The project and its constituent work packages and tasks have now been completed. Some work continues in analysis of the data from final NEU-ChiP prototype in preparation for publication. The project developed methods to grow human stem cell derived neuronal networks and approaches for their electrical and optical interrogation in fabricated devices, together with analysis and inference methods to determine network structure and plasticity changes. A prototype was developed as a proof-of-concept, with continuing analysis and experiments beyond the project life. In addition to these scientific advances, we organised and participated in significant symposia and workshops, namely a joint symposium with a Japanese biological AI consortium at the 2023 NOLTA conference in Sicily, a satellite symposium at the 2024 FENS meeting on “Cortical neurons for biological computing”, which we believe is the first of its kind, and a workshop on “Maturation and Plasticity in Neural Networks” in October 2024 in Cargèse, Corsica. Other dissemination activities included a number of journal publications with four having contributions from multiple NEU-ChiP partners. Six innovations were identified for potential exploitation. These are at different readiness levels, with one having additional local accelerator funding and one being pursued for exploitation by the academic institution. The success of the project also contributed to the generation of new funding from the UK national centre for neuromorphic computing.
Main results:
• Generation of human iPSC-derived neural progenitor cell lines stably expressing a genetically encoded calcium indicator (GCaMP6).
• Fabrication and characterisation of micro-structured devices.
• Use of PDMS micro-structured devices to direct neurite growth between cell populations.
• Capacity to monitoring the development of human neuronal cultures during the processes of differentiation and maturation, using calcium imaging and MEAs, along several weeks and within micro-structured devices.
• Determination of response of neurons upon stimulation on the MEAs and characterization of functional remodelling due to plasticity.
• Investigation of effect of chemical stamping and PDMS mould patterning on neuronal network dynamic behaviour.
• Development of stimulation protocols to imprint tasks in evolving neural networks.
• Devise methods for inferring functional connectivity from neuronal activity data.
• Realizing probabilistic and statistical models of signal transmission by neurotransmitters between a presynaptic neuron and a postsynaptic neuron.
• Demonstration of long-term potentiation in neuronal networks in response to specific spatially defined electrical stimulation using the 3Brain CMOS MEA.
• Successful integration and optimisation of PDMS devices with 3Brain CMOS MEA system, and development of protocols shared across partner laboratories for the formation and long-term culture of neural cell circuits.
• Integration of ‘digital mirror devices’ with PDMS devices and 3Brain CMOS MEAs to provide bespoke patterned optical stimulation.
• Development of a ‘numerical simulations package’ (in silico neuronal networks) to
include aspects inspired by neuroengineering, such as guidance of axons, topographical modulation of substrate where neurons sit, and modular organization. This package has helped to understand experimental results, provide feedback and make predictions, and support theoretical models.
• Investigation of ‘reservoir computing’ in silico and in vitro, observing a good agreement between simulations and experiments in preparations of rat primary cultures.
• Advancement in the theoretical understanding of some empirical observations about the accessibility of computational tasks under stimulation protocols, which have been done in the context of the first algorithm.
• Development a method for visual informatics analysis of neuronal activity patterns, which allows to identify key developmental stages and the impact of stimulation on network behaviour.
• Development of a Bayesian method for inferring effective synaptic strength, neuronal type, as well as connectivity.
• Utilization of the aforementioned methodologies to study the effects of Long-Term Potentiation (LTP) stimulation on neuronal network plasticity.
• Development of a reservoir computing system using volatile memristors as leaky integrators for MNIST digit recognition with different input methods (1D, 2D, and Parity), achieving classification accuracies of 84%, 91%, and 93% respectively.
• Benchmarking of the neuromorphic electronic circuit implementation against biological
performance.
