Biology subjects

Malthe-Sorenssen, A.

Publications and source records attributed to Malthe-Sorenssen, A..

9 recordsLinked to original sources

bioRxiv2025

Deep Learning for Predicting Stem Cell Efficiency for use in Beta Cell Differentiation

Recent clinical trial data have shown that cell therapy holds curative potential for type-1 diabetes, however the large amounts of lab-grown cells required is a substantial bottleneck. The necessary cell differentiation process exhibits substantial variability, even among clones of induced pluripotent stem cells generated from the same patient. In contrast to many established data sets in medical imaging, human experts struggle to see the difference between highly- and lowly-efficient cell clones. We therefore propose an image-based deep learning model to guide the selection of the most efficient stem cell clones. We apply different deep learning model architectures to learn the morphological differences between good and bad stem cell clones and classify them based on only phase-contrast imaging. To gain critical insight into the learned features and enhance trust in our model, we use layer-wise relevance propagation (LRP), and Fourier-based frequency analysis. Using an EfficientNet-V2-S model, we obtain a novel early prediction for the outcome of the differentiation process from patient-derived stem cell to {beta}-cells using only phase-contrast images. Clone level accuracy is 76.7 % at 24 hours and 96.7 % at 53 hours after start of differentiation. The LRP-generated attribution maps and structure factor analysis show that the structure of the cell population is an important predictive feature. All in all, we present a highly predictive model for successful stem cell differentiation from phase-contrast images, which learns biologically relevant features. This study is a proof-of-concept that deep learning combined with label-free imaging can guide selection of induced pluripotent stem cell clones, thereby reduce cost of {beta}-cell production, and bring curative treatments one step closer to the clinic.

bioinformatics↗

bioRxiv2024

Task and Behavior-Related Variables Are Encoded by the Postrhinal and Medial Entorhinal Cortex During Non-Spatial Associative Learning

The medial entorhinal cortex (MEC) is pivotal in spatial computations and episodic memory. In particular, an animals position can be decoded from the activity of entorhinal grid cells. However, it remains elusive whether MEC could play a more general role in different types of associative learning and how the representations develop during the learning process. It has been shown that the postrhinal cortex (POR), which is directly connected to MEC, integrates visual stimuli with salient outcomes. Here, we use a non-spatial visual association task to investigate whether MEC neurons represent low-level visual cues during learning. Using a Go/NoGo visual association task, we recorded neural activity in MEC and POR throughout the learning phase as mice associated drifting gratings with rewarded, aversive, or neutral outcomes. Our findings reveal that the neural tuning curves in both the POR and MEC change with the learning of the task. From the start of training, the POR neurons exhibited response tuning to the visual cues, and the tuning was stable to cue orientations during learning. In contrast, MEC neurons did not initially respond very strongly to visual cues but developed a robust tuning toward the rewarded trials. While the MECs representation of visual information was limited, it encoded other task elements. A large fraction of the neurons formed distinct functional clusters that were either activated or suppressed by reward-related behavior. Remarkably, these clusters segregated anatomically in MEC and maintained strong within-cluster correlations before and after training. Notably, although the same functional clusters were apparent in the POR, they did not show any anatomical structure as in the MEC. Task reversal induced significant changes in network responses across both regions, with a decrease in overall task-responsive neurons but a slight increase in stimulus representation. Strikingly, information about the choice to lick emerged with learning in both brain areas, and most significantly within the functional cell clusters representing reward consumption and plus-cue stimulus. Our results demonstrate that although neurons in MEC and POR develop behavior-modulated tuning during learning of a non-spatial visual association task, the MEC exhibits stronger within-cluster correlations and anatomical organization. Conversely, the POR population exhibits less structural organization and more specific stimulus-tuning, which is reflective of being a higher visual association area. Our findings reveal that the MEC can encode task- and behavior-related variables beyond spatial information.

neuroscience↗

bioRxiv2024

Cortical Reactivations Modulated by Local InhibitoryCircuits Mediate Memory Consolidation

Highly salient events activate neurons across various brain regions. During subsequent rest or sleep, the activity patterns of these neurons often correlate with those observed during the preceding experience. Growing evidence suggests that these reactivations play a crucial role in memory consolidation, the process by which experiences are solidified in cortical networks for long-term storage. Here, we demonstrate that reactivations in the lateral visual cortex are vital for the consolidation of visual association learning. By employing longitudinal two-photon Ca2+ imaging alongside paired LFP recordings in the hippocampus and cortex, we show that targeted manipulation of PV+ inhibitory neurons in the lateral visual cortex after daily training selectively attenuated cue-specific reactivations and learning, with no apparent effect on normal network function during training. In contrast, reactivations in the control group were biased towards salient cues, aligned with learning process and persisted for hours after training had ended. Overall, our results underscore a crucial role for cortical reactivations in memory consolidation.

neuroscience↗

bioRxiv2024

Self-Supervised Grid Cells Without Path Integration

Grid cells, found in the medial Entorhinal Cortex, are known for their regular spatial firing patterns. These cells have been proposed as the neural solution to a range of computational tasks, from performing path integration, to serving as a metric for space. Their exact function, however, remains fiercely debated. In this work, we explore the consequences of demanding distance preservation over small spatial scales in networks subject to a capacity constraint. We consider two distinct self-supervised models, a feedforward network that learns to solve a purely spatial encoding task, and a recurrent network that solves the same problem during path integration. We find that this task leads to the emergence of highly grid cell-like representations in both networks. However, the recurrent network also features units with band-like representations. We subsequently prune velocity inputs to subsets of recurrent units, and find that their grid score is negatively correlated with path integration contribution. Thus, grid cells emerge without path integration in the feedforward network, and they appear substantially less important than band cells for path integration in the recurrent network. Our work provides a minimal model for learning grid-like spatial representations, and questions the role of grid cells as neural path integrators. Instead, it seems that distance preservation and high population capacity is a more likely candidate task for learning grid cells in artificial neural networks.

neuroscience↗

bioRxiv2024

Decoding the Cognitive map: Learning place cells and remapping

Hippocampal place cells are known for their spatially selective firing and are believed to encode an animals location while forming part of a cognitive map of space. These cells exhibit marked tuning curve and rate changes when an animals environment is sufficiently manipulated, in a process known as remapping. Place cells are accompanied by many other spatially tuned cells such as border cells and grid cells, but how these cells interact during navigation and remapping is unknown. In this work, we build a normative place cell model wherein a neural network is tasked with accurate position reconstruction and path integration. Motivated by the notion of a cognitive map, the networks position is estimated directly from its learned representations. To obtain a position estimate, we propose a non-trainable decoding scheme applied to network output units, inspired by the localized firing patterns of place cells. We find that output units learn place-like spatial representations, while upstream recurrent units become boundary-tuned. When the network is trained to perform the same task in multiple simulated environments, its place-like units learn to remap like biological place cells, displaying global, geometric and rate remapping. These remapping abilities appear to be supported by rate changes in upstream units. While the model does not learn grid-like units, its place unit centers form clusters organized in a hexagonal lattice in open fields. When we decode the center locations of CA1 place fields in mice, we find preliminary evidence of a similar clustering tendency. This suggests a potential mechanism for the interaction between place cells, border cells, and grid cells. Our model provides a normative framework for learning spatial representations previously reserved for biological place cells, providing new insight into place cell field formation and remapping.

neuroscience↗

bioRxiv2024

Hexagons all the way down: Grid cells as a conformal isometric map of space

The brains ability to navigate is often attributed to spatial cells in the hippocampus and entorhinal cortex. Grid cells, found in the entorhinal cortex, are known for their hexagonal spatial activity patterns and are traditionally believed to be the neural basis for path integration. However, recent studies have cast grid cells as a distance-preserving representation. We further investigate this role in a model of grid cells based on a superposition of plane waves. In a module of such grid cells, we optimise their phases to form a conformal isometry (CI) of two-dimensional flat space. With this setup, we demonstrate that a module of at least seven grid cells can achieve a CI, with phases forming a regular hexagonal arrangement. This pattern persists when increasing the number of cells, significantly diverging from a random uniform distribution. In particular, when optimised for CI, the phase distribution becomes distinctly regular and hexagonal, offering a clear experimentally testable prediction. Moreover, grid modules encoding a CI maintain constant energy expenditure across space, providing a new perspective on the role of energy constraints in normative models of grid cells. Finally, we investigate the minimum number of grid cells required for various spatial encoding tasks, including a unique representation of space, the population activity forming a torus, and achieving a CI, where we find that all three are achieved when the module encodes a CI. Our study not only underscores the versatility of grid cells beyond path integration but also highlights the importance of geometric principles in neural representations of space.

neuroscience↗

bioRxiv2023

Acoustic wave-induced stroboscopic optical mechanotyping of adherent cells

In this study, we present a novel, high content technique using an innovative cylindrical acoustic transducer, stroboscopic fast imaging and homodyne detection to recover the mechanical properties (dynamic shear modulus) of living adherent cells at low ultrasonic frequencies. By analyzing the micro-oscillations of cells we were able to simultaneously mechanotype whole populations of cells with sub-cellular resolution. The technique can be combined with standard fluorescence imaging allowing to further cross-correlate biological and mechanical information. We demonstrate the potential of the technique by mechanotyping co-cultures of different cell types with significantly different mechanical properties.

biophysics↗

bioRxiv2018

A Kirchhoff-Nernst-Planck framework for modeling large scale extracellular electrodiffusion surrounding morphologically detailed neurons

1Many pathological conditions, such as seizures, stroke, and spreading depression, are associated with substantial changes in ion concentrations in the extracellular space (ECS) of the brain. An understanding of the mechanisms that govern ECS concentration dynamics may be a prerequisite for understanding such pathologies. To estimate the transport of ions due to electrodiffusive effects, one must keep track of both the ion concentrations and the electric potential simultaneously in the relevant regions of the brain. Although this is currently unfeasible experimentally, it is in principle achievable with computational models based on biophysical principles and constraints. Previous computational models of extracellular ion-concentration dynamics have required extensive computing power, and therefore have been limited to either phenomena on very small spatiotemporal scales (micrometers and milliseconds), or simplified and idealized 1-dimensional (1-D) transport processes on a larger scale. Here, we present the 3-D Kirchhoff-Nernst-Planck (KNP) framework, tailored to explore electrodiffusive effects on large spatiotemporal scales. By assuming electroneutrality, the KNP-framework circumvents charge-relaxation processes on the spatiotemporal scales of nanometers and nanoseconds, and makes it feasible to run simulations on the spatiotemporal scales of millimeters and seconds on a standard desktop computer. In the present work, we use the 3-D KNP framework to simulate the dynamics of ion concentrations and the electrical potential surrounding a morphologically detailed pyramidal cell. In addition to elucidating the single neuron contribution to electrodiffusive effects in the ECS, the simulation demonstrates the efficiency of the 3-D KNP framework. We envision that future applications of the framework to more complex and biologically realistic systems will be useful in exploring pathological conditions associated with large concentration variations in the ECS.\n\n2 Author summaryMany pathological conditions, such as epilepsy and cortical spreading depression, are linked to abnormal extracellular ion concentrations in the brain. Understanding the underlying principles of such conditions may prove important in developing treatments for these illnesses, which incur societal costs of tens of billions annually. In order to investigate the role of ion-concentration dynamics in the pathological conditions, one must measure the spatial distribution of all ion concentrations over time. This remains challenging experimentally, which makes computational modeling an attractive tool. We have previously introduced the Kirchhoff-Nernst-Planck framework, an efficient framework for modeling electrodiffusion. In this study, we introduce a 3-dimensional version of this framework and use it to model the electrodiffusion of ions surrounding a morphologically detailed neuron. The simulation covered a 1 mm3 cylinder of tissue for over a minute and was performed in less than a day on a standard desktop computer, demonstrating the frameworks efficiency. We believe this to be an important step on the way to understanding phenomena involving ion concentration shifts at the tissue level.

neuroscience↗

bioRxiv2018

Experimental Directory Structure (Exdir): An alternative to HDF5 without introducing a new file format

Natural sciences generate an increasing amount of data in a wide range of formats developed by different research groups and commercial companies. At the same time there is a growing desire to share data along with publications in order to enable reproducible research. Open formats have publicly available specifications which facilitate data sharing and reproducible research. Hierarchical Data Format 5 (HDF5) is a popular open format widely used in neuroscience, often as a foundation for other, more specialized formats. However, drawbacks related to HDF5s complex specification have initiated a discussion for an improved replacement. We propose a novel alternative, the Experimental Directory Structure (Exdir), an open standard for data storage in experimental pipelines which amends drawbacks associated with HDF5 while retaining its advantages. HDF5 stores data and metadata in a hierarchy within a complex binary file which, among other things, is not human-readable, not optimal for version control systems, and lacks support for storing raw data. Exdir, one the other hand, uses file system directories to represent the hierarchy, with metadata stored in human-readable YAML files, datasets stored in binary NumPy files, and raw data stored directly in subdirectories. Furthermore, storing data in multiple files makes it easier to track for version control systems. Exdir is not a file format in itself, but a standard for organizing files in a directory structure. Exdir uses the same abstractions as HDF5 and is compatible with the HDF5 Abstract Data Model. Several research groups are already using data stored in a directory hierarchy as an alternative to HDF5, but no common standard exists in the field. This complicates and limits the opportunity for data sharing and development of common tools for reading, writing, and analyzing data. Exdir facilitates improved data storage, data sharing, reproducible research, and novel insight from interdisciplinary collaboration. With the publication of Exdir, we invite the scientific community to join the development to create an open standard that will serve as many needs as possible and that will serve as a foundation for open access to and exchange of data.\n\nSIGNIFICANCE STATEMENTExperimental Directory Structure (Exdir) is a proposal to standardize a storage solution that has become an increasingly popular alternative to Hierarchical Data Format 5 (HDF5), namely to use directories to define a hierarchy, store data in binary files, and metadata in text files. While this strategy is deployed locally by several research groups, no common standard exists. We envision the establishment of such a standard and present Exdir to the community as a starting point.

bioinformatics↗