Cognitive calculation method and system based on biological neural network

Abstract

The invention discloses a cognitive calculation method and system based on a biological neural network. A biological neural network BNN core unit comprises a nerve cell culture, an input stimulation unit and an output readout unit, and can be controlled through various life cycles of the nerve cell culture, so as to provide a data processing function. An automated system includes environmental and chemical controller units adapted to operate with BNN stimulation and readout data interfaces, facilitating monitoring and adjustment of BNN core unit parameters. The pre-processing and post-processing of the BNN interface signal can further promote the training and reinforcement learning of the BNN. A plurality of BNN core units may also be assembled together as a stack. The proposed system provides a BNN operating system as a core component for a wet piece server to receive, process, and send data for different client applications without exposing the BNN core unit component to client users, while improving the performance of the system as compared to conventional silicon-based hardware and software information processing for advanced cognitive computing tasks. And obviously less energy is needed.

Description

[0001] This application is a divisional application of the patent application filed on September 8, 2019, with application number 2019800584579 and invention title "Cognitive Computing Method and System Based on Biological Neural Network". Technical Field

[0002] This disclosure relates to cognitive computing systems, methods, and processes for performing various advanced and complex cognitive tasks that mimic and extend the functions of the biological brain. The proposed cognitive computing systems, methods, and processes employ multiple neural cells as core biotechnological processing elements that integrate with on-chip brain interfaces and controllers to complement and interact with more traditional information technology networks and computing architectures. Background Technology

[0003] With the tremendous development of information technology (IT) over the past few decades, numerous methods and systems are now available for performing a wide range of computational tasks, such as computation, data optimization, data classification, natural language processing and translation, and image and video processing and recognition. Recent advancements in cognitive computing include machine learning, deep learning, and artificial neural computing (ANLC), which mimics biological neural networks. ANC further seeks to provide artificial intelligence to humans in homes, social media, and workplaces, as well as in harsh environments like the deep sea and space, to assist with tasks such as object and facial recognition, natural language processing (NLP) and sentiment analysis, and even nuclear reactors. Consequently, large companies like Google (Google Cloud Platform), Amazon (Amazon Web Services (AWS), and Microsoft (Azure) offer high-performance computing, high-throughput computing, and highly available systems and infrastructure as cloud computing services. Facebook and Apple also operate their own private data centers on carefully selected data farm sites worldwide, where they can benefit from affordable and reliable electricity. However, a major limitation of current silicon-based computing environments (including software and hardware) is the excessive power required to perform complex cognitive tasks. Silicon-based systems consume an order of magnitude more power than biological systems with similar performance, such as the brain. Another limitation is the requirement for at least some explicit logic programming and structured data representation, which makes them unsuitable today for realizing concepts that have not yet been understood and modeled by human science, such as higher-level cognitive processes, creative thinking, and consciousness.

[0004] Neural networks can be viewed as a means of creating a spatiotemporal mapping between two spaces of different dimensions. In configurations where the number of outputs is less than the number of inputs, neural networks thus create simplified expressions for problems in spaces with finite dimensions. Well-known spatial transformations include: the Hough transform, which maps a two-dimensional space to a two-parameter voting space; the Fourier transform, which maps a time-periodic signal to its frequency representation; and wavelet transforms, which map an image to a scale-space representation. While these transformations are well-described mathematically and can be directly implemented, even though neural networks can be used for this, they are unlikely to be the most computationally efficient choice. However, explicit mathematical solutions are often overly complex, for example, in cases where a 1000-dimensional multivariate input space produces only 10 relevant variables at output. Neural networks offer a good approach to these problems. Nevertheless, the internal state space dimension required to solve such problems can be enormous and demands excessive computational resources with current hardware technology.

[0005] As an alternative to traditional software and hardware information technology, as early as the late 20th century, scholars primarily explored wet software solutions based on biological components (such as cultured cells) rather than transistors. https: / / www.technologyreview.com / s / 400707 / biologicalcomputing / To date, most scientists have focused on how to implement basic functions similar to the core IT processing logic gates, computation, and memory storage, such as DNA computing. https: / / www.nature.com / subjects / dna-computing Despite the promising future of this path, especially with the recent advancements in synthetic biology and DNA editing technologies, the engineering pathways for constructing higher-order cognitive processes from these fundamental functions remain just as challenging as traditional silicon-based logic.

[0006] Examples of advanced cognitive functions are those required for a typical intelligent machine to pass the Turing test. Current research in this area using artificial neural networks (ANNs) faces at least two major limitations:

[0007] Compared to biological neural networks (BNNs), existing multilayer networks used for deep learning lack temporal flexibility. The human brain is essentially a biological multinucleus system, and no mathematical model is available. Most of the powerful mathematics used in the greatest scientific models, such as quantum mechanics or general relativity, is useless in this context. Existing recurrent spike neural networks (SNNs) cannot be trained on complex tasks or simulated on a large scale, primarily due to the lack of computational efficiency in digital processing. Since all neurons work in parallel by nature, they essentially represent up to 100 billion processors (for example, let's assume the human brain has an average of 100 billion neurons). Replicating this processing power would be possible if each neuron represented only one floating-point operation, but in reality, the accuracy of the neuron simulations required to achieve the Turing test target is unknown: if each neuron is on the order of 100 Mflops, it is impossible to achieve with current technology. In fact, the best computing power is 1E16 Flops and 100E9 x 100E6 = 1E11 x 1E8 = 1E19, which is about 1000 times greater than the computing power of the world's best high-performance computing systems. Even non-real-time simulations are too slow to yield any meaningful results in a reasonable amount of time. The computational efficiency of a biological brain is also orders of magnitude higher than that of a digital computer: for example, a brain typically consumes 20W of 100 billion neurons (5 billion neurons / W), while digital simulations require several orders of magnitude more power, even for simple neuron models such as integral and ignition spike neurons.

[0008] Therefore, a novel alternative approach is to use biological neural networks instead of silicon-based digital computing to replicate high levels of cognitive processes. Recent advances in biotechnology now facilitate the cultivation and assembly of biological neural networks derived from embryonic stem cells, such as rat embryonic stem cells, and differentiated human induced pluripotent stem cells (IPS). CThe cultured BNN can then be stimulated and read using a multi-electrode array (MEA). To date, the primary applications of MEAs have been in developing neuronal and brain models for pharmacological, drug testing, and toxicological studies, as well as in better understanding common brain diseases such as Alzheimer's and Parkinson's. Other industrial applications have also been proposed in recent years. The application of BNNs in aircraft control was, for example, proposed by DeMarse et al. in their 2005 IEEE International Conference on Neural Networks proceedings entitled "Adaptive Flight Control with Live Neuron Networks on Microelectrode Arrays." US Navy Patent 7,947,626 discloses a passaged progenitor cell-derived neural network MEA that can be used to detect and / or quantify various biological or chemical toxins. On the same track, Koniku (www.koniku.com) was the first company to develop a wet-chip computing chip based on cultured neural cells and has been commercializing highly specialized products since 2017 for the detection of odor compounds in the security, military, and agricultural / food markets. As described in their patent application WO 2018 / 081657, neurons can be modified through various biotechnological processes (e.g., gene editing, methylation editing, etc.) to express a unique spectrum of odor receptors with cell surface receptors, as is known in the art. The neurons can be interfaced with a computer via a state-of-the-art neurophysiological interface, such as a MEA (multi-electrode array) electrode. The computer can then measure the electrical signals generated by the neurons when exposed to odorant compounds in a dedicated chamber and detect the presence of certain odorant compounds using conventional signal processing methods, which may include artificial neural networks (ANNs) and machine learning classifiers for training and classification. ANNs are particularly useful for distinguishing multiple signals when used in real-world environments with networks of different neurons possessing different odor receptor properties, potentially involving complex combinations of multiple detectable compounds. A major limitation of this approach is its restriction to very specific sensory applications.

[0009] Baker Hughes' US patent application US20140279772 discloses the use of a cultured biological neural network (BNN) in an apparatus for processing downhole signals transmitted into the formation through a borehole within a container. The BNN is connected to MEA electrodes to receive input signals from sensors and output measurements outside the neural network. The proposed BNN system also includes environmental control components, such as a nutrient dispenser. While this disclosure mentions the advantages of BNNs over traditional computing systems in challenging environments in terms of parallel computing power, robustness to vibration and electrical noise, and self-healing capabilities, it does not explicitly describe how the system is managed over time, particularly the dynamic learning process. Its application-specific nature suggests that preprocessing learning could be applied as an offline preparation process rather than in a challenging runtime environment.

[0010] In his 2013 doctoral dissertation, "Simulating and Cultivating Neural Network Computation," Ju Han from the University of Singapore explored the ability of isolated neuronal cultures from 18-day-old rat embryonic cortical cells within the context of a state-dependent computing paradigm. His research demonstrated that random networks composed of live neurons can handle complex spatiotemporal information, making them suitable for prototyping neural computing devices based on the LSM (Liquid State Machine) paradigm. To test the ability of neuronal cultures to classify temporal and complex spatiotemporal patterns, two types of stimulation were designed: a jittery spike training template classification task (a benchmark for LSM) and a randomly combined piano music classification task. Processing temporal inputs is dependent on memory decay; Ju Han observed that in his setup, short-term memory in isolated neuronal cultures could exceed 4 seconds. To control neuronal culture, Ju Han employed optogenetics (photostimulation of transgenic neurons) combined with a multi-electrode array (MEA) for both stimulating neuronal inputs and recording neuronal outputs. While MEA can provide electrical stimulation (BNN input) and measurement (BNN output) at low spatial and temporal resolution, optogenetics enables more precise stimulation control, contactless manipulation, and repeated interrogation of neurons, thus further developing into an intensive research area over the past five years. Examples of recent advances in optogenetic stimulation research can be found in the following literature: "Optogenetic Approaches in Drug Screening: Techniques and Applications" by Agus and Janovjak in Biotechnology Curr. Opin., December 2017, and "Optogenetic Stimulation and Recording of Primary Cultured Neurons with Spatiotemporal Control" by Barral and Reyes in BioProtoc, August 2017, describing a rapid video projector based on the working principle of a digital micromirror device (DMD) that can spatially focus optical stimulation onto a single neuron while achieving temporal display modes of 1.44 kHz and above.

[0011] To optimize the overall functionality of the BNN system, Ju Han proposed using standard machine learning methods (especially genetic algorithms) to further control the biological culture units, defining electrical stimulation and processing to ultimately achieve higher-level functions. Instead of the individual neuron readouts often considered in neuroscience experiments, network layer outputs can be read and processed as multivariate signals (e.g., using MEA electrophysiological measurement probes combined with signal processing software such as Matlab).

[0012] Further research in neuroscience and neural computation emphasizes that biological neuronal systems require a learning process, much like the brain develops its computational capabilities. In his 2013 doctoral dissertation, Ju Han pointed out that biological neuronal systems can learn or be designed, possibly through light stimulation. This learning ability of networks, combined with drug manipulation, forms possible steps for optimizing neural circuits to perform computations. BNN systems can be trained to generate behaviors specified by a reference model through reinforcement learning, such as releasing global reward or punishment signals based on behavioral outcome feedback. For example, Ju Han proposed using NDMA receptor antagonist drugs to treat reinforcement learning.

[0013] More generally, in addition to open-loop systems, which are still widely used in neuroscience experiments, closed-loop systems can be designed in the same way as in conventional automated systems engineering. Current state-of-the-art BNN systems are tailored for small-scale, very specific applications, and therefore face two main limitations when used as a single component in a general-purpose wet-piece computing system with adaptive (different needs) and evolutionary (over time) advanced cognitive capabilities: their architecture is limited to a single input (single neuron or layer stimulation using predefined settings and protocols) and output layer design (single neuron or layer measurements using predefined biological or signal processing methods); and their inherent network connectivity means they can only process one function at a time, with or without initial training.

[0014] Therefore, new solutions and architectures are needed to leverage the advanced cognitive processing capabilities inherent in biological neural networks as a low-power alternative or complementary processing solution to more traditional silicon-based information technology processing systems, devices, software, and electronic chips. Summary of the Invention

[0015] An automated processing system for converting spatiotemporal input data signals into spatiotemporal output data signals is described. The system includes: an in vitro biological culture of nerve cells (BNN core unit); an input stimulation unit (SU) adapted to apply input spatiotemporal stimulation signals to a first group of nerve cells; an output readout unit (RU) adapted to capture output spatiotemporal readout signals from a second group of nerve cells; one or more nutrient tanks connected to one or more nutrient dispensers to inject one or more nutrients into the biological nerve cell culture; one or more additive tanks, each additive tank connected to one or more additive dispensers to inject one or more additives into the BNN culture; one or more nutrient waste collectors for filtering and discharging nutrient waste from the BNN culture; and one or more additive waste collectors for discharging nutrient waste from the BNN culture. The system includes: a filter for filtering and removing additive waste; one or more vascularized networks connecting the nutrient dispenser; an additive dispenser, a nutrient waste collector, and an additive waste collector for the BNN culture; one or more sensors for measuring at least one environmental parameter of the BNN culture; and an automated controller configured to adapt the stimulus signal to the input data signal, adapt the output data signal to the readout signal, and control at least one of the BNN core unit environmental parameters, BNN neural cell culture nutrient supply, BNN neural cell culture additive supply, BNN neural cell culture nutrient waste collection, and BNN neural cell culture additive waste collection, thereby maintaining the homeostasis of the BNN neural cell culture over time, such that the spatiotemporal input data signal is continuously converted into the spatiotemporal output data signal.

[0016] The automated controller may include a preprocessing unit that uses at least one of the following to transform the input data signal into the stimulus signal: a spatiotemporal signal filter, a spatiotemporal signal classifier, a machine learning algorithm based on a mathematical or statistical model, an artificial neural network, a convolutional neural network, a support vector machine classifier, a random forest classifier, a genetic algorithm, a genetic programming algorithm, or a reservoir calculation method.

[0017] The automated controller may include a post-processing unit for transforming the readout signal into the output data signal using at least one of a spatiotemporal signal filter, a spatiotemporal signal classifier, a machine learning algorithm based on a mathematical or statistical model, an artificial neural network, a convolutional neural network, a support vector machine classifier, a random forest classifier, a genetic algorithm, a genetic programming algorithm, or a reservoir calculation method.

[0018] A method for converting spatiotemporal input data signals into spatiotemporal output data signals using an automated controller and a biological neural network (BNN) core unit, the BNN core unit comprising at least an in vitro culture of nerve cells, the in vitro culture of nerve cells being adapted to feed a stimulus spatiotemporal signal to a first group of nerve cells using an input stimulus unit (SU) and to read out a spatiotemporal signal from a second group of nerve cells using an output readout unit (RU), the method comprising: preprocessing the spatiotemporal input data signals using the automated controller to form the stimulus spatiotemporal signal; postprocessing the readout spatiotemporal signal using the automated controller to form the spatiotemporal output data signal; and controlling at least one of the following: environmental parameters of the BNN core unit, nutrient supply to the BNN nerve cell culture, additive supply to the BNN nerve cell culture, nutrient waste collection to the BNN nerve cell culture, additive waste collection to the BNN nerve cell culture, preprocessing parameters, or postprocessing parameters, thereby maintaining the homeostasis of the BNN nerve cell culture over time, such that the spatiotemporal input data signals are continuously converted into the spatiotemporal output data signals.

[0019] The method may further include minimizing the error between the output data signal and the target output data signal by adjusting at least one of the following: the environmental parameters of the BNN core unit, the nutrient supply for the BNN neural cell culture, the additive supply for the BNN neural cell culture, the nutrient waste collection for the BNN neural cell culture, the additive waste collection for the BNN neural cell culture, preprocessing parameters, or postprocessing parameters. Attached Figure Description

[0020] Figure 1 The core BNN unit is described as a building block of a biological computing server.

[0021] Figure 2a ), Figure 2b ), Figure 2c ), Figure 2d ), Figure 2e Different layout and manufacturing schemes are proposed to assemble the core BNN unit.

[0022] Figure 3 An exemplary schematic diagram of an automated vascularization system (AVS) operating using the BNN core unit is described.

[0023] Figure 4 and Figure 5 The following are side cross-sectional views of two exemplary spongy structures that serve as potential hosts for neural cells with an intrinsic vascularized network supporting growth in 3DBNN cultures.

[0024] Figure 6A schematic diagram of a possible automated BNN growth, maintenance, and control system is shown.

[0025] Figure 7 Another schematic diagram is shown, illustrating the potential for automated BNN growth, maintenance, and control as a biological operating system (BOS).

[0026] Figure 8 The learning process can be implemented using real-time processing software.

[0027] Figure 9a )and Figure 9b This illustrates a possible implementation of the BNN Biological Computing Stack (BCS).

[0028] Figure 10a )and Figure 10b This paper illustrates a possible embodiment of a BNN biological computing stack (BCS) (T-BCS) suitable for training.

[0029] Figure 11 An exemplary T-BCS implementation with ANN is shown.

[0030] Figure 12 The possible maintenance process for T-BCS is shown.

[0031] Figure 13 The host server operating using the T-BCS wet software architecture is shown.

[0032] Figure 14 It describes the possible functions that can be operated by the master server to manage user clients.

[0033] Figure 15 It also describes a general architecture with load balancing for serving multiple clients from the same T-BCS server host.

[0034] Figure 16 The proposed BNN server for image processing is compared with applications featuring deep learning architectures.

[0035] Figure 17 The study showed a neurosphere approximately 200 meters wide four days after rat cortical stem cells matured.

[0036] Figure 18 , Figure 19 , Figure 20 An example of 12 electrodes regularly distributed around the periphery of the neurosphere of cortical neural stem cells at three different growth stages is shown schematically.

[0037] Figure 21 A photograph of a biological neural network on an MEA circuit is shown.

[0038] Figure 22 Adherent cells and Microscopic image of the matrix. Detailed Implementation

[0039] BNN core unit

[0040] Figure 1 The core BNN unit 100 is described as a building block of a biological computing server. The biological material in the BNN unit typically consists of, but is not limited to, an active biological culture 120, which is typically a combination of multiple active nerve and glial cells. Cells can be assembled through various processes, such as, but not limited to, cell culture or organogenesis-like methods. Throughout this disclosure, the term cell culture is used indiscriminately for in vitro cell growth and lifespan maintenance detached from their native in vivo environment. The BNN unit can be arranged in 2D or 3D. A stimulation unit 110 (SU) represents the input interface between the digital data input signal 105 and the biological culture 120. It is used to selectively stimulate different neurons, dendrites, or axons. The stimulation unit 110 can control the digital data input signal transferred to the biological culture by spatially (addressing different neurons), temporally (stimulating with a time-varying signal), and / or spatially-temporally altering the digital data input signal transferred to the biological culture. Examples of implementations of SU 110 include, but are not limited to, multi-electrode arrays (MEAs), patch clamps, photoinduced stimulation (such as optogenetic systems), magnetic or electric fields, ion stimulation, focused lasers, optical tweezers, or mechanically induced stimulation through changes in gravity or pressure. Readout unit 130 represents the output interface between the biological culture 120 and the digital data output signal 135. It is used to selectively measure the activity of different neurons, dendrites, or axons. Readout unit 130 can control the conversion of biological culture activity into a digital data output signal (potentially multidimensional) by sampling it spatially (capturing individual activity of different neurons), temporally (capturing signals that vary over time), and / or spatiotemporally. Examples of implementations of RU130 include, but are not limited to, multi-electrode arrays (MEAs), patch clamps, imaging systems, ion-sensitive sensors, electrical or magnetic-sensitive sensors, chemical sensors, and other sensors suitable for neuron culture.

[0041] A subset of neurons can be excited with an input signal via a BNN input interface such as a multi-electrode array (MEA) for receiving electrical signals. Alternatively, a subset of neurons can be transgenic to receive optical stimulation as an input signal from an optogenetic system. Using a measurement system such as a multi-electrode array (MEA) for receiving electrical signals, the electrical activity of neuronal cells can be monitored at several locations in the biological material as the output of the BNN unit. Optionally, a subset of neurons can be transgenic to express fluorescence as an output signal from the BNN to an imaging sensor system. However, many other systems exist that can realize or facilitate the connection of BNNs, such as processes utilizing concepts similar to those in the human body, such as the conversion of electrochemical stimulation into mechanical motion and observable in muscle movement or speech.

[0042] In a possible embodiment, from Maxwell Biosystems ( https: / / www.mxwbio.com The MaxOne MEA can be used as the host platform for the BNN core unit 100. Isolate cell cultures BNN120 can be electroplated and grown on the MaxOne microsensor in CMOS technology according to the scheme proposed at the following URL: https: / / www.mxwbio.com / applications / neuronal-networks / applications / neuronal-networks /

[0043] Sample cell culture electroplating process

[0044] ○ The electrode array surface was pre-coated with a thin film of 0.05 wt% poly(ethyleneimine) (PEI) (Sigma, Missouri, USA) in 8.5 pH borate buffer (Chemie Brunschwig, Basel, Switzerland).

[0045] ○ Add one drop of 0.02 mg / ml laminin (Sigma) to the culture medium (Ingenie, California, USA) for cell adhesion.

[0046] ○ The seed cell suspension was dropped onto the array in 6-1 droplets.

[0047] ○ Add 1ml of electroplating medium after 20-30 minutes.

[0048] ○ After 24 hours, replace the plate medium with 1-2 ml of growth medium and keep the culture in the incubator to control the environmental conditions (37°C, 65% humidity, 5% carbon dioxide).

[0049] ○ Change 50% of the growth medium twice a week.

[0050] In the protocol proposed by Maxwell Biosystems, the plate culture medium can consist of 850 ml of medium supplemented with 10% horse serum (HyClone, Utah, USA), 0.5 mM GlutaMAX (Ingenieur, California, USA), and 2% B27 (Ingenieur, California, USA). However, it is obvious to those skilled in the field of cell culture that other formulations are also possible.

[0051] In the proposed protocol by Maxwell Biosystems, the growth medium can consist of 850 ml of DMEM (Ingenie, California, USA) supplemented with 10% horse serum, 0.5 mM GlutaMAX, and 1 mM sodium pyruvate (Ingenie, California, USA). However, it is obvious to those skilled in the field of cell culture that other formulations are also possible.

[0052] In a possible embodiment, the Maxwell MEA microsensor can operate as stimulation unit 110. The Maxwell MEA microsensor is capable of stimulating BNN activity from an input digital data pattern using a subset of active stimulation electrode sites. The input digital data pattern 105 can then be prepared using various data processing methods and software. In a possible embodiment, the Maxwell stimulation module can be used as SU 110 to provide 32 stimulation channels. Each stimulation channel can provide a voltage amplitude of up to ±1.6V or a current amplitude of ±1.5mA, with an amplitude resolution of 2nA and a time resolution of 2s. The MaxLab Live software component can generate various digital data stimulation patterns suitable for these resolutions, such as monophasic, biphasic, triphasic pulses, ramp waveforms, and other custom pulse shapes.

[0053] In a possible embodiment, the Maxwell MEA microsensor can operate as a readout unit 130. The Maxwell MEA microsensor is capable of outputting digital data readouts of BNN activities, which can be simultaneously recorded using multiple active electrode sites on configurable timescales ranging from microseconds to months. The digital data readouts 135 can then be processed using various signal processing methods. The readouts can also be visualized in an imaging system, for example, as a raster image. While the current Maxwell Biosystems MEA technology, as described in "A 1024-channel CMOS microelectrode array with 26,400 electrodes for in vitro recording and stimulation of electrogenic cells," published in IEEE Solid State Circuits, Vol. 49, No. 11, pp. 2705-2719, 2014, embeds a high-resolution CMOS-based microelectrode array as a two-dimensional plated array with 1024 low-noise readout channels, 26,400 electrodes, a density of 3,265 electrodes per mm², including an on-chip 10-bit ADC, and consumes only 75 mW, other arrangements are also possible.

[0054] While the above possible embodiments have been described using Maxwell Biosystems MEA solutions as exemplary implementations of a core BNN unit with high density and high throughput based on recent technological advancements, it will be apparent to those skilled in the art that other neurotechnologies, electrophysiological and / or optogenetic systems, circuits, devices, probes, components, software, protocols, and methods may also be used individually or in combination to provide a functional core BNN unit 100, for example, those from Harvard Biosciences' Multichannel Systems division. www.multichannelsystems.com ), 3Brain www.3brain.com NeuroNexus, a subsidiary of Nuvectra www.neuronexus.com ), AxionBiosystems https: / / www.axionbiosystems.com ), Charles River Laboratories( www.criver.com ), Plexon www.plexon.com ), by Koniku www.koniku.com Georgia Tech Porter Lab https: / / sites.google.com / site / neurorighter / ), Harvard University Lieber Lab's reticular electronics for chronic recording at the single neuron level ( http: / / cml.harvard.edu Products developed by [various entities].

[0055] It will be apparent to those skilled in the art that different types of neurons can be used as the biological basis of BNN 120. Furthermore, the biological basis can, of course, include a single neuronal cell type, or a predetermined combination of different neuronal cell types or even other cell types. Moreover, the cell type composition can vary throughout the entire 2D or 3D structure of BNN 120. In a possible implementation, rat embryonic neural stem cells (NSCs), such as those from ThermoFisher Scientific and supplied by Ingenium, can be used. Cell lines with catalog numbers N7744-100 and N7744-200. In an alternative embodiment, human neural stem cells (hmNPC), such as those from Lonza Poietics, may be used. TM Neural progenitor cells (NHNP), MilporeSigma VM or Cx, ThermoFisher Scientific StemPro™ neural stem cells, etc. Neuronal cells can be maintained in vitro using biological culture media such as MEM (modified Eagle Medium Gibco) or DMEM (Dulbecco's modified Eagle Medium-Gibco, Invitrogen, ThermoFisher).

[0056] In possible embodiments, the BNN 120 may employ a neurosphere or neural volume system, an adhesion monolayer system, or other specialized arrangements and configurations for immobilizing BNN cells and / or other cells (e.g., neural stem cells (NSCs)). It will be apparent to those skilled in the art of biomaterials that the BNN 120 can be arranged in 2D, or preferably in 3D with various scaffold types, to provide a suitable living environment for the cells. This includes, but is not limited to, processes such as growth into brain organoids or simply maintaining the position of living cells. The resulting in vitro brain organoids can operate sustainably in a manner as similar as possible to the mammalian brain environment. Examples of state-of-the-art scaffold forms and materials, such as hydrogels, for the three-dimensional culture and differentiation of various rat, mouse, and human neural cells, are shown in Table 1, “Three-Dimensional In Vitro Culture Scaffolds for Neural Lineage Cells,” by A. Murphy et al., 54(2017)1-20. Hybrid hydrogels, such as those recently described by Mauri et al., can also be used. BiomaterSci.February 27, 2018; 6(3):501-510, "Functional Evaluation of Hybrid Hydrogels as Three-Dimensional Neural Stem Cell Culture Systems" describes those described therein. Commercially available hydrogel scaffolds from Coming, Lonza, Qgel, Ibidi, and other companies can also be used. In a further embodiment, the scaffold itself can also be constructed from biomaterials, which allows it to grow during neural stem cell replication and differentiation.

[0057] Biocompatible materials are also particularly suitable for microelectronic components of the stimulation unit SU 110 and / or readout unit RU 130. As a possible embodiment, Figure 2a This illustrates a possible application of mask etching to create a biocompatible layer 251 that closely matches the underlying multi-electrode array structure 250. In practice, readout or stimulating electrodes simultaneously read / inject signals into multiple neurons without precise positioning. Etching using a mask 252 opposite the MEA array structure 250 can be used to create a biocompatible layer 251 that induces more precise neuronal positioning by aligning the biocompatible material 251 with the underlying MEA 250, allowing the stimulating unit 110 to stimulate the BNN 120 at the neuron level and / or the readout unit 130 to read the BNN 120 at the neuron level.

[0058] In a possible embodiment, an adhesion layer may be specifically coated at the location of the RU sensor and / or SU probe, such that nerve cells preferably adhere to and / or grow onto these areas, thereby facilitating their control through the RU and SU interface. As a possible alternative embodiment, Figure 2b The diagram illustrates a schematic of a BNN core unit assembly support 270 suitable for ensuring maximum efficiency communication between the RU and SU interfaces and neural cells. In conventional methods, the RU and SU interface sites 271 can be coated with an adhesion layer 272, and an additional layer 274 with masking properties can be deposited further outside the RU and SU interface sites to prevent the units 273 from adhering to and unfolding outside the RU and SU interface control. The interface layer can also consist of special films that allow for optimal adhesion and exchange of molecules and atoms for various purposes, including, for example... A matrix or any similar material that simulates growth and connection.

[0059] Figure 2c The image shows a side view of a possible embodiment of a 3D layered stack of neurospheres 280. Nutrient solutions and additive liquid solutions suitable for the growth and maintenance of BNN cultures flow optimally through the entire neurosphere stack joint. Flow can be promoted and controlled in various ways, such as by natural gravity, centrifugation, electric or magnetic forces, and / or pumping.

[0060] Figure 2d)Shows a possible embodiment of an optogenetic stimulation unit SU interface for generating spikes on transgenic photosensitive neurons 293. A number of light beams 290, 291, and 292 can be aimed at the same neuron in such a way that all the light beams cross at the location of the nerve cell 293, which in turn generates spikes. This method is capable of generating spikes in a three-dimensional aggregate of nerve cells. Let T be the intensity threshold of the light stimulation, above which the illuminated neuron emits a spike signal. Then the intensity values I of the p light beams can be defined p , such that sum(I p ) > T, even though the individual intensity of each light beam is lower than the spike threshold I p < T. It is obvious to those skilled in the art of optogenetics that various illumination devices can be used according to the characteristics of transgenic photosensitive nerve cells. In a possible embodiment, a laser for optogenetics can be used as the illumination beam. To excite a specific neuron at a given depth within a 3D aggregate of neurons, the laser (or multi-spectral) beam can be focused so as to precisely reach the maximum energy of light at the neuron depth position. The target position can also be adjusted by changing the position of the light-emitting device and / or by adjusting the orientation of the mirror used to deflect the direction of the light beam.

[0061] As a possible alternative embodiment, Figure 2e ) shows an alternative embodiment in which the neuron 260 is deposited onto a digital display support 261. The digital display support can be, for example, an LCD screen or an OLED screen. Current display technologies, such as the screens developed for the smartphone industry, typically have pixel sizes on the order of tens of micrometers. In addition, other technologies such as LCoS (liquid crystal on silicon), DLP, or DMD micromirror methods are less than 5 μm, while neuron sizes range from 4 to 100 μm, so it is possible to selectively control pixel illumination outside the neuron 262, at the boundary 263 of the neuron, or in the line 264 with the neuron. In addition, color pixels can be selected to obtain a specific effect on neuron activity according to the characteristics of transgenic photosensitive nerve cells.

[0062] Automated BNN fabrication, growth, and maintenance

[0063] To develop high levels of cognitive abilities, the core units of the BNN need to be maintained in a stable, sustainable, and safe environment for a sufficient period of time to optimize the growth and training of their internal biological networks according to their biological needs, just as in vivo. It is readily apparent to those skilled in the biotechnology field that various automated systems have been developed for efficient biomanufacturing using well-mastered cell lines in many biotechnology industrial applications; however, the in vitro processing of mammalian neural cells outside their natural brain environment is particularly vulnerable, as the natural brain environment is exceptionally well protected from potential external disturbances such as mechanical strain or shock, contamination from pathogens or chemicals, exposure to light radiation, etc. Another issue is that conventional cell line cultures typically operate at the cellular level (e.g., cells are the basic factory components for producing certain proteins of interest), while neural cells need to operate at a higher dimensional level, often through interconnections in multidimensional networks. Yet another issue is that the human brain requires years to express its high levels of cognitive potential in vivo, and there is currently no research evidence that this process can be accelerated in vitro (Pascal, The Rise of the Three-Dimensional Human Brain Culture, *Nature*, Vol. 553, January 2018).

[0064] Therefore, there is a need for specialized automated methods and systems for biological neural networks to manufacture, maintain, and control neural cell cultures in a sustainable and repeatable process that can be automated and run in parallel. The manufacturing process includes the assembly of BNN core unit components and the growth of BNN cultures into functional BNN core units, including automated learning processes. Maintenance and control processes include various activities such as feeding, stimulation, measurement, and curing and cleaning. Possible embodiments of such automated methods and systems will now be described in more detail.

[0065] To expand the development of BNN core units beyond current manual laboratory setups, the manufacturing process can include the automated assembly of BNN core unit components. In a possible embodiment, neural cells can be 3D bioprinted together with a scaffold and the necessary elements for their growth. Examples of 3D biocompatible tissue engineering are products developed by Revotek Ltd., Biosynsphere, Organovo, Aspect Biosystems, and others. In one possible embodiment, BNN core unit components can be derived from stem cell-derived neural progenitor cells, such as those described by Joung et al. in Adv. Funct. Mater. 2018, "3D Printed Stem Cell-Derived Neural Progenitor Cells for Spinal Cord Scaffold Generation." This facilitates optimal cell localization directly on the scaffold. Cell clusters in bio-inks, such as cell-laden hydrogels, can be deposited in continuous channel layers at a resolution of approximately 200 μm. As reported by the authors in their studies of tissue models and future spinal cord injury implants, this approach facilitates axonal proliferation and maintains cell viability and mechanical stability during in vitro central nervous system tissue construction.

[0066] To automate the maintenance of pre-assembled BNN core units in operation beyond current manual laboratory maintenance tasks, the second stage of the automation process could include automated vascularization via microfluidic loops to biologically supply BNN core unit cells and collect their biowaste. To date, only limited solutions have been proposed for the in vitro vascularization of 3D brain organoids. Therefore, due to central necrosis of cells, peripheral nutrients can no longer reach the central necrosis after the organoids reach a certain depth and density, thus limiting their size. Consequently, research biologists have recently proposed transplanting them into the adult mouse brain (Mansour et al., "In vivo models of functional and vascularized human brain organoids," *Nature Biotechnology*, Vol. 36, No. 5, May 2018; and Lancaster, "Vascularization of brain organoids," *Nature Biotechnology*, Vol. 36, No. 5, May 2018). The feasibility of vascularizing brain organs using the patient's own endothelial cells was recently demonstrated in Pham et al.'s paper, "Generation of Human Vascularized Brain Organs" (Neurology Reports, Vol. 29, No. 7, pp. 588-593, May 2018). However, this latter approach is inherently limited to natural jet distribution and can only be controlled externally in a black-box manner, which limits automation capabilities to very simple models. To overcome this limitation of existing BNN cultures, various possible alternative implementations, individually or in combination, can be considered, as will now be described in further detail.

[0067] Figure 3A schematic diagram of a first possible embodiment of an automated vascularization system (AVS) 300 operating together with a BNN core unit 100 is provided, including:

[0068] One or more nutrient tanks 319 are connected to one or more nutrient dispensers 320 to inject one or more nutrients into the BNN culture 120. The nutrient dispenser 320 may include a mixer-syringe active module, which is responsible for delivering the correct dose of nutrients using mechanical means 351, such as valves, syringes, or pumps, according to the specific nutritional needs of the BNN culture.

[0069] One or more additive containers 321, 323, each connected to one or more additive dispensers 322, 324, for injecting one or more additives into BNN culture 120. Additive dispensers 322, 324 may include a mixer-syringe active module responsible for delivering the correct dose of additive according to the growth requirements of the BNN culture using mechanical devices 352, 353 such as valves, syringes, or pumps.

[0070] One or more nutrient waste collectors 325 have mechanical devices 331, such as valves, syringes or pumps, for filtering nutrient waste and discharging it into tanks 335 and / or back into the BNN.

[0071] One or more additive waste collectors 326, 327 have mechanical means 332, 333 for filtering and discharging additive waste into tanks 336, 337 and / or back into the BNN, such as valves, syringes or pumps.

[0072] One or more angiogenesis networks connect nutrient and additive dispensers 320, 322, 324 and nutrient and additive waste collectors 325, 326, 327 to the BNN culture 120. Different angiogenesis streams can also be re-injected into the system to minimize loss.

[0073] Nutrients can include amino acids, carbohydrates, vitamins, and minerals, which can be the same or in separate liquid solutions.

[0074] Additives may include chemicals, pharmaceuticals, or other elements, such as dopaminergic stimuli enhancers that increase the dopaminergic response in the brain nerve impulse response (BNN) and dopaminergic stimuli inhibitors that decrease it. It will be apparent to those skilled in the art of automation that this enables the reproduction and control of the dopaminergic / anti-dopaminergic system. More generally, additives may include chemicals known to affect neurotransmitters in the central nervous system, such as botulinum toxin, nicotine, curry, amphetamines, cocaine, MDMA, strychnine, THC, caffeine, benzodiazepines, barbiturates, alcohol, opioids, etc. Additives may also include growth factors, hormones, and gases (such as carbon dioxide, oxygen, etc.).

[0075] Each nutrient dispenser 320 is interconnected with the BNN culture 120 via a nutrient vascularization network 328, which delivers nutrients to the BNN cells. Each additive dispenser 322, 324 is interconnected with the BNN culture 120 via an additive vascularization network 329, 330, which delivers additives to the BNN cells. The nutrient vascularization network 328 and the additive vascularization networks 329, 330 can be the same or different networks.

[0076] Each nutrient waste collector is interconnected with BNN cells 120 via a nutrient waste vascularization network that delivers nutrient waste from the BNN cells. Each additive waste collector is interconnected with BNN cultures 120 via an additive waste vascularization network that delivers additive waste from the BNN cells.

[0077] The nutrient waste vascularization network can be the same as or different from the nutrient vascularization network 328. The additional waste vascularization network can be the same as or different from the additional vascularization networks 329 and 330. The vascularization network can be manufactured using 3D bioprinting with biocompatible materials or grown from stem cells on BNN 120 culture supports.

[0078] It should be noted that vascularization can be considered as a similar system in the human brain or a completely different type of system, such as the structure of porous materials. Figure 4A side cross-sectional view of an exemplary embodiment of a sponge-like structure serving as a possible host for nerve cells 440, the nerve cells 440 having an inherent vascularized network to support their growth within a 3D BNN 120, is shown. The sponge-like material 439 is preferably soft and compressible to mechanically accommodate the BNN cells 440 (as shown in 410) within its pores 442, while subsequently adapting to their growth and development as a 3D BNN culture. The sponge-like material 439 is further preferably porous and absorbent to deliver nutrients and / or additives and / or waste throughout the BNN culture. In a possible embodiment, a solution containing nerve cells 440 may be immersed in the sponge-like structure 439, such that the nerve cells are regularly distributed within its pores and channels. In a possible embodiment, a polyethyleneimine (PEI) polymer may be used to enhance adhesion factors. In another possible embodiment, neural stem cells 440 may be immersed in a solution containing growth factors (e.g., fibroblast growth factor FGF2) to prevent cell specialization until the stem cells are dispensed into the sponge-like structure 439. Once the assigned cells have adhered to the sponge-like structure 439, the BNN culture 120 is assembled, and the sponge-like structure 439 can be further used as its angiogenesis system. The adhesion and growth factor solution can be washed out and replaced with an additional nutrient and additive solution 441 suitable for the specialization, growth, and maintenance of BNN cultured cells. In an alternative embodiment (not shown), induced pluripotent stem cells can be mechanically inserted and directly cultured onto the sponge-like structure 339 as its inherent vascularization network system.

[0079] Figure 5 A top cross-sectional view of another exemplary embodiment of a spongy structure serving as a possible host for nerve cells 544 is shown. The nerve cells 544 have an inherent vascularized network supporting their growth in a 3D BNN culture, while facilitating the interface connection of electrical and / or optical components of the stimulation and / or readout units with the 3D BNN culture. For this purpose, the spongy structure can be shaped as a sphere, with electrophysiological clamps or electrodes penetrating the spongy structure from various locations distributed throughout the sphere and penetrating it at different depths.

[0080] In another possible exemplary embodiment (not shown), the 3D BNN can be cultured in a sandwich-like structure along biocompatible filaments suspended between two planes. Nerve cells can attach to the biocompatible filaments in various ways. In a possible embodiment, the filaments can be coated with an adhesion factor such as PEI, but other embodiments are also possible.

[0081] BNN controller

[0082] Figure 6A schematic diagram of a possible automated BNN growth, maintenance, and control system is shown, comprising a BNN core unit 120, a stimulation module 110, a readout module 130, and an automated angiogenesis system (AVS) 300, which may include nutrient and additive containers and dispensers, as well as a waste collector. In a preferred embodiment, the operation of the AVS is coordinated by a BNN automation controller 600, which is responsible for real-time calculation of the selection and amount of nutrients and additives to be injected into the BNN culture by the AVS, and the amount of waste to be collected from the BNN culture by the AVS. The BNN automation controller may be a computer processor built with electronic hardware and adapted to execute software algorithms. In a simple embodiment, the BNN automation controller may operate in an open loop, identifying and quantifying necessary nutrients and additives, and deriving the resulting waste values ​​based on scientific expertise from neurophysiological parameterization. In another embodiment, the BNN automation controller may further continuously monitor readout information 635 from the BNN readout module 130 to determine the latest BNN culture status and adjust AVS parameters accordingly.

[0083] In another embodiment, the BNN automation controller can control the stimulation signal 605 to synchronize the stimulation unit with the BNN state. For example, the nutrients and additives injected into the BNN culture medium by the AVS, as well as the waste collected, vary depending on the maturation stage of the BNN cell life cycle: differentiation stage (assembly), growth stage (learning), operational stage (stabilizing function), and death stage (generating additional waste). The proportion of re-injected or treated nutrients or drugs can be controlled using nutrient dispensers, additive dispensers, and their respective waste collectors. Through these feedback loops, the amounts of nutrients and drugs can be dynamically adjusted to: 1 / optimize lifespan, 2 / minimize manual care, 3 / adjust responsiveness, and stabilize BNN performance. Therefore, the homeostasis of the BNN can be maintained over time.

[0084] In a possible embodiment, the BNN automation controller 600 can directly feed the raw data input signal 105 as the stimulus signal 605. In another embodiment, the BNN automation controller may further include a preprocessing unit 610 responsible for converting the data input signal 105 into the stimulus signal 605. This allows the BNN automation controller to better adapt the raw input to the actual stimulus unit SU capabilities and BNN capabilities, thereby optimally performing end-to-end BNN processing tasks without requiring the end user to specifically adapt his / her input signal to every possible configuration. This also facilitates learning tasks, as the capabilities of the BNN can evolve over time, and therefore the signal preprocessing 610 can adapt accordingly.

[0085] In a possible embodiment, the BNN automation controller can directly output the raw BNN readout signal 635 as the signal processed by the end-to-end BNN system. In another embodiment, the BNN automation controller 600 may further include a post-processing unit 630 responsible for converting the raw BNN readout signal 635 into an output signal 135. This allows the BNN automation controller to better adapt the raw readouts to the needs of practical applications, where these readouts may be too noisy to be easily interpreted. One exemplary application is to extract the relevant signal from a series of spikes using a spike sorting signal processing algorithm, but other methods, such as statistical models, classifiers, or even machine learning methods, can also be used. This also facilitates learning tasks, as the end-to-end BNN system can be trained to correlate the synthesized target output signal 135 with a predefined input signal 105 without requiring the end user to interpret the measured output signal specifically based on each possible internal BNN culture or readout unit configuration. This also facilitates learning tasks, as the capabilities of the BNN can evolve over time, and therefore the signal post-processing 630 can adapt accordingly.

[0086] Figure 7Another schematic diagram illustrates a possible automated BNN growth, maintenance, and control as a biological operating system (BOS, analogous to a computer operating system). Similar to a traditional computer operating system or cloud computing service that abstracts the underlying hardware-specific devices and operations, this BOS can host real-time functional control software for end-users to execute BNN processing unit functions through abstracted BNN automation processes, thus remaining unnoticed by the end-user. The proposed BOS operates in conjunction with the BNN core unit culture 120, stimulation and readout modules SU / RU 110 / 130, and a BNN health control unit 710, as previously described, for an automated angiogenesis system (AVS). The BNN health control unit 710 may include a chemical control unit with nutrient and additive tanks and dispensers, as well as a waste collector, and an environmental control unit responsible for controlling other BNN culture environment parameters (e.g., temperature, pressure, humidity, oxygen or carbon dioxide ratio) and other parameters. It may include, for example, digitally controlled micropumps for delivering chemicals and temperature, humidity, pH, or carbon dioxide sensors, or it may even alter any other environmental parameters, including, for example, sound or light waves in any frequency band. The BNN health control unit is responsible for monitoring the health of BNN cells in real time under the supervision of the BNN health control software 700. Therefore, this health monitoring system operates as a steady-state system responsible for adjusting BNN performance over time, as BNN performance can drift naturally and transparently over time to the end user. It operates by adjusting environmental parameters (chemicals, nutrients, temperature, carbon dioxide, etc.) to ensure the normal functioning of the BNN processing unit. This may include periodically checking whether the SU input signal training set is still producing the desired RU output signal. The BNN health control software 700 can be managed by the system's user administrator through the administrator interface 740.

[0087] The real-time function control software 701 is also responsible for managing the functions of the BNN as a processing unit through the BNN function interface 730. The BNN interface 730 can be used to process the inputs and outputs of the BNN in real time, and the inputs and outputs of the BNN are interfaced using the previously described SU / RU system. The BNN interface 730 can, for example, apply any necessary stimulus signal preprocessing tasks (610) and / or readout signal postprocessing tasks (630).

[0088] End users can interact with real-time control software 701 through user interface 741. User interface 741 may be the same or different for end users and / or administrator users, but if they are the same, the administrator user may have access to more functions. Parameters used for real-time function control software 701 and for BNN health control software and / or BNN function interfaces can be stored in database 720.

[0089] For advanced cognitive BNN processing unit functions that may require very low latency and high data throughput bandwidth, the BNN function interface 730 can apply additional data preprocessing 610 and / or data postprocessing algorithms 630 under the control of the real-time function control software 701. In a possible embodiment, the real-time control software operates locally instead of on the Internet. Therefore, the function control software 701 can be uploaded by an administrator using the administrator interface 740, or directly by an end user using the user interface 741.

[0090] The BNN controller computer system (also referred to herein as the "system" or automated system) can be programmed or otherwise configured to implement different BNN processing methods, such as receiving and / or combining stimulus input signals, processing such stimulus input signals, and generating and / or combining readout output signals according to a given application.

[0091] The BNN controller may be a computer system or part of a computer system that includes a central processing unit (CPU, here meaning "processor" or "computer processor"), memory such as RAM, and storage units such as hard disks, and a communication interface for communicating with other computer systems via a communication network (e.g., the Internet or a local area network). Examples of computing systems, environments, and / or configurations include, but are not limited to, personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputer systems, mainframe computer systems, etc. In some embodiments, the computer system may include one or more computer servers that operate alongside many other general-purpose or special-purpose computing systems and may enable distributed computing, such as cloud computing, for example, in a BNN data field. In some embodiments, the BNN controller may be integrated into a massively parallel system.

[0092] BNN controller systems can be applied to the general context of computer-executable instructions (e.g., program modules) executed by a computer system. Typically, program modules can include routines, programs, objects, components, logic, data structures, etc., that perform specific tasks or implement specific abstract data types. As is well known to those skilled in the art of computer programming, program modules can utilize native operating system and / or file system functionalities; standalone applications; browser or application plugins, applets, etc.; commercial or open-source libraries and / or library tools that can be programmed in Python, Biopython, C / C++, or other programming languages; custom scripts, such as Perl or Bioperl scripts; and highly specialized languages ​​suitable for linear genetic programming and cognitive computing, such as SlashA or machine code, or any other data structures describing computational steps (such as those used for advancing genetic programming, Cartesian genetic programming, or tree-based genetic programming).

[0093] Instructions can be executed in a distributed cloud computing environment where tasks are performed by remote processing devices linked via a communication network. In this environment, program modules can reside on local and remote computer system storage media, including memory storage devices.

[0094] Therefore, the entire system can operate as a biological operating system 750, as it incorporates all the systems required to ensure the operation of the BNN computing system.

[0095] In possible embodiments, the BNN functional interface 730 may include various input and / or output signal processing algorithms, such as preprocessing or post-processing filters, classifiers, machine learning algorithms based on mathematical or statistical models, and the functional control software 701 may control these algorithms and their parameters as needed by the end user. Signal preprocessing may include transforming the signals to be learned by the BNN so that they are optimally suited to the SU stimulus capability and format. Signal postprocessing may include converting the signals output from the BNN into a more comprehensive format so that they provide the target function. In possible embodiments, the BNN may be modeled as a nonlinear system, and signal preprocessing may include applying a nonlinear gain to the input signal. In possible embodiments, the BNN functional interface may include one or more artificial neural networks as a preprocessor and / or postprocessor, and parameters may include weight values, the selection of activation functions, and other parameters for learning and / or classifying the signal. For example, the preprocessor may be used to repeat the input applied signal over a period of time or slightly modify these input applied signals for re-enhancement of a more robust training and learning process.

[0096] Typically, automation aims to produce a BNN processing system that generates the correct set of spatial-temporal output signals Oi(t) for a given spatiotemporal input signal Sj(t) (where i represents the stimulus input electrode, j represents the sensing output electrode, and t represents the sampling time). Typically, a BNN can be trained using a set of k distinct functions {O(t), S(t)}(k), collectively called the training set, where O(t) and S(t) represent vectors of components Oi(t) and Si(t), respectively. After successful training, the BNN can be used to predict the correct output for inputs S(t) that are not part of the training set. Successful training and successful prediction can be measured as minimizing the difference between the target signal O(t) and the measured BNN output signal. Various metrics can be used for this, such as, for example, mean squared error (MSE) or more generally, n-norm distance, such as Euclidean distance or 1-norm distance. Formally, this means being able to output the correct values ​​of a set of p distinct functions {O(t), S(t)}(p), which together are referred to as the test set, which is not part of the training set.

[0097] In some embodiments, to achieve such successful training, machine learning algorithms (Artificial Neural Networks (ANNs), Convolutional Neural Networks (CNNs), Support Vector Machines (SVMs), and deep learning, but this also includes any ML methods such as random forests, genetic algorithms, genetic programming, reservoir computation, etc.) can be used to optimally represent the input and optimally measure the output. For example:

[0098] Some of the k functions {O(t), S(t)}(k) may need to be repeated more frequently than the other functions;

[0099] - Some inputs can be mapped to different electrodes over time;

[0100] - It may be necessary to read several electrodes simultaneously;

[0101] -etc.

[0102] More generally, depending on application requirements, the raw signal input 135 can be fed to the first preprocessing subunit 610 of the BNN interface 730, which processes the raw signal input 135 before feeding it to the BNN 120 with SU 110. The BNN 120 output signal 635 read from RU 130 can itself be further transformed by the second post-processing subunit 630 of the BNN interface 730 to produce a more suitable output signal 135. In an open-loop architecture, the preprocessing and post-processing subunits can operate independently of each other. In a closed-loop architecture, the preprocessing and post-processing subunits can operate jointly; for example, some output from the post-processing unit can be fed back to the stimulus preprocessing unit.

[0103] Preferably, the preprocessing unit can perform space-time processing on the original input Sr(t) to generate a processed signal S(t) that will be fed into a BNN with SU. Then, the BNN outputs the original signal O. r (t), the original signal O r The signal (t) is then processed by the post-processing unit to produce a processed output signal O(t). Each of these signals can have a different dimension. For example, to make a BNN output a spike for a simple, raw one-dimensional sine wave at a frequency of 1 Hz, it might be necessary to excite 100 different electrodes one after another, each with a different time signal defined at a resolution of 100 Hz. The pre-processing unit 610 can be adjusted accordingly. The pre-processing unit 610 and the post-processing unit 630 can also utilize machine learning. In a later embodiment, the three subsystems can be trained to achieve the objective of minimizing the overall difference between the target signal of the predefined input signal 105 and the measured final output signal 135. Note that as BNN technology develops, it is also possible to implement the pre-processing unit 610 and / or the post-processing unit 630 in software rather than hardware, for example, as a simple BNN, to gradually build more complex systems suitable for learning higher cognitive functions.

[0104] Therefore, the BOS 750 integrates health monitoring and functional control of BNN culture 120 to provide end users with real-time, sustainable, and reliable BNN computational operations (including learning) by enabling the following functions:

[0105] The steady-state function ensures a constant performance level by optimizing the operation of the BNN function interface 730 and ECCU parameters 710.

[0106] The learning function works by optimizing the BNN interface (potentially including input-side signal preprocessing and / or output-side signal postprocessing algorithms) along with the ECCU parameters required to learn new input / output associations. Specifically, the function control software 701 is capable of real-time, coordinated with the closed-loop stimulation and readout signals processed by the BNN function interface, adjusting the values ​​of the additive supply and / or environmental parameters used for BNN health control to facilitate optimal learning with respect to the training set.

[0107] Maintaining functionality may require cloning aged BNN cultures so that these BNN cultures can operate directly on training sets already learned by pre-existing BNNs. Then, at least a portion of the BNN interface and the ECCU parameters of pre-existing BNNs can be retrieved from database 720 and copied to accelerate learning time on new BNNs.

[0108] BNN Automated Learning

[0109] The inherent dynamic interconnectivity of neurons is considered a means of creating (re)programmable functions based on data feedback, an essential component of machine learning. It is readily apparent to those skilled in neuroscience that a BNN system can be trained initially with input data signal 105 until it reaches a steady state that produces the desired output data signal 135, and the same applies to inputs not fed during training. When the BNN can correlate the target output data signal with the input data signal as desired, the system can be said to have learned and generalized, because it can produce a deterministic response when fed a given input.

[0110] To train the BNN, the automated controller can adapt the input spatiotemporal signal 605 of the stimulation unit SU to the input data signal 105 and the output spatiotemporal data signal 135 until the output spatiotemporal data signal 135 matches the target output data signal. The automated controller can further adjust at least one or more of the following:

[0111] -BNN core unit environment parameters;

[0112] -BNN neural cell culture nutrition supply;

[0113] -BNN neural cell culture additive supply;

[0114] -BNN neural cell culture nutrient waste collection;

[0115] -BNN neural cell culture additive waste collection;

[0116] -BNN interface signal preprocessing algorithm parameters;

[0117] -BNN interface signal post-processing algorithm parameters;

[0118] This continues until the output spatiotemporal data signal matches the desired output data signal.

[0119] Figure 8 A learning process that can be implemented by real-time processing software is illustrated, wherein a feedback loop H 800 adapts a stimulus signal 605 to a measured readout signal 635. In a possible embodiment, the real-time processing software can trigger periodic repetition of the stimulus signal 605 on the BNN to promote long-term potentiation. More generally, environmental and chemical parameters controlled by a health monitoring system can also be part of a closed-loop learning system. In a possible embodiment, when the measured readout signal 635 coincides with the expected signal of the training group for a given stimulus signal input, the real-time processing software can trigger the delivery of a known drug dose to enhance the BNN in return.

[0120] In other possible embodiments (not shown), BNN can connect to external information sources, such as internet web databases, via the BNN automation controller interface. BNN can further learn to access and use this external information as a source to learn more, connect new concepts, and further develop higher levels of cognitive abilities over time.

[0121] BNN Biological Computing Stack

[0122] For the feedback loop to function, closed-loop control of the BNN core unit is necessary. The proposed automated vascularization system is limited to the growth and maintenance of BNN cultures that are relatively small compared to the mammalian brain, for example, on the order of 10,000 interconnected neurons. Therefore, simply integrating a single BNN core unit into a BNN computational system to learn and manipulate higher cognitive functions may not be sufficient. This limitation can be overcome using a BNN biological computing stack (BCS), such as... Figure 9a )and Figure 9b As shown in the diagram. In a possible embodiment, one or more BNN core units can be arranged serially. The readout unit RU of the first BNN core unit can be connected to the stimulation unit SU of the next BNN core unit, or alternatively, they can be combined in a single piece of hardware. At each interface between the blocks, the system can also accept, as shown in the diagram. Figure 9b The external stimulus signal input is shown as 900 (ES).

[0123] As a possible embodiment, 2D BNN core units can be stacked vertically. In another possible embodiment, BNN core units fabricated as stacks of neurospheres or layers of 3D bioprinted material can be mechanically arranged in series, with electrophysiological probes inserted at the interface between two adjacent units.

[0124] like Figure 9b As shown in the diagram, in an arrangement similar to a deep learning hierarchical architecture, the learning process can be controlled end-to-end via a feedback loop operating between the final readout unit and the first stimulus unit. In this configuration, due to the inherent characteristics of biological neural networks, the feedback element acts as a stabilizer, altering its internal state even in the absence of external stimuli.

[0125] Figure 10a )and Figure 10b Two additional embodiments are shown, one with and one without the additional post-processing block O 1010, of an adaptive closed-loop system controlling a BNN computation stack with an additional nonlinear gain P 1000. It will be apparent to those skilled in the art of deep learning that the pre-processing block P facilitates the learning process, and the post-processing block O can represent the topology used for reservoir computation. This configuration supports a trainable biological computation stack (T-BCS).

[0126] Figure 11 It shows the relationship with Figure 10b The specific implementation of the related data is as follows: the O output signal post-processing block 1010 is implemented by an artificial neural network (ANN). The biological neural network (BNN) 120 is connected to the digital interface readout unit (RU) and the stimulation unit (SU) via a multi-electrode array (MEA) 1135. The feedback function (HI) 1137 corresponds to the learning processing of the ANN, such as, but not limited to, backpropagation. The external feedback function (H2) 1136 corresponds to the average value of the long-term potential applied to the BNN. The external feedback function H2 can be implemented using genetic programming or machine learning to determine the correct spatiotemporal sequence at the output of the BNN culture 120 to perform the desired spike. Another way to apply long-term enhancement is to use the RU as the stimulation unit with the desired spike sequence, which will strengthen the internal connections within the BNN. Using this method, the ANN can even be omitted in some embodiments.

[0127] Let S(t) be different stimulus units S that depend on MEA. i The time (t) is a periodic data input function. Let O(t) be the time (t) of different readout units O of the MEA for a given S(t). i For example, the data output function depends on time (t).

[0128] Then, H2 is a function or algorithm (found by example using genetic programming or machine learning) that minimizes (or maximizes) the scalar metric L, such that the BNN always produces the same spike-time function O(t) for a given input S(t) when L is minimized (or maximized) for any number of periods P. L is called the loss (or reward) function.

[0129] For example, L could be:

[0130]

[0131] Here, N is the period of the input function S(t). In this case, the function or algorithm H2 will minimize L.

[0132] Figure 12 It shows Figure 11Another possible implementation of this approach facilitates long-term maintenance of T-BCS operation. In practice, the BNN culture 120 can evolve over time and deviate from its initial (learned) structure, potentially causing drift in the T-BCS data output. This can be monitored by periodically testing the T-BCS operation using a validation set that includes one or more of the initial signal training set as inputs to the SU. The output of the T-BCS readout unit RU can then be checked against the expected prediction set, and if the difference is too large, the T-BCS must reapply the learning process. Since the learning process is time-consuming and disrupts T-BCS functionality, in a possible implementation, the T-BCS system is replaced with a new system that has already been prepared.

[0133] BNN server

[0134] The proposed automated BNN system, BNN operating system (BOS), and trainable BNN computation stack (T-BCS) can provide the core architecture for network-based computing servers. Compared to traditional software and hardware-based server architectures, such BNN servers may be more suitable for providing advanced cognitive processing more efficiently.

[0135] like Figure 13 As shown, the T-BCS wetware architecture can operate together with hosts that provide different services to user clients. Figure 14 It also describes possible functions that can be operated by the host server to manage user clients, such as processing requests, scheduling jobs, reporting and logging, maintaining dashboards, and issuing invoices. Figure 15 It also describes a general architecture with load balancing for serving multiple clients from the same T-BCS server host.

[0136] Preferably, the BNN server supports redundant T-BCS operations, allowing one T-BCS to be removed for maintenance (e.g., relearning) while at least one other T-BCS remains operational to serve client requests.

[0137] Example application of BNN maintenance and updates

[0138] Figure 16 b) illustrates a possible application of the T-BCS server as a reverse image search service compared to a traditional deep learning server 16a). Figure 16 a) In traditional applications, such as those implemented by Google, reverse image search services take an image as input and tell you where the image is located on the web, give a description of the image, and may also return a collection of similar images. Figure 16The existing reverse image search backend of a) uses multiple different image processing or image vision systems 1613 to obtain a robust and compact representation of the image, typically as a hash value 1614. The proposed application replaces the complete backend processing with T-BCS 1615, significantly reducing power consumption with the same computational power. In particular, once properly trained, the Biocomputing Stack (BCS) can advantageously replace the feature extraction portion, including machine learning AI networks, or feature descriptor extractor algorithms, or perceptual hash functions, or additional image processing tasks, or any combination of these tasks, as well as the so-called post-hash processing required to obtain a compact representation of the image to be stored in the database.

[0139] Other experiments, examples, and applications

[0140] Figure 17 This study demonstrates how rat cortical stem cells, after maturing for 4 days in DMEM-F12 medium supplemented with glutamine (TM), fibroblast growth factor, epidermal growth factor, and Stempro (R), form a neurosphere approximately 200 meters wide. In a possible embodiment, 3D electrodes can be distributed on the surface of the neurosphere (or any other type of neuronal aggregate). Following growth stimulation, the tips of the electrodes will gradually fit into the growing neurosphere. Figure 18 , Figure 19 and Figure 20 An example of 12 electrodes arranged in a virtual ring-like pattern along the neurosphere of cortical neural stem cells that grow from 400 μm to over 1 mm in 15 days is schematically shown, allowing the tip of each electrode to naturally become more deeply embedded in the neurosphere. In this particular case, growth is caused by… Matrix-stimulated, for example, matrix extracted from Engelbreth-Holm-Swarm (EHS) mouse sarcoma.

[0141] A matrix can also be used to stimulate the 3D growth of nerve cells directly on the MEA. This is in Figure 21 The diagram shows that adherent cells 2120 grow such that these adherent cells 2120 extend their axons 2121, 2122, or these adherent cells 2120 migrate through Matrix 2100. This makes it possible to obtain much thicker networks that can extend more than a few millimeters above the MEA surface 210. Figure 22 Adherent cells 2201 and Microscopic images of 2202 include nerve cells growing on the surface of the MEA. It will be apparent to those skilled in the art that the same principle applies to neurospheres or any aggregate of nerve cells.

[0142] While various implementation methods have been described above, it should be understood that they are presented by way of example rather than limitation. It will be apparent to those skilled in the art that various changes in form and detail can be made without departing from the spirit and scope. In fact, after reading the above description, those skilled in the art will understand how to implement alternative implementation methods.

[0143] It will be apparent to those skilled in the art of digital data communications that the methods described herein can be applied indiscriminately to various data structures such as data files or data streams. Therefore, the terms “data,” “data structure,” “data field,” “file,” or “stream” can be used interchangeably in this specification.

[0144] Although the above detailed description contains many specific details, these details should not be construed as limiting the scope of the embodiments, but rather as providing illustration of some embodiments among several embodiments.

[0145] While various implementation methods have been described above, it should be understood that they are presented by way of example rather than limitation. It will be apparent to those skilled in the art that various changes in form and detail can be made without departing from the spirit and scope. In fact, after reading the above description, those skilled in the art will understand how to implement alternative implementation methods.

[0146] Furthermore, it should be understood that any accompanying drawings highlighting the features and advantages are presented for illustrative purposes only. The disclosed methods are flexible and configurable enough that they can be used in ways other than those shown.

[0147] Although the term "at least one" is frequently used in the specification, claims, and drawings, the terms "a," "an," "the," "the," etc., also mean "at least one" or "the at least one" in the specification, claims, and drawings.

[0148] Throughout this specification, multiple instances can be implemented as components, operations, or structures described as a single instance. While individual operations of one or more methods are described and presented as separate operations, one or more of these operations may be performed concurrently and do not need to be performed in the order stated. Structures and functionalities presented as separate components in the example configuration can be implemented as combined structures or components. Similarly, structures and functionalities presented as single components can be implemented as separate components. These and other variations, modifications, additions, and improvements fall within the scope of this document.

[0149] This document describes certain embodiments as including logic or components, modules, units, or mechanisms. Modules or units may constitute software modules (e.g., code embodied on a machine-readable medium or embodied in transmitted signals) or hardware modules. A hardware module is a tangible unit capable of performing a particular operation and may be configured or arranged in a particular physical manner. In various example embodiments, one or more computer systems (e.g., standalone computer systems, client computer systems, or server computer systems) or one or more hardware modules of a computer system (e.g., processors or a group of processors) may be configured by software (e.g., an application or an application portion) to operate as hardware modules to perform certain operations described herein.

[0150] In some embodiments, the hardware module may be implemented mechanically, electronically, biologically, or any suitable combination thereof. For example, the hardware module may include dedicated circuitry or logic permanently configured to perform certain operations. For example, the hardware module may be a dedicated processor, such as a field-programmable gate array (FPGA) or an ASIC. The hardware module may also include programmable logic or circuitry temporarily configured by software to perform certain operations. For example, the hardware module may include software contained within a general-purpose processor or other programmable processor. It will be understood that the decision to implement the hardware module mechanically, in a dedicated and permanently configured circuit, or in a temporarily configured (e.g., software-configured) circuit may be driven by cost and time considerations. With the development of synthetic biology, all or part of the hardware module may be made from biological cells, such as neurospheres and / or genetically engineered cells (also known as wet components).

[0151] The various operations of the example methods described herein can be performed, at least in part, by one or more processors (e.g., via software) that are temporarily or permanently configured to perform the relevant operations. Whether temporarily or permanently configured, such processors can constitute processor-implemented modules that operate to perform one or more of the operations or functions described herein. As used herein, a processor-implemented module refers to a hardware module implemented using one or more processors.

[0152] Similarly, the methods described herein can be implemented at least in part by a processor, which is an example of hardware. For example, at least some operations of the methods can be performed by one or more processors or modules implemented by processors.

[0153] Some parts of the topics discussed herein can be presented in terms of algorithms or symbolic representations of operations on data stored as bits or binary digital signals in machine memory (e.g., computer memory). Such algorithms or symbolic representations are examples of techniques used by those of ordinary skill in the field of data processing to convey the essence of their work to others of similar skill in the art. As used herein, an algorithm is a self-consistent sequence of operations or similar processes that leads to a desired result. In this context, algorithms and operations involve the physical manipulation of physical quantities.

[0154] While an overview of the subject matter of the invention has been described with reference to specific exemplary embodiments, various modifications and variations can be made to these embodiments without departing from the broader spirit and scope of the embodiments of the invention. For example, those skilled in the art can mix and match various embodiments or features thereof, or make them optional. These embodiments of the subject matter of the invention may be referred to herein individually or collectively by the term "invention," which is merely for convenience and not intended to limit the scope of this application to any single invention or inventive concept if more than one is actually disclosed.

[0155] It is believed that the embodiments described herein have been described in sufficient detail to enable those skilled in the art to practice the disclosed teachings. Other embodiments may be used and derived therefrom, such that structural and logical substitutions and changes may be made without departing from the scope of the invention. Therefore, the specific embodiments should not be construed as limiting, and the scope of the various embodiments is defined only by the appended claims and the full scope of their equivalents.

[0156] Furthermore, multiple instances may be provided for a resource, operation, or structure described herein as a single instance. Additionally, the boundaries between various resources, operations, modules, engines, and data stores are somewhat arbitrary, and specific operations are illustrated within the context of a particular illustrative configuration. Other functional assignments are predictable and may fall within the scope of various embodiments of the invention. Generally, structures and functions presented as individual resources in the example configuration can be implemented as combined structures or resources. Similarly, structures and functions presented as single resources can be implemented as individual resources. These and other variations, modifications, additions, and improvements fall within the scope of embodiments of the invention as defined by the appended claims. Therefore, the specification and drawings are considered illustrative rather than restrictive.

[0157] Finally, the applicant's intention is to interpret claims containing the explicit language "apparatus for..." or "steps for..." only in accordance with paragraph 6 of 35U.SC112. Claims that do not explicitly include the phrases "apparatus for..." or "steps for..." shall not be interpreted in accordance with paragraph 6 of 35U.SC112.

Claims

1. An automated biological computer system, comprising: (a) A core unit of a biological neural network (BNN), comprising 2D or 3D in vitro neural cell cultures; (b) A stimulation unit configured to apply spatiotemporal stimulation signals to a first group of nerve cells using a multi-electrode array (MEA) or an optogenetic stimulation system; (c) A readout unit configured to acquire spatiotemporal activity signals from a second group of nerve cells using an MEA, an imaging system, or an electrical sensor; (d) An automated vascularization system, comprising: (i) One or more additive tanks, each additive tank being connected to one or more additive dispensers to inject one or more additives into BNN cell cultures; (ii) One or more vascularized networks for connecting the one or more additive dispensers to the BNN core unit; (e) A control unit connected to the stimulation unit, the readout unit, and the automated vascularization system, the control unit being configured to monitor the health status of the BNN cell culture and adjust the supply of additives in real time so that the spatiotemporal input data signal can be continuously converted into a spatiotemporal output data signal; and (f) A user interface configured to enable end users to manage the control unit using real-time function control software.

2. The automated biological computer system according to claim 1 further includes a preprocessing unit configured to convert digital input data signals into spatiotemporal stimulation signals.

3. The automated biological computer system according to claim 1 or 2 further includes a post-processing unit configured to convert the spatiotemporal readout signal into a digital output data signal.

4. The automated biocomputer system according to any one of claims 1 to 3, wherein the one or more additives are selected from: dopamine stimulators for enhancing the dopamine response of the brain neural network (BNN), dopamine stimulators for reducing the dopamine response of the BNN, known agents for enhancing the BNN as a reward when the spatiotemporal output data signal matches the desired output data signal of a given stimulus input, botulinum toxin, nicotine, tubocurarine, amphetamine, cocaine, MDMA, strychnine, THC, caffeine, benzodiazepines. Drugs, barbiturates, alcohol, opioids, growth factors, hormones, gases, or combinations thereof.

5. The automated biological computer system according to any one of claims 1 to 4 further includes a database for storing parameters of the real-time function control software.

6. A server for performing automated processing tasks, the server comprising the automated biological computer system of any one of claims 1 to 5.

7. A method for operating an automated biological computer system according to any one of claims 1 to 5, the method comprising: (a) Real-time function control software is hosted in the control unit of the automated biological computer system, and the real-time function control software is configured by the end user to perform BNN processing functions; (b) Receive user input for performing BNN processing functions via the user interface of the automated biological computer system; (c) Using a multi-electrode array (MEA) or optogenetic stimulation system, a spatiotemporal stimulation signal is applied to the first group of nerve cells through the stimulation unit of the automated biocomputer system; (d) Using MEA, imaging system or electrical sensor, spatiotemporal activity signals are acquired from the second group of nerve cells through the readout unit of the automated biocomputer system; (e) The health status of the BNN cell cultures of the automated biological computer system is monitored by the control unit; as well as (f) The supply of additives to the BNN cell culture is adjusted in real time by the control unit so that the spatiotemporal input data signal can be continuously converted into the spatiotemporal output data signal.

8. The method according to claim 7 further includes converting the digital input data signal into a spatiotemporal stimulation signal of the stimulation unit.

9. The method according to claim 7 or 8 further comprises converting the spatiotemporal activity signal from the readout unit into a digital output data signal.

10. The method according to any one of claims 7 to 9, further comprising storing parameters of the real-time function control software in a database.

11. The method according to any one of claims 7 to 10, further comprising real-time function control software for the control unit uploaded by the end user through the user interface.

Images