Vaiva Vasiliauskaite

Longitudinal electrophysiological recordings of neuronal networks are essential for studying network maturation, plasticity, and pharmacological responses. Yet, current microelectrode array (MEA) approaches are limited by evaporation‐induced drift in culture conditions, exacerbated by heat dissipation from active recording electronics on CMOS‐based high‐density MEAs. We present a cell culture lid featuring a water compartment at its interface that eliminates evaporation whilst maintaining gas exchange. Combined with a custom incubator that uses independent temperature control of the MEA to prevent condensation, the system enables stable, uninterrupted recordings for weeks. We show that perturbations in firing rate and functional connectivity following medium exchange are significantly reduced by suppressing evaporation. We demonstrate continuous 35‐day recordings of patterned human iPSC‐derived neuronal networks with a single half‐medium exchange, revealing multiple spatiotemporal firing patterns that are held concurrently and evolve over weeks. All design files are provided, enabling adoption across culturing platforms for longitudinal studies of development within individual cultures, and establishing the conditions required for studies of plasticity and chronic pharmacology.

We investigate the impact of transitive reduction on citation networks. Our hypothesis is that documents which lose fewer citations under transitive reduction are likely to be multi-class (our definition of interdisciplinary), while a large loss of citations suggests a document is primarily cited within a single class. We test this hypothesis by using an artificial model of a citation network and by using data on citations from three sources: academic papers, court decisions and patents. Where needed, we applied modularity-based clustering techniques on a network defined using the undirected version of the citation network to classify documents by communities. A cluster-dependent measure was then used to classify the nodes as multi-class or single-class. Our results provide strong support for our hypothesis in all four cases.

Directed cycles are fundamental motifs in natural, social and artificial networks, yet their distinct roles in processing information remain poorly understood, particularly as higher-order structures. Furthermore, it is unclear how the function of a cycle depends on the wider network in which it is embedded. Here we show, using information-theoretic measures, that network size, sparsity and directionality critically shape how directed cycles process information. In networks with no preferred direction, feedforward cycles enable greater information flow, while feedback cycles support stronger information integration. The relative orientation of a feedforward cycle, and the structural incoherence it induces, tunes its capacity to generate higher-order behaviour. Introducing feedback into otherwise feedforward architectures increases the diversity of network activity while modifying the computational behaviour. In a supervised learning task, feedforward networks train faster and tolerate random node ablations, whereas feedback makes them robust to input noise. These findings reveal directed cycles as tunable computational units whose behaviour is set by their structural context.

Glioblastoma (GBM) is the most aggressive primary brain tumour and is frequently accompanied by severe neurological symptoms, including epilepsy and cognitive impairment. Neurological symptoms often persist after surgical resection, indicating that GBM induces durable and self-sustaining changes in the surrounding neuronal networks. However, the mechanisms by which GBM reshapes network structure and function in the tumour periphery remain poorly understood. We present a compartmentalised in vitro platform enabling long-term coculture of iPSC-derived neurons and primary GBM cells to investigate these changes. Placed on high-density microelectrode arrays, the platform permits longitudinal electrophysiological recordings at single-neuron resolution. Using effective network inference, we find that GBM drives a reproducible structural progression: first toward a hyperconnected, hub-dominated architecture, then a collapse of community structure accompanied by a widespread neuron loss. This evolving structure shapes population dynamics, constraining features such as network burst rate and instantaneous synchrony. The reorganisation also carries computational consequences: signal propagation becomes progressively redundant and synergistic rather than unique. As a result, neurons lose the capacity to encode distinct input combinations independently, and the repertoire of accessible network states contracts. Together, these findings reframe GBM as a driver of neuronal network reorganisation rather than uniform hyperexcitability, and establish a compartmentalised, single-neuron-resolution platform for the longitudinal observation, dissection, and ultimately targeting of the network processes that underlie disease progression.

Understanding how neurodegenerative diseases initiate and propagate through neural circuits remains a fundamental challenge in neuroscience. The earliest stages occur years before symptoms emerge, making them inaccessible to study in patients. Microfluidic platforms, where neurons communicate across chambers through microchannels accessible only to their axons, have opened new experimental avenues. However, existing models lack the complexity and precision needed to track how individual circuit components respond to focal pathological changes over time. Here we present a 33-chamber cortical network-on-chip integrating human iPSC-derived excitatory neurons, inhibitory neurons, and astrocytes in a six-layer feedforward architecture recapitulating the laminar structure of the neocortex. Amyloid-β is applied globally, while progerin-induced accelerated ageing in a single chamber establishes a defined disease core. Continuous recordings using high-density microelectrode arrays reveal progressive, layer-dependent changes in firing dynamics and network topology. Machine-learning-based feature analysis identifies a multiparametric electrophysiological signature distinguishing healthy from disease-affected chambers, enabling studies of the earliest timepoint at which pathology becomes detectable. This establishes a scalable framework for mechanistic studies of neurodegeneration and identification of electrophysiological biomarkers of disease progression.

Although data generation is often straightforward, extracting information from data is more difficult. Object-centric representation learning can extract information from images in an unsupervised manner. It does so by segmenting an image into its subcomponents: the objects. Each object is then represented in a low-dimensional latent space that can be used for downstream processing. Object-centric representation learning is dominated by autoencoder architectures (AEs). Here, we present ORGAN, a novel approach for object-centric representation learning, which is based on cycle-consistent Generative Adversarial Networks instead. We show that it performs similarly to other state-of-the-art approaches on synthetic datasets, while at the same time being the only approach tested here capable of handling more challenging real-world datasets with many objects and low visual contrast. Complementing these results, ORGAN creates expressive latent space representations that allow for object manipulation. Finally, we show that ORGAN scales well both with respect to the number of objects and the size of the images, giving it a unique edge over current state-of-the-art approaches.

Understanding how neuronal circuits transform inputs into outputs requires systematic perturbation under controlled conditions. In vitro neuronal networks can be cultured on microelectrode arrays (MEAs) that allow stimulation and recording, and microfluidic patterning can constrain network topology to yield stable stimulation-evoked responses. Yet, the space of possible spatiotemporal stimulation patterns remains too large for exhaustive exploration. Additionally, the evoked responses depend on prior stimulation history. Here, we embedded topologically constrained biological neuronal networks in a closed-loop reinforcement learning (RL) framework that sends electrical stimuli to the MEA and evaluates the evoked responses to efficiently identify stimulation patterns that evoke specific target activity motifs. We extend inkube, a low-cost, open-source electrophysiology system from off-the-shelf components, with closed-loop electrophysiology. This enables reliable delivery of stimulation at single-sample precision with millisecond-range round-trip times. It also allows independent RL agents to control multiple networks simultaneously. We first demonstrated that stimulation-evoked responses in engineered recurrent networks were stable and separable across the action space over hours of continuous operation. We characterised the dependency of responses on prior stimulation history, finding state dependence in a subset of stimulus pairs. We then benchmarked different RL agents, multi-armed bandits (MABs) and linear contextual bandits (LCBs), on the task of identifying stimulation patterns that maximise the length of clockwise-circular firing sequences. All agents learned to improve reward during training with respect to random stimulation. Agents converged on non-trivial stimulation patterns that span the full action space rather than mirroring the target motif. LCBs exploited the identified state dependence through action switching, yielding measurable reward benefits for specific action pairs, though this did not translate into overall performance gains over MABs. All hardware designs and software are publicly available, providing an accessible platform for goal-directed functional characterisation of engineered neuronal networks at single-spike resolution.

Longitudinal electrophysiological recordings of neuronal networks are essential for studying network maturation, plasticity, and pharmacological responses. Yet current microelectrode array (MEA) approaches are limited by evaporation-induced drift in culture conditions, exacerbated by heat dissipation from active recording electronics on CMOS-based high-density MEAs. We present a cell culture lid featuring a water compartment at its interface that eliminates evaporation whilst maintaining gas exchange. Combined with a custom incubator that uses independent temperature control of the MEA to prevent condensation, the system enables stable, un-interrupted recordings for weeks. We show that perturbations in firing rate and functional connectivity following medium exchange are significantly reduced by suppressing evaporation. We demonstrate continuous 35-day recordings of patterned human iPSC-derived neuronal networks with a single medium exchange, revealing the spontaneous emergence and consolidation of spatiotemporal firing patterns during maturation. All design files are provided to facilitate adoption across culturing platforms, enabling un-interrupted longitudinal interfacing with network dynamics for studies of plasticity, chronic pharmacology, and developmental trajectories in individual cultures.

Corrigendum: How do circadian rhythms and neural synchrony shape networked cooperation?