings provided is not intended to limit the scope of the application claimed, but merely represents selected embodiments of the application. Based on the embodiments of the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the application.

[0064] It should be noted that the relational terms such as first and second and the like are used solely to distinguish one entity or action from another, without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements recited, but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without more limitations, an element defined by an "including a" statement does not exclude the existence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0065] Referring to Figure 1 is a flowchart of an analysis method of a population of neurons on a brain-on-a-chip provided by an embodiment of the application.

[0066] In step S202, a structural analysis is performed on the neural dynamics model to obtain potential neural structure relationships. The neural structure relationships represent the connection relationships between the neurons in the target population of neurons.

[0067] It can be understood that an embodiment of the application pre-constructs a simulation platform, which pre-stores a neural dynamics model, and the neural dynamics model can be used to simulate a target population of neurons in an in-vitro cultured brain-on-a-chip. Therefore, the neural dynamics model can be used to analyze the population of neurons.

[0068] In this embodiment, a structural analysis can be performed on the target population of neurons simulated by the neural dynamics model by performing a structural analysis on the neural dynamics model, that is, to obtain potential neural structure relationships. The neural structure relationships can reflect the network structure characteristics of the target population of neurons and the connection relationships between the neurons.

[0069] In step S204, a dynamic prediction and analysis are performed on the neural dynamics model according to a pre-set initial state to obtain dynamic prediction information corresponding to each initial state. The dynamic prediction information represents the relationship between the discharge activity of the target population of neurons starting from the initial state and the time change.

[0070] In this embodiment, multiple different initial states can be pre-set to dynamically predict and analyze the neurodynamic model. Specifically, the dynamic evolution of the target neuron population simulated by the neurodynamic model can be analyzed to obtain dynamic prediction information corresponding to each initial state. The dynamic prediction information can reflect the temporal changes in the discharge activity of the target neuron population starting from the initial state.

[0071] Step S206 , performing feasible domain analysis on the neural dynamics model according to the preset stimulation parameter range to obtain corresponding feasible domain information; the feasible domain information represents the state change range of the target neuron population under the stimulation parameter range.

[0072] In this embodiment, a stimulation parameter range of a stimulation type can be pre-set based on analysis needs, and the stimulation type can be electrical stimulation, light stimulation, or drug stimulation, etc., which is not limited in this embodiment of the present invention. Then, based on the stimulation parameter range, a feasible domain analysis is performed on the neurodynamic model, that is, the response of the target neuron population simulated by the neurodynamic model when subjected to stimulation of corresponding intensity is analyzed to obtain the corresponding feasible domain information. This feasible domain information can reflect the state changes that occur when the target neuron population is stimulated within a certain range.

[0073] Step S208: The neural structure relationship, all dynamic prediction information, and feasible domain information are used as the analysis results of the target neuron group.

[0074] In this embodiment, the neural structure relationship is obtained through structural analysis, dynamic prediction information is obtained through dynamic prediction and analysis, and feasible domain information is obtained through feasible domain analysis, that is, the analysis results of the target neuron group are obtained.

[0075] Based on the above steps, a structural analysis of the neural dynamics model is performed to obtain potential neural structural relationships. These neural structural relationships represent the connection relationships between neurons in the target neuron population. Based on each preset initial state, the neural dynamics model is dynamically predicted and analyzed to obtain dynamic prediction information corresponding to each initial state. The dynamic prediction information represents the relationship between the discharge activity of the target neuron population starting from the initial state and the time-varying relationship. Based on the preset stimulation parameter range, a feasible domain analysis of the neural dynamics model is performed to obtain the corresponding feasible domain information. The feasible domain information represents the range of state changes of the target neuron population within the stimulation parameter range. The neural structural relationships, all dynamic prediction information, and feasible domain information are used as the analysis results of the target neuron population. Simulating and analyzing the neuron population through the neural dynamics model simplifies the analysis of the neuron population and improves its efficiency.

[0076] Optionally, for step S202, an embodiment of the present invention provides a possible implementation method.

[0077] Step S202-1, obtain the connection matrix from the model parameters of the neural dynamics model. The connection matrix is ​​expressed as follows:

[0078]

[0079] Among them, w ij Represents the connection strength between the i-th neuron and the j-th neuron in the target neuron population.

[0080] Step S202-3: Calculate the eigenvalues ​​of the connection matrix, analyze the distribution of the eigenvalues ​​to obtain a spectrum distribution map, and use the spectrum distribution map as the neural structure relationship.

[0081] It is understandable that since the neurodynamic model is used to simulate the target neuron population, some of the model parameters of the neurodynamic model can reflect the characteristics of the target neuron population. The embodiments of the present invention utilize this to perform structural analysis using the model parameters of the neurodynamic model.

[0082] In this embodiment, a connection matrix can be obtained from the model parameters of the neurodynamic model. The connection matrix can reflect the connection strength between each neuron in the target neuron population. For example, if there is a connection between two neurons, the element at the corresponding position in the connection matrix is ​​a non-zero element. And this connection strength can also be understood as a weight. Since neurons transmit signals through neurotransmitters, the size of the connection strength is the size of the weight, which determines the strength of signal transmission between neurons.

[0083] Then, based on the connectivity matrix, the eigenvalues ​​are calculated and their distribution is analyzed. This distribution is then graphically displayed, yielding a spectral distribution diagram, which serves as the neural structure. It can be understood that the distribution of eigenvalues ​​shown in the spectral distribution diagram can be used to analyze information such as the stability, robustness, and oscillatory behavior of the target neuron population.

[0084] For ease of understanding, the present invention provides an example of a spectrum distribution diagram. Figure 2 . Among them, the horizontal coordinate in the spectral distribution diagram represents the real part of the eigenvalue, and the vertical coordinate represents the imaginary part of the eigenvalue. Based on the fact that the eigenvalues ​​in the spectral distribution diagram are mainly concentrated in the area where the real part is less than or equal to zero, this indicates that the target neuron population has good stability. Based on the fact that the distribution of eigenvalues ​​in the spectral distribution diagram is compact, this indicates that the target neuron population has a certain robustness to small disturbances. Based on the fact that the eigenvalues ​​in the spectral distribution diagram are mainly concentrated near the real axis and there is no significant imaginary part, this indicates that the target neuron population will not exhibit strong oscillation behavior.

[0085] Optionally, for step S204, an embodiment of the present invention provides a possible implementation method.

[0086] Step S204-1: For any initial state, a neural dynamics model is used to make predictions based on the initial state to obtain multiple neural state sequences.

[0087] Step S204 - 3 , performing signal analysis and dimensionality reduction operations on multiple neural state sequences to obtain a dynamic behavior matrix, and using the dynamic behavior matrix as dynamic prediction information corresponding to the initial state.

[0088] Step S204-5: traverse each initial state to obtain dynamic prediction information corresponding to each initial state.

[0089] In this embodiment, for each initial state, a neural dynamics model can be used to predict the initial state to obtain multiple neural state sequences. These multiple neural state sequences can be understood as the predicted activity states of the target neuron population in the future if it is active from the initial state.

[0090] These multiple neural state sequences are then subjected to signal analysis and dimensionality reduction. For example, a PCA (Principal Component Analysis) algorithm can be used to process these multiple neural state sequences, extracting the principal components and converting them into low-dimensional data, such as 2D or 3D data. This results in a dynamic behavior matrix, which is used as the dynamic prediction information corresponding to the initial state. Similarly, each initial state is processed to obtain the dynamic prediction information corresponding to each initial state.

[0091] It is understandable that in order to facilitate the intuitive display of dynamic prediction information, visualization operations can be performed based on the dynamic behavior matrix to display the spatial distribution of these main components. For example, analyzing whether its dynamic information presents topological structures such as limit cycle diagrams or attractor diagrams is convenient for analyzing the dynamic characteristics of neuron groups. For ease of understanding, the embodiment of the present invention provides an example low-dimensional manifold diagram for on-chip brain neural activity, please refer to Figure 3 The three coordinate axes in this low-dimensional manifold example represent the first, second, and third principal components. Furthermore, the behavioral trajectory shown in this low-dimensional manifold example is closed and cyclical, indicating that the discharge activity of the target neuron population changes periodically over time, and its discharge pattern is regular, i.e., periodic.

[0092] Optionally, for the process of predicting based on the initial state by using the neural dynamics model to obtain the plurality of neural state sequences in step S204-1, an embodiment of the present application provides a possible implementation manner.

[0093] In step S204-1-1, the neural dynamics model is used to predict based on the initial state to obtain a first neural state sequence, and the first neural state sequence is taken as a current neural state sequence.

[0094] In step S204-1-3, the neural dynamics model is used to predict based on the current neural state sequence to obtain a next neural state sequence.

[0095] In step S204-1-5, a total number of the obtained neural state sequences is counted, and the total number is compared with a preset value.

[0096] In step S204-1-7A, if the total number does not reach the preset value, the next neural state sequence is taken as a new current neural state sequence, and then step S204-1-3 is repeated.

[0097] In step S204-1-7B, if the total number reaches the preset value, all the neural state sequences are obtained to obtain the plurality of neural state sequences.

[0098] It can be understood that the neural dynamics model processes each initial state in a similar manner, and for brevity, an initial state is taken as an example for description below.

[0099] In the embodiment, the neural dynamics model is used to predict based on the initial state to obtain a first neural state sequence; then the neural dynamics model is used to predict based on the first neural state sequence to obtain a second neural state sequence; then the neural dynamics model is used to predict based on the second neural state sequence to obtain a third neural state sequence. That is, the neural dynamics model is used to predict based on an nth neural state sequence to obtain an (n+1) th neural state sequence, and the cycle is repeated until the total number of the obtained neural state sequences reaches a preset value, and then the plurality of neural state sequences are obtained.

[0100] It can be understood that, by giving an initial state and using the neural dynamics model to continuously predict based on the initial state, an embodiment of the present application can predict a next neural state sequence according to a last output neural state sequence. Thus, the activity state of a target neuron group in a future long period of time can be simulated.

[0101] Optionally, for step S206, an embodiment of the present application provides a possible implementation manner.

[0102] Step S206 - 1 : generating a stimulation parameter sequence including a plurality of stimulation parameters based on the stimulation parameter range.

[0103] Step S206-3: Perform stimulation operations on the neural dynamics model according to each stimulation parameter to obtain a neural response sequence corresponding to each stimulation parameter.

[0104] Step S206-5: Perform state analysis based on the neural response sequence corresponding to each stimulation parameter to obtain the neural state vector corresponding to each stimulation parameter, and use the set containing all neural state vectors as feasible domain information.

[0105] In this embodiment, based on a preset stimulation parameter range, multiple stimulation parameters may be generated at certain intervals based on the minimum and maximum values ​​of the stimulation parameter range and the multiple stimulation parameters may be sorted to obtain a stimulation parameter sequence.

[0106] Then, based on the stimulation parameter sequence, we select one stimulation parameter at a time to stimulate the neural dynamics model and output the corresponding neural response sequence. This simulates the activity state of the target neuron population after receiving the corresponding stimulation. Based on each stimulation parameter, we can then obtain the neural response sequence corresponding to each stimulation parameter.

[0107] Based on the neural response sequence corresponding to each stimulation parameter, a state analysis is performed to obtain the neural state vector corresponding to each stimulation parameter. For example, the neural state vector may include the firing rate of three channels, where the firing rate is a parameter used to represent the frequency of neuronal population activity. In other words, the neural state vector represents the activity state of the neuronal population in three dimensions. Obtaining the neural state vector corresponding to all stimulation parameters is to obtain the feasible domain information.

[0108] It is understandable that in order to facilitate the intuitive display of feasible domain information, visualization operations can be performed based on all the obtained neural state vectors to show the state change range of the neuron group. For ease of understanding, the embodiment of the present invention provides an example diagram, please refer to Figure 4 The three coordinate axes in this example graph represent the firing rate of the first channel, the firing rate of the second channel, and the firing rate of the third channel, respectively. Different colors represent different stimulation types, and different color ranges of the same color represent the feasible domain corresponding to different stimulation ranges under the same stimulation type, that is, the range of state changes of the neuronal population.

[0109] Optionally, for step S206-3, an embodiment of the present invention provides a possible implementation method.

[0110] Step S206-3-1, taking the first stimulation parameter in the stimulation parameter sequence as the target stimulation parameter;

[0111] Step S206-3-3, using the neural dynamics model to make predictions based on the preset original state, obtaining an original neural state sequence, and using the original neural state sequence as the current neural response sequence.

[0112] Step S206-3-5: Generate a target stimulation sequence based on the target stimulation parameters, and use the neural dynamics model to predict based on the target stimulation sequence and the current neural response sequence to obtain the next neural response sequence.

[0113] Step S206-3-7: Use the next neural response sequence as the neural response sequence corresponding to the target stimulation parameters.

[0114] Step S206-9A, if the target stimulation parameter is not the last stimulation parameter of the stimulation parameter sequence, then the next stimulation parameter of the target stimulation parameter is used as the new target stimulation parameter, and the next neural response sequence is used as the new current neural response sequence, and then step S206-3-5 is repeated.

[0115] Step S206-9B: if the target stimulation parameter is the last stimulation parameter in the stimulation parameter sequence, then a neural response sequence corresponding to each stimulation parameter is obtained.

[0116] In this embodiment, the first stimulation parameter in the stimulation parameter sequence is first obtained, and the neural dynamics model is used to predict based on the preset original state to obtain the original neural state sequence. Then, the first stimulation sequence is generated based on the first stimulation parameter. For example, if the first stimulation parameter is 5Hz and 10mV, the first stimulation sequence generated is (10, 10, 10, 10, 10), which represents an electrical stimulation pulse with a duration of 1 second and an intensity of 10mV every 200 milliseconds. And the neural dynamics model is used to predict based on the first stimulation sequence and the original neural state sequence to obtain the neural response sequence corresponding to the first stimulation sequence.

[0117] Then, a second stimulation sequence is generated based on the second stimulation parameter, and the neural dynamics model is used to predict the neural response sequence corresponding to the second stimulation parameter and the neural response sequence corresponding to the first stimulation sequence. That is, the neural dynamics model is continuously used to predict the stimulation sequence generated based on the nth stimulation parameter and the neural response sequence corresponding to the n-1th stimulation parameter to obtain the neural response sequence corresponding to the nth stimulation parameter. This cycle is repeated until all stimulation parameters in the stimulation parameter sequence are traversed, and the neural response sequence corresponding to each stimulation parameter is obtained.

[0118] It can be understood that, the embodiment of the present application is to simulate the multiple stimulation operations on the target neuron group by sequentially inputting multiple stimulation sequences to the neural dynamics model, and to predict the next neural response sequence based on the last output neural response sequence and the current input stimulation sequence by using the neural dynamics model. Thus, the reaction of the target neuron group after receiving the cumulative number of stimulations can be simulated.

[0119] Alternatively, based on the above neural dynamics model, the embodiment of the present application further provides an implementation of obtaining the neural dynamics model, please refer to Figure 5 It should be understood that, the embodiment of the present application is to train the neural dynamics model according to the following steps, and then to realize the above-mentioned analysis method of the on-chip brain neuron group by using the neural dynamics model.

[0120] Step S212, a plurality of neural state sequence samples are obtained, and each neural state sequence sample is split into a first neural state sequence and a second neural state sequence with equal length.

[0121] Step S214, a plurality of neural response sample groups are obtained; each neural response sample group includes a stimulation sequence, an original neural state, and a real neural response sequence.

[0122] Step S216, for each first neural state sequence, a third neural state sequence is obtained by using the to-be-trained model to predict based on the first neural state sequence, to obtain each third neural state sequence.

[0123] Step S218, for each neural response sample group, an estimated neural response sequence is obtained by using the to-be-trained model to predict based on the stimulation sequence and the original neural state in the neural response sample group, to obtain each estimated neural response sequence.

[0124] Step S220, the to-be-trained model is trained based on all second neural state sequences, all third neural state sequences, all real neural response sequences, and all estimated neural response sequences, to obtain the neural dynamics model.

[0125] In the embodiment, first, the real activity state of the neuron group in the on-chip brain cultured in vitro can be recorded, and then a plurality of neural state sequence samples are obtained. Each neural state sequence sample is split, i.e., the first half of the neural state sequence sample is taken as the first neural state sequence, and the second half is taken as the second neural state sequence, to obtain the first neural state sequence and the second neural state sequence with the same length. That is, the first neural state sequence and the second neural state sequence represent the same length.

[0126] Multiple stimulation sequences can then be set based on different stimulation types. For each stimulation sequence, the neuronal population in the cultured brain-on-chip is stimulated using the stimulation sequence, and the state of the neuronal population before stimulation is recorded to obtain the original neural state. The activity state of the neuronal population after stimulation is also recorded over a period of time to obtain the true neural response sequence. A neural response sample group consisting of the stimulation sequence, the original neural state, and the true neural response sequence is then obtained, i.e., each neural response sample group is obtained.

[0127] The model to be trained is then used to make predictions based on each first neural state sequence to obtain each third neural state sequence. Furthermore, the model to be trained is used to make predictions based on the stimulus sequence and original neural state in each neural response sample set to obtain each estimated neural response sequence. Finally, the model to be trained is trained based on all second neural state sequences, all third neural state sequences, all true neural response sequences, and all estimated neural response sequences to obtain a neural dynamics model.

[0128] It is understood that the model to be trained can adopt an encoder-recurrent neural network-decoder structure. The recurrent neural network (RNN) can process sequential data and has a memory function. The encoder is composed of two layers of neural networks and is used to encode the activity state of the neuron group to obtain the initial hidden state; the recurrent neural network is used to generate multi-step hidden states based on the initial hidden state prediction and the stimulus signal; the encoder is used to decode the multi-step hidden states to output the state sequence of the neuron group.

[0129] Furthermore, during model training, the recurrent neural network primarily learns the state transition rule f, which has biological information priors. This is based on the Dale principle to constrain the weighted connections of the recurrent neural network. This weighted connection can be masked to classify some units in the recurrent neural network into multiple excitatory neuron groups or multiple inhibitory neuron groups. The constraints on the weighted connection are as follows:

[0130] (1) Non-negativity constraints on the input weight matrix Win and the output weight matrix Wout. That is, all elements in the input weight matrix Win (the weight matrix from the input layer to the hidden layer of the recurrent neural network) are non-negative, and all elements in the output weight matrix Wout (the weight matrix from the hidden layer to the output layer of the recurrent neural network) are non-negative. This constraint can be achieved by using the ReLU (Rectified Linear Unit) activation function.

[0131] (2) Iterative constraints on the cyclic connection weight matrix Wrec (i.e., the connection matrix mentioned above) within the hidden layer:

[0132]

[0133] in, represents the connection matrix after iteration; W rec represents the connection matrix before iteration; [W rec ] + Represents the connection matrix W before iteration using ReLU activation rec Calculation is performed to satisfy the non-negative constraint condition; D represents a diagonal matrix, the elements on the diagonal of the diagonal matrix are 1 and -1, and the ratio between the total number of elements that are 1 and the total number of elements that are -1 is 4:1.

[0134] The Dale principle states that a neuron typically releases only one type of neurotransmitter: either excitatory or inhibitory. Using 1 to represent excitability and -1 to represent inhibition, the diagonal matrix D constrains the ratio of excitatory to inhibitory neurons in the simulated neuron population to 4:1. By combining nonnegative constraints and the Dale principle, the model simulates both excitatory and inhibitory neural connections, improving its predictive capabilities.

[0135] In order to execute the corresponding steps in the above embodiments and various possible methods, an implementation method of an on-chip brain neuron population analysis device is provided below. Figure 6 , is a functional module diagram of an on-chip brain neuron population analysis device provided in an embodiment of the present invention. It should be noted that the basic principles and technical effects of the on-chip brain neuron population analysis device 300 provided in this embodiment are the same as those of the above-mentioned embodiments. For the sake of simplicity, any matters not mentioned in this embodiment can be referred to the corresponding contents of the above-mentioned embodiments. The on-chip brain neuron population analysis device 300 includes:

[0136] The neural analysis module 310 is used to perform structural analysis on the neural dynamics model to obtain potential neural structural relationships; the neural structural relationships represent the connection relationships between neurons in the target neuron population;

[0137] Based on each preset initial state, the neural dynamics model is dynamically predicted and analyzed to obtain dynamic prediction information corresponding to each initial state; the dynamic prediction information represents the relationship between the discharge activity of the target neuron group starting from the initial state and the change over time;

[0138] Based on the preset stimulation parameter range, the feasible domain analysis of the neural dynamics model is performed to obtain the corresponding feasible domain information; the feasible domain information represents the state change range of the target neuron population under the stimulation parameter range;

[0139] The result acquisition module 330 is used to take the neural structure relationship, all dynamic prediction information and feasible domain information as the analysis results of the target neuron group.

[0140] Optionally, the neural analysis module 310 is further configured to obtain a connection matrix from model parameters of the neural dynamics model. The connection matrix is ​​represented as follows:

[0141]

[0142] Among them, w ij Represents the connection strength between the i-th neuron and the j-th neuron in the target neuron population; calculates the eigenvalues ​​of the connection matrix, and analyzes the distribution of the eigenvalues ​​to obtain a spectral distribution map, and uses the spectral distribution map as the neural structure relationship.

[0143] Optionally, the neural analysis module 310 is also used to: for any initial state, use the neural dynamics model to make predictions based on the initial state to obtain multiple neural state sequences; perform signal analysis and dimensionality reduction operations on the multiple neural state sequences to obtain a dynamic behavior matrix, and use the dynamic behavior matrix as the dynamic prediction information corresponding to the initial state; traverse each initial state to obtain the dynamic prediction information corresponding to each initial state.

[0144] Optionally, the neural analysis module 310 is also used to: use the neural dynamics model to make predictions based on the initial state to obtain a first neural state sequence, and use the first neural state sequence as the current neural state sequence; use the neural dynamics model to make predictions based on the current neural state sequence to obtain the next neural state sequence; count the total number of neural state sequences obtained, and compare the total number with a preset value; if the total number does not reach the preset value, use the next neural state sequence as the new current neural state sequence, and repeat the process of using the neural dynamics model to make predictions based on the current neural state sequence to obtain the next neural state sequence; if the total number reaches the preset value, obtain all neural state sequences to obtain multiple neural state sequences.

[0145] Optionally, the neural analysis module 310 is also used to: generate a stimulation parameter sequence including multiple stimulation parameters based on the stimulation parameter range; perform stimulation operations on the neural dynamics model according to each stimulation parameter to obtain a neural response sequence corresponding to each stimulation parameter; perform state analysis based on the neural response sequence corresponding to each stimulation parameter to obtain a neural state vector corresponding to each stimulation parameter, and use the set containing all neural state vectors as feasible domain information.

[0146] Optionally, the neural analysis module 310 is further configured to: take a first stimulation parameter in the stimulation parameter sequence as a target stimulation parameter; predict, based on the preset original state, an original neural state sequence by using the neural dynamics model, and take the original neural state sequence as a current neural response sequence; generate a target stimulation sequence based on the target stimulation parameter, and predict, based on the target stimulation sequence and the current neural response sequence, a next neural response sequence by using the neural dynamics model; take the next neural response sequence as a neural response sequence corresponding to the target stimulation parameter; if the target stimulation parameter is not a last stimulation parameter in the stimulation parameter sequence, take a next stimulation parameter of the target stimulation parameter as a new target stimulation parameter, and after taking the next neural response sequence as a new current neural response sequence, repeat the operations of generating a target stimulation sequence based on the target stimulation parameter, and predicting, based on the target stimulation sequence and the current neural response sequence, a next neural response sequence by using the neural dynamics model; and if the target stimulation parameter is the last stimulation parameter in the stimulation parameter sequence, obtain the neural response sequence corresponding to each stimulation parameter.

[0147] Referring to Figure 7 is a block schematic diagram of the simulation platform provided by the embodiment of the present application. The simulation platform 100 comprises a bus 110, a processor 120, a memory 130, an I / O module 150, and a communication module 170.

[0148] The bus 110 is a circuit for connecting the above-mentioned elements to each other and transmitting signals between the above-mentioned elements.

[0149] The processor 120 can receive a command from the above-mentioned other elements (for example, the memory 130, the I / O module 150, the communication module 170, etc.) through the bus 110, can interpret the received command, and can perform calculation or data processing according to the interpreted command. The processor 120 can be an integrated circuit chip with signal processing capability. The processor 120 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a ready-to-use programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0150] The memory 130 can store commands or data received from the processor 120 or other components (e.g., the I / O module 150, the communication module 170, etc.), or commands or data generated by the processor 120 or other components. The memory 130 can be, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), or an electrically erasable programmable read-only memory (EEPROM).

[0151] The I / O module 150 can receive commands or data input from the user via input-output means (e.g., sensors, keyboards, touch screens, etc.), and can transmit the received commands or data to the processor 120 or the memory 130 via the bus 110. It is also used to display various information (e.g., multimedia data, text data) received, stored, and processed by the above components, and can display videos, images, data, etc. to the user.

[0152] The communication module 170 may be used to communicate signals or data with other devices.

[0153] It is understandable that Figure 7 The structure shown is only a schematic diagram of the structure of the simulation platform 100. The simulation platform 100 may also include Figure 7 More or fewer components than shown, or with Figure 7 Different configurations shown. Figure 7 Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0154] The simulation platform provided by the embodiment of the present invention may be a personal computer, a tablet computer, a netbook, etc., and the embodiment of the present invention is not limited thereto.

[0155] The memory in the simulation platform provided by the embodiment of the present invention stores a computer program, and when the processor executes the computer program, the on-chip brain neuron population analysis method disclosed by the embodiment of the present invention is implemented.

[0156] An embodiment of the present invention further provides a storage medium storing a computer program. When the computer program is executed by a processor, the method for analyzing a brain-on-chip neuron population disclosed in an embodiment of the present invention is implemented.

[0157] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of the devices, methods, and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or part of the code, which contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, as well as the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or actions, or can be implemented using a combination of dedicated hardware and computer instructions.

[0158] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0159] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0160] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A method for analyzing a brain-on-a-chip neuron population, characterized in that: Applied to a simulation platform, the simulation platform pre-stores a neural dynamics model for simulating a target neuron population, and the analysis method of the brain-on-chip neuron population includes: Performing structural analysis on the neural dynamics model to obtain potential neural structural relationships; the neural structural relationships represent the connection relationships between neurons in the target neuron population; According to each preset initial state, the neural dynamics model is dynamically predicted and analyzed to obtain dynamic prediction information corresponding to each initial state; the dynamic prediction information represents the relationship between the discharge activity of the target neuron population starting from the initial state and time; Performing feasible domain analysis on the neural dynamics model according to a preset stimulation parameter range to obtain corresponding feasible domain information; the feasible domain information represents the state change range of the target neuron population under the stimulation parameter range; Taking the neural structure relationship, all dynamic prediction information and the feasible domain information as the analysis results of the target neuron population; The neural dynamics model is obtained in the following manner: Acquire multiple neural state sequence samples, and split each of the neural state sequence samples into a first neural state sequence and a second neural state sequence of equal length; Acquire multiple neural response sample groups; each of the neural response sample groups includes a stimulation sequence, an original neural state, and a true neural response sequence; For each of the first neural state sequences, using the to-be-trained model to predict based on the first neural state sequence to obtain a third neural state sequence, thereby obtaining each third neural state sequence; For each of the neural response sample groups, using the to-be-trained model to make predictions based on the stimulation sequence and the original neural state in the neural response sample group to obtain an estimated neural response sequence, thereby obtaining each estimated neural response sequence; The model to be trained is trained based on all second neural state sequences, all third neural state sequences, all true neural response sequences, and all estimated neural response sequences to obtain the neural dynamics model.

2. The method for analyzing brain-on-a-chip neuron populations according to claim 1, wherein: Each neuron in the target neuron population has a corresponding number; The step of performing structural analysis on the neural dynamics model to obtain potential neural structural relationships includes: The connection matrix is ​​obtained from the model parameters of the neural dynamics model, and the connection matrix is ​​expressed as follows: ； in, represents the connection strength between the i-th neuron and the j-th neuron in the target neuron population; The eigenvalues ​​of the connection matrix are calculated, and the distribution of the eigenvalues ​​is analyzed to obtain a spectral distribution map, and the spectral distribution map is used as the neural structure relationship.

3. The method for analyzing brain-on-a-chip neuron populations according to claim 1, wherein: The step of dynamically predicting and analyzing the neural dynamics model according to each preset initial state to obtain dynamic prediction information corresponding to each initial state includes: For any of the initial states, using the neural dynamics model to make predictions based on the initial state to obtain multiple neural state sequences; Performing signal analysis and dimensionality reduction operations on the multiple neural state sequences to obtain a dynamic behavior matrix, and using the dynamic behavior matrix as dynamic prediction information corresponding to the initial state; Traverse each of the initial states to obtain dynamic prediction information corresponding to each of the initial states.

4. The method for analyzing brain-on-a-chip neuron populations according to claim 3, wherein: The step of using the neural dynamics model to predict based on the initial state to obtain multiple neural state sequences includes: Using the neural dynamics model to make predictions based on the initial state to obtain a first neural state sequence, and using the first neural state sequence as a current neural state sequence; Using the neural dynamics model to predict based on the current neural state sequence, to obtain a next neural state sequence; Counting the total number of neural state sequences obtained, and comparing the total number with a preset value; If the total number does not reach the preset value, after taking the next neural state sequence as a new current neural state sequence, repeatedly performing the step of using the neural dynamics model to predict based on the current neural state sequence to obtain the next neural state sequence; If the total number reaches the preset value, all neural state sequences are acquired to obtain the multiple neural state sequences.

5. The method for analyzing brain-on-a-chip neuron populations according to claim 1, wherein: The step of performing feasible domain analysis on the neural dynamics model according to a preset stimulation parameter range to obtain corresponding feasible domain information includes: generating a stimulation parameter sequence including a plurality of stimulation parameters based on the stimulation parameter range; performing a stimulation operation on the neural dynamics model according to each of the stimulation parameters to obtain a neural response sequence corresponding to each of the stimulation parameters; A state analysis is performed based on the neural response sequence corresponding to each of the stimulation parameters to obtain a neural state vector corresponding to each of the stimulation parameters, and a set containing all neural state vectors is used as the feasible domain information.

6. The method for analyzing brain-on-a-chip neuron populations according to claim 5, characterized in that: The step of performing a stimulation operation on the neural dynamics model according to each stimulation parameter to obtain a neural response sequence corresponding to each stimulation parameter includes: using the first stimulation parameter in the stimulation parameter sequence as the target stimulation parameter; Using the neural dynamics model to make predictions based on a preset original state to obtain an original neural state sequence, and using the original neural state sequence as a current neural response sequence; generating a target stimulation sequence based on the target stimulation parameters, and using the neural dynamics model to predict based on the target stimulation sequence and the current neural response sequence to obtain a next neural response sequence; using the next neural response sequence as the neural response sequence corresponding to the target stimulation parameter; If the target stimulation parameter is not the last stimulation parameter of the stimulation parameter sequence, then after using the next stimulation parameter of the target stimulation parameter as a new target stimulation parameter and the next neural response sequence as a new current neural response sequence, repeatedly performing the steps of generating a target stimulation sequence based on the target stimulation parameter, and performing prediction based on the target stimulation sequence and the current neural response sequence using the neural dynamics model to obtain a next neural response sequence; If the target stimulation parameter is the last stimulation parameter of the stimulation parameter sequence, a neural response sequence corresponding to each stimulation parameter is obtained.

7. An on-chip brain neuron population analysis device, characterized in that: Applied to a simulation platform, the simulation platform pre-stores a neural dynamics model for simulating a target neuron population, and the analysis device of the brain-on-chip neuron population includes: A neural analysis module, configured to perform structural analysis on the neural dynamics model to obtain potential neural structural relationships; the neural structural relationships represent the connection relationships between neurons in the target neuron population; According to each preset initial state, the neural dynamics model is dynamically predicted and analyzed to obtain dynamic prediction information corresponding to each initial state; the dynamic prediction information represents the relationship between the discharge activity of the target neuron population starting from the initial state and time; Performing feasible domain analysis on the neural dynamics model according to a preset stimulation parameter range to obtain corresponding feasible domain information; the feasible domain information represents the state change range of the target neuron population under the stimulation parameter range; A result acquisition module, configured to use the neural structure relationship, all dynamic prediction information, and the feasible domain information as analysis results of the target neuron population; The neural dynamics model is obtained in the following manner: Acquire multiple neural state sequence samples, and split each of the neural state sequence samples into a first neural state sequence and a second neural state sequence of equal length; Acquire multiple neural response sample groups; each of the neural response sample groups includes a stimulation sequence, an original neural state, and a true neural response sequence; For each of the first neural state sequences, using the to-be-trained model to predict based on the first neural state sequence to obtain a third neural state sequence, thereby obtaining each third neural state sequence; For each of the neural response sample groups, using the to-be-trained model to make predictions based on the stimulation sequence and the original neural state in the neural response sample group to obtain an estimated neural response sequence, thereby obtaining each estimated neural response sequence; The model to be trained is trained based on all second neural state sequences, all third neural state sequences, all true neural response sequences, and all estimated neural response sequences to obtain the neural dynamics model.

8. A simulation platform, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and when the processor executes the computer program, the method for analyzing a brain-on-chip neuron population according to any one of claims 1 to 6 is implemented.

9. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, implements the method for analyzing a brain-on-a-chip neuron population according to any one of claims 1 to 6.

Images