Identifying Interpretable Latent Factors with Sparse Component Analysis

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Feb. 6, 2024

Abstract In many neural populations, the computationally relevant signals are posited to be a set of ‘latent factors’ – shared across individual neurons. Understanding relationship between activity and behavior requires identification factors that reflect distinct computational roles. Methods for identifying such typically require supervision, which can suboptimal if one is unsure how (or whether) grouped into distinct, meaningful sets. Here, we introduce Sparse Component Analysis (SCA), an unsupervised method identifies interpretable latent factors. SCA seeks sparse in time occupy orthogonal dimensions. With these simple constraints, facilitates surprisingly clear parcellations range behaviors. We applied motor cortex from reaching cycling monkeys, single-trial imaging data C. elegans , multitask artificial network. consistently identified sets were useful describing network computations.

Language: Английский

---

An emerging view of neural geometry in motor cortex supports high-performance decoding

eLife, Journal Year: 2025, Volume and Issue: 12

Published: Feb. 3, 2025

Decoders for brain-computer interfaces (BCIs) assume constraints on neural activity, chosen to reflect scientific beliefs while yielding tractable computations. Recent advances suggest that the true especially its geometry, may be quite different from those assumed by most decoders. We designed a decoder, MINT, embrace statistical are potentially more appropriate. If accurate, MINT should outperform standard methods explicitly make assumptions. Additionally, competitive with expressive machine learning can implicitly learn data. performed well across tasks, suggesting assumptions well-matched outperformed other interpretable in every comparison we made. 37 of 42 comparisons. MINT’s computations simple, scale favorably increasing neuron counts, and yield quantities such as data likelihoods. performance simplicity it strong candidate many BCI applications.

Language: Английский

---

Mind In Vitro Platforms: Versatile, Scalable, Robust, and Open Solutions to Interfacing with Living Neurons

Advanced Science, Journal Year: 2023, Volume and Issue: 11(11)

Published: Dec. 31, 2023

Abstract Motivated by the unexplored potential of in vitro neural systems for computing and corresponding need versatile, scalable interfaces multimodal interaction, an accurate, modular, fully customizable, portable recording/stimulation solution that can be easily fabricated, robustly operated, broadly disseminated is presented. This approach entails a reconfigurable platform works across multiple industry standards enables complete signal chain, from substrates sampled through micro‐electrode arrays (MEAs) to data acquisition, downstream analysis, cloud storage. Built‐in modularity supports seamless integration electrical/optical stimulation fluidic interfaces. Custom MEA fabrication leverages maskless photolithography, favoring rapid prototyping variety configurations, spatial topologies, constitutive materials. Through dedicated analysis management software suite, utility robustness this system are demonstrated cultures applications, including embryonic stem cell‐derived primary neurons, organotypic brain slices, 3D engineered tissue mimics, concurrent calcium imaging, long‐term recording. Overall, technology, termed “mind vitro” underscore inspiration, provides end‐to‐end widely deployed due its affordable (>10× cost reduction) open‐source nature, catering expanding needs both conventional unconventional electrophysiology.

Language: Английский

---

An emerging view of neural geometry in motor cortex supports high-performance decoding

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: April 6, 2023

Decoders for brain-computer interfaces (BCIs) assume constraints on neural activity, chosen to reflect scientific beliefs while yielding tractable computations. Recent advances suggest that the true especially its geometry, may be quite different from those assumed by most decoders. We designed a decoder, MINT, embrace statistical are potentially more appropriate. If accurate, MINT should outperform standard methods explicitly make assumptions. Additionally, competitive with expressive machine learning can implicitly learn data. performed well across tasks, suggesting assumptions well-matched outperformed other interpretable in every comparison we made. 37 of 42 comparisons. MINT’s computations simple, scale favorably increasing neuron counts, and yield quantities such as data likelihoods. performance simplicity it strong candidate many BCI applications.

Language: Английский

---

Geometry of population activity in spiking networks with low-rank structure

PLoS Computational Biology, Journal Year: 2023, Volume and Issue: 19(8), P. e1011315 - e1011315

Published: Aug. 7, 2023

Recurrent network models are instrumental in investigating how behaviorally-relevant computations emerge from collective neural dynamics. A recently developed class of based on low-rank connectivity provides an analytically tractable framework for understanding structure determines the geometry low-dimensional dynamics and ensuing computations. Such however lack some fundamental biological constraints, particular represent individual neurons terms abstract units that communicate through continuous firing rates rather than discrete action potentials. Here we examine far theoretical insights obtained rate networks transfer to more biologically plausible spiking neurons. Adding a top random excitatory-inhibitory connectivity, systematically compare activity integrate-and-fire with statistically equivalent connectivity. We show mean-field predictions allow us identify at constant population-average networks, as well novel non-linear regimes such out-of-phase oscillations slow manifolds. finally exploit these results directly build perform nonlinear

Language: Английский

---

Linking Neural Manifolds to Circuit Structure in Recurrent Networks

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Feb. 28, 2024

Abstract The classic view of cortical circuits composed precisely tuned neurons hardly accounts for large-scale recordings indicating that neuronal populations are heterogeneous and exhibit activity patterns evolving on low-dimensional manifolds. Using a modelling approach, we connect these two contrasting views. Our recurrent spiking network models explicitly link the circuit structure with dynamics population activity. Importantly, show different can lead to equivalent dynamics. Nevertheless, design method retrieving from test it simulated data. approach not only unifies established collective dynamics, but also paves way identifying elements experimental recordings.

Language: Английский

---

Flexible Control of Motor Units: Is the Multidimensionality of Motor Unit Manifolds a Sufficient Condition?

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: July 24, 2024

Abstract Understanding flexibility in the neural control of movement requires identifying distribution common inputs to motor units. In this study, we identified large samples units from two lower limb muscles: vastus lateralis (VL; up 60 units/participant) and gastrocnemius medialis (GM; 67 units/participant). First, applied a linear dimensionality reduction method assess manifolds underlying unit activity. We subsequently investigated under conditions: sinusoidal contractions with torque feedback, online visual feedback on firing rates. Overall, found that activity GM was effectively captured by single latent factor defining unidimensional manifold, whereas VL were better represented three factors multidimensional manifold. Despite difference dimensionality, recruitment muscles exhibited similarly low levels flexibility. Using spiking network model, tested hypothesis derived factorization does not solely represent descending cortical commands but is also influenced spinal circuitry. demonstrated heterogeneous units, or specific configurations recurrent inhibitory circuits, could produce This study clarifies an important debated issue, demonstrating while firings non-compartmentalised muscle can lie central nervous system may still have limited capacity for flexible these Key points To generate movement, distributes both excitatory The level remains topic debate significant implications smallest control. By combining experimental data silico models, sample be manifold; however, show very their recruitment. manifold directly reflect instead relate organisation local circuits.

Language: Английский

---

Representation of a perceptual bias in the prefrontal cortex

Proceedings of the National Academy of Sciences, Journal Year: 2024, Volume and Issue: 121(50)

Published: Dec. 5, 2024

Perception is influenced by sensory stimulation, prior knowledge, and contextual cues, which collectively contribute to the emergence of perceptual biases. However, precise neural mechanisms underlying these biases remain poorly understood. This study aims address this gap analyzing recordings from prefrontal cortex (PFC) monkeys performing a vibrotactile frequency discrimination task. Our findings provide empirical evidence supporting hypothesis that can be reflected in activity PFC. We found state-space trajectories PFC neuronal encoded warped representation first presented during Remarkably, distorted aligned with predictions its Bayesian estimator. The identification correlates expands our understanding basis highlights involvement shaping experiences. Similar analyses could employed other delayed comparison tasks various brain regions explore where how reflects different stages trial.

Language: Английский

---

Hippocampome.org 2.0 is a knowledge base enabling data-driven spiking neural network simulations of rodent hippocampal circuits

eLife, Journal Year: 2024, Volume and Issue: 12

Published: Feb. 12, 2024

Hippocampome.org is a mature open-access knowledge base of the rodent hippocampal formation focusing on neuron types and their properties. Previously, v1.0 established foundational classification system identifying 122 based axonal dendritic morphologies, main neurotransmitter, membrane biophysics, molecular expression (Wheeler et al., 2015). Releases v1.1 through v1.12 furthered aggregation literature-mined data, including among others counts, spiking patterns, synaptic physiology, in vivo firing phases, connection probabilities. Those additional properties increased online information content this public resource over 100-fold, enabling numerous independent discoveries by scientific community. v2.0, introduced here, besides incorporating 50 new types, now recenters its focus extending functionality to build real-scale, biologically detailed, data-driven computational simulations. In all cases, freely downloadable model parameters are directly linked specific peer-reviewed empirical evidence from which they were derived. Possible research applications include quantitative, multiscale analyses circuit connectivity neural network simulations activity dynamics. These advances can help generate precise, experimentally testable hypotheses shed light mechanisms underlying associative memory spatial navigation.

Language: Английский

---

Orthogonality of sensory and contextual categorical dynamics embedded in a continuum of responses from the second somatosensory cortex

Proceedings of the National Academy of Sciences, Journal Year: 2024, Volume and Issue: 121(29)

Published: July 11, 2024

How does the brain simultaneously process signals that bring complementary information, like raw sensory and their transformed counterparts, without any disruptive interference? Contemporary research underscores brain’s adeptness in using decorrelated responses to reduce such interference. Both neurophysiological findings artificial neural networks support notion of orthogonal representation for signal differentiation parallel processing. Yet, where, how are into more abstract representations remains unclear. Using a temporal pattern discrimination task trained monkeys, we revealed second somatosensory cortex (S2) efficiently segregates faithful subspaces. Importantly, S2 population encoding signals, but not ones, disappeared during nondemanding version this task, which suggests transformation decoding from downstream areas only active on-demand. A mechanistic computation model points gain modulation as possible biological mechanism observed context-dependent computation. Furthermore, individual activities underlie exhibited continuum responses, with no well-determined clusters. These advocate brain, while employing heterogeneous splits subspaces fashion enhance robustness, performance, improve coding efficiency.

Language: Английский

---