Biology subjects

Charles, A.

Publications and source records attributed to Charles, A..

17 recordsLinked to original sources

bioRxiv2026

Uncovering internal states with a robust shared-state multi-neuron GLM-HMM framework

Neural systems exhibit multiple firing states that reflect an organisms internal state and modulate the relationship between external environmental stimuli and behavior. Several studies have inferred these latent states by supplementing the traditional hidden Markov Model (HMM) with generalized linear models (GLMs) with non-Poisson behavioral observations. However, understanding the relationship between internal brain states and behavior also requires modeling the neural activity. Nonetheless, fitting multi-neuron GLM-HMMs is non-trivial due to high sparsity, collinearity, and low trial counts in neuronal datasets. Therefore, we built a robust multi-neuron GLM-HMM framework that uncovers latent states from population activity while incorporating the influence of time-stamped task variables and spike histories. To obtain reliable model parameters, we employ a modified expectation-maximization procedure. Specifically, we show that incorporating neuron-adaptive penalization in the maximization step overcomes the covariate co-linearity issues typical of time-stamped events and sparse spiking, yielding stable estimates of Poisson GLM coefficients. Furthermore, we incorporate a trust-region algorithm to ensure stable M-step convergence in the presence of ill-conditioned Hessians that can lead to unstable Newton-Raphson updates. We further demonstrate the utility of leave-one-out cross-validation analysis for evaluating model performance on datasets with low trial counts and without breaking their temporal structure. We evaluate our framework on three electrophysiological datasets from primates and rodents as they perform a decision-making task, demonstrate stable model convergence, and discuss the behavioral relevance of the inferred states. Author SummaryNeural systems evolve over time: not only do the individual neurons influence each other across the network, but the network and interconnections themselves change as an animal enters different behaviors (e.g., attentive vs. disengaged) or states (e.g., hungry or tired). Analyzing the neural activity that guides behavior thus must incorporate the time-varying nature of the brain. Recent modeling work has extended the popular Generalized Linear Model, a model that can connect task and behavior to recorded neural action potentials, to incorporate a latent Hidden Markov Model. This extension allows the resulting GLM-HMM to exhibit several different relationships (different GLMs) that are switched between over time to account for the animals changing patterns. While GLM-HMMs have been applied extensively on behavioral data (e.g., task choice in a decision making paradigm), neural data is much more difficult due to the smaller sample sizes, sparser activity, and larger parameter space. Our work presents a new fitting approach and best practices to robustly fit GLM-HMMs to neural data. We demonstrate through numerous applications to a variety of neural datasets that by robustly fitting GLM-HMMs to data, we can identify important features of neural activity that let us better understand its relationship to behavior.

neuroscience↗

bioRxiv2026

PN6047 Demonstrates Broad Preclinical Efficacy in Headache Models as a Novel Delta-Opioid Receptor Agonist

BackgroundThe Delta-opioid receptor (DOR) has gained attention as a promising target for the treatment of migraine and headache disorders. This is largely attributed to its unique pharmacological profile, which suggests that DOR-targeting treatment offers effective therapeutic benefit with a lower risk of medication overuse headache (MOH), reduced abuse liability, and minimal potential for physical dependence. These advantages have driven the development of a novel DOR agonist PN6047 (3-[[4-(dimethylcarbamoyl) phenyl]-[1-(thiazol-5-ylmethyl)-4-piperidylidene] methyl]benzamide), which has completed Phase I clinical trial and showed a favorable safety and tolerability profile. Although PN6047 has shown promising effects in neuropathic pain models, its efficacy in preclinical models of headache-associated pain remains to be evaluated. Here, we investigated the effects of PN6047 in models of migraine-associated pain and aura as well as post-traumatic headache (PTH) and MOH. MethodsC57BL6/J mice were used to examine the effects of PN6047 in the following migraine models: chronic intermittent nitroglycerin (NTG)-induced migraine-associated pain, PTH, KCl-induced cortical spreading depression (CSD), and optogenetic evoked CSD in a freely behaving transgenic mice expressing ChR2-eYFP. In addition, we tested whether chronic PN6047 induced MOH and whether it could prevent the development of MOH induced by sumatriptan. ResultsA single injection of PN6047 blocked chronic cephalic allodynia established by chronic intermittent NTG and PTH. Moreover, chronic PN6047 treatment prevented the development of MOH induced by sumatriptan, without causing MOH itself. In addition, PN6047 significantly reduced the number of CSD events in the KCl-induced CSD model, and delayed CSD onset triggered in freely behaving mice along with subsequent CSD-evoked allodynia. ConclusionPN6047, a novel DOR agonist, strikingly blocks headache-associated mechanism and symptoms in preclinical models of chronic migraine, migraine aura, PTH, and MOH. Importantly, prolonged PN6047 treatment did not induce MOH or analgesic tolerance. Together, these data demonstrate that despite the distinct mechanisms underlying migraine and headache disorder, PN6047 exhibits robust efficacy without inducing MOH, and displays a favorable safety and tolerability profile.

neuroscience↗

bioRxiv2025

Compositional Analysis of Mouse Sperm Subpopulations During Capacitation In Vitro

Mammalian sperm must undergo capacitation to become competent for fertilization, yet this process is marked by substantial phenotypic heterogeneity among sperm cells. How such variability emerges and how it relates to fertilizing potential remain unresolved, in part because sperm subpopulations are typically analyzed independently despite being intrinsically interdependent. Here, we combine large-scale single-cell spectral flow cytometry with compositional statistical modeling to quantify how sperm population structure responds to controlled capacitation signals in vitro. Using cauda epididymal mouse sperm, we implemented a two-dimensional assay that systematically varies extracellular bicarbonate and free calcium--key regulators of capacitation--and classified millions of individual cells into four irreversible physiological states defined by cell viability and acrosome reaction status. We show that bicarbonate and calcium interact nonlinearly to redistribute sperm across these subpopulations, revealing structured responses that would be otherwise obscured in measurements lacking single-cell resolution. Elevated intracellular calcium was associated with increased cell death, while the highest proportions of live, acrosome-reacted sperm occurred under relatively low extracellular calcium conditions. To enable subpopulation-level analysis, we applied a hierarchical Dirichlet-multinomial regression model that accounts for multinomial sampling noise and between-male variability, yielding posterior probability surfaces that describe how sperm subpopulations reallocate across functional states as microenvironmental signaling conditions change. Together, these results demonstrate that capacitation is a stochastic, cell-population level process shaped by structured phenotypic heterogeneity. This framework provides a quantitative foundation for linking sperm subpopulation composition to measures of fertility competence and for improving existing interpretation of flow cytometry-based assessments of male fertility.

cell biology↗

bioRxiv2025

CUSP: Complex Spike Sorting from Multi-electrode Array Recordings with U-net Sequence-to-Sequence Prediction

BackgroundComplex spikes (CSs) in cerebellar Purkinje cells convey unique signals complementary to Simple spike (SS) action potentials, but are infrequent and variable in waveform. Their variability and low spike counts, combined with recording artifacts such as electrode drift, make automated detection challenging. New MethodWe introduce CUSP (CS sorting via U-net Sequence Prediction), a fully automated deep learning framework for CS sorting in high-density multi-electrode array recordings. CUSP uses a U-Net architecture with hybrid self-attention inception blocks to integrate local field potential and action potential signals and outputs CS event probabilities in a sequence-to-sequence manner. Detected events are clustered and paired with concurrently detected SSs to reconstruct the complete Purkinje cell activity. ResultsTrained on cerebellar neuropixels recordings in rhesus macaques, CUSP achieves human-expert performance (F1 = 0.83 {+/-} 0.03) and even captures valid CS events overlooked during manual annotation. Comparison with Existing MethodsCUSP outperforms traditional and state-of-the-art CS and SS sorting algorithms on CS detection. It remains robust to waveform variability, spikelet composition, and electrode drift, enabling accurate CS tracking in long-term recordings. In contrast, existing methods often show false-positive biases or degrade under drift. ConclusionsCUSP provides a scalable, robust framework for analyzing burst-like or dynamically complex spike patterns. Its generalizability makes it valuable for large-scale cerebellar datasets and other neural systems, such as hippocampal pyramidal cells, where complex bursts are critical for computation. By combining expert-level accuracy with automation, CUSP offers a broadly applicable solution for studying information coding across circuits.

neuroscience↗

bioRxiv2025

Improved inference of latent neural states from calcium imaging data

Calcium imaging (CI) is a standard method for recording neural population activity, as it enables simultaneous recording of hundreds-to-thousands of individual somatic signals. Accordingly, CI recordings are prime candidates for population-level latent variable analyses, for example, using models such as Gaussian Process Factor Analysis (GPFA), hidden Markov models (HMMs), and latent dynamical systems. However, these models have been primarily developed and fine-tuned for electrophysiological measurements of spiking activity. To adapt these models for use with the calcium signals recorded with CI, per-neuron fluorescence time-traces are typically either de-convolved to approximate spiking events or analyzed directly under Gaussian observation assumptions. The former approach, while enabling the direct application of latent variable methods developed for spiking data, suffers from the imprecise nature of spike estimation from CI. Moreover, isolated spikes can be undetectable in the fluorescence signal, creating additional uncertainty. A more direct model linking observed fluorescence to latent variables would account for these sources of uncertainty. Here, we develop accurate and tractable models for characterizing the latent structure of neural population activity from CI data. We propose to augment HMM, GPFA, and dynamical systems models with a CI observation model that consists of latent Poisson spiking and autoregressive calcium dynamics. Importantly, this model is both more flexible and directly compatible with standard methods for fitting latent models of neural dynamics. We demonstrate that using this more accurate CI observation model improves latent variable inference and model fitting on both CI observations generated using state-of-the-art biophysical simulations and imaging data recorded in an experimental setting. We expect the developed methods to be widely applicable to many different analyses of population CI data.

neuroscience↗

bioRxiv2025

Age- and Sex- Driven Transcriptional and Metabolic Diversity in Myeloid-Derived Suppressor Cells After Mouse Sepsis

Sepsis induces profound immune dysregulation, often resulting in chronic critical illness characterized by persistent immunosuppression and poor outcomes. Myeloid-derived suppressor cells (MDSCs) are central mediators of this immunosuppressive phenotype, yet the influence of age and sex on their transcriptional and metabolic states remain poorly understood. Here, we employed single-cell RNA sequencing of splenic leukocytes from young (3-4 months) and older (18-24 months) adult male and female mice subjected to a clinically relevant murine sepsis model to define age- and sex-specific MDSC phenotypes. We identified significant differences regarding age and sex in MDSC expansion, transcriptome, canonical pathway activation, RNA velocity, mitochondrial metabolism, and predicted cell-cell communication after sepsis. Using drug2cell analysis of total leukocytes we also identified cohort-specific drug target profiles. These findings underscore the importance of age and sex in shaping sepsis-induced MDSC biology and suggest that personalized immunomodulatory strategies targeting MDSCs could improve sepsis outcomes.

immunology↗

bioRxiv2025

Precise Extraction of Neural Motifs Reveals Multiscale, Parallel Encoding Schemes in Auditory Cortex

Neural population activity is often stereotyped into recurring activity patterns, i.e., neural motifs, which can be seen as the fundamental building blocks in sensory processing and cognition. In this work we study the codes carried by such neural motifs in primary auditory cortex A1, and analyze how they build on and complement traditional views of single-unit coding. In particular, we study, using two-photon calcium imaging (CI), how activity in A1 differentially represents both sensory stimuli and task and behavioral variables in each of the parallel single neuron and population motif scales. While CI enables the study of neural motifs by capturing the activity of large neural populations, identifying motifs in CI is hampered by the temporal imprecision of current motif-detection algorithms when applied to CI data. We thus developed a new algorithm for motif detection, which enabled us to identify widespread stimulus-encoding motifs as well as a small number of motifs jointly encoding stimulus and choice in Layer 2/3 of A1. These motifs consist of small groups of neurons that are neither clustered nor regularly ordered in space. Interestingly, active neurons within task-encoding motifs exhibit mixed encoding properties inconsistent with the motifs they participate in. Together, these results demonstrate how single unit activity and neural motifs in A1 L2/3 provide different levels of coding granularity containing different information in parallel within the greater neural population. Generally, our results indicate that downstream populations, by selecting which scale of a population drives them, can be selective in the information collected for later cognitive processing.

neuroscience↗

bioRxiv2025

Fast and accessible morphology-free functional fluorescence imaging analysis

Optical calcium imaging is a powerful tool for recording neural activity across a wide range of spatial scales, from dendrites and spines to whole-brain imaging through two-photon and widefield microscopy. Traditional methods for analyzing functional calcium imaging data rely heavily on spatial features, such as the compact shapes of somas, to extract regions of interest and their associated temporal traces. This spatial dependency can introduce biases in time trace estimation and limit the applicability of these methods across different neuronal morphologies and imaging scales. To address these limitations, the Graph Filtered Temporal Dictionary Learning (GraFT) uses a graph-based approach to identify neural components based on shared temporal activity rather than spatial proximity, enhancing generalizability across diverse datasets. Here we present significant advancements to the GraFT algorithm, including the integration of a more efficient solver for the L1 least absolute shrinkage and selection operator (LASSO) problem and the application of compressive sensing techniques to reduce computational complexity. By employing random projections to reduce data dimensionality, we achieve substantial speedups while maintaining analytical accuracy. These advancements significantly accelerate the GraFT algorithm, making it more scalable for larger and more complex datasets. Moreover, to increase accessibility, we developed a graphical user interface to facilitate running and analyzing the outputs of GraFT. Finally, we demonstrate the utility of GraFT to imaging data beyond meso-scale imaging, including vascular and axonal imaging.

bioinformatics↗

bioRxiv2025

Automatic detection of fluorescently labeled synapses in volumetric in vivo imaging data

Synapses are submicron structures that connect and enable communication between neurons. Many forms of learning are thought to be encoded by synaptic plasticity, wherein the strength of specific synapses is regulated by modulating expression of neurotransmitter receptors. For instance, regulation of AMPA-type glutamate receptors is a central mechanism controlling the formation and recollection of long-term memories. A critical step in understanding how synaptic plasticity drives behavior is thus to directly observe, i.e., image, fluorescently labeled synapses in living tissue. However, due to their small size and incredible density -- with one [~]0.5 {micro}m diameter synapse every cubic micron -- accurately detecting individual synapses and segmenting each from its closely abutting neighbors is challenging. To overcome this, we trained a convolutional neural network to simultaneously detect and separate densely labeled synapses. These tools significantly increased the accuracy and scale of synapse detection, enabling segmentation of hundreds of thousands of individual synapses imaged in living mice.

neuroscience↗

bioRxiv2024

Decomposed Linear Dynamical Systems (dLDS) modelsreveal context-dependent dynamic connectivity in C. elegans

Mounting empirical evidence indicates that neural "tuning" can be highly variable within an individual across time and especially across individuals. Furthermore, modulatory effects can change the relationship between neurons in the brain as a function of behavioral or other conditions, meaning that the changes in activity (the derivative) may be as important as the activity itself. However, current computational models fail to capture the nonstationarity and variability of neural coding, preventing the quantitative evaluation of these effects, especially during individuals adaptation to their environments. Here we present a novel way to study the effects of adaptation in one of the most well-studied organisms, C. elegans, leveraging recent advances in dynamical systems modeling, specifically decomposed Linear Dynamical Systems (dLDS).Our approach enables the discovery of multiple parallel neural processes on different timescales using a low-dimensional set of linear operators that can be recombined in different ratios. Our model identifies "dynamic connectivity," describing patterns of dynamic neural interactions in time. We use these patterns to identify instantaneous, contextually-dependent, hierarchical roles of neurons; discover the underlying variability of neural representations even under seemingly discrete behaviors; and learn a single aligned latent space underlying multiple individual worms activity. By analyzing individual worms and neurons, we found evidence that 1) changes in interneuron connectivity mediate efficient task-switching and 2) changes in sensory neuron connectivity show a mechanism of adaptation.

neuroscience↗

bioRxiv2023

Multi-day Neuron Tracking in High Density Electrophysiology Recordings using EMD

Accurate tracking of the same neurons across multiple days is crucial for studying changes in neuronal activity during learning and adaptation. Advances in high density extracellular electrophysiology recording probes, such as Neuropixels, provide a promising avenue to accomplish this goal. Identifying the same neurons in multiple recordings is, however, complicated by non-rigid movement of the tissue relative to the recording sites (drift) and loss of signal from some neurons. Here we propose a neuron tracking method that can identify the same cells independent of firing statistics, that are used by most existing methods. Our method is based on between-day non-rigid alignment of spike sorted clusters. We verified the same cell identity in mice using measured visual receptive fields. This method succeeds on datasets separated from one to 47 days, with an 84% average recovery rate.

neuroscience↗

bioRxiv2023

Revealing abrupt transitions from goal-directed to habitual behavior

A fundamental tenet of animal behavior is that decision-making involves multiple controllers. Initially, behavior is goal-directed, driven by desired outcomes, shifting later to habitual control, where cues trigger actions independent of motivational state. Clark Hulls question from 1943 still resonates today: "Is this transition abrupt, or is it gradual and progressive?"1 Despite a century-long belief in gradual transitions, this question remains unanswered2,3 as current methods cannot disambiguate goal-directed versus habitual control in real-time. Here, we introduce a novel volitional engagement approach, motivating animals by palatability rather than biological need. Offering less palatable water in the home cage4,5 reduced motivation to work for plain water in an auditory discrimination task when compared to water-restricted animals. Using quantitative behavior and computational modeling6, we found that palatability-driven animals learned to discriminate as quickly as water-restricted animals but exhibited state-like fluctuations when responding to the reward-predicting cue--reflecting goal-directed behavior. These fluctuations spontaneously and abruptly ceased after thousands of trials, with animals now always responding to the reward-predicting cue. In line with habitual control, post-transition behavior displayed motor automaticity, decreased error sensitivity (assessed via pupillary responses), and insensitivity to outcome devaluation. Bilateral lesions of the habit-related dorsolateral striatum7 blocked transitions to habitual behavior. Thus, volitional engagement reveals spontaneous and abrupt transitions from goal-directed to habitual behavior, suggesting the involvement of a higher-level process that arbitrates between the two.

neuroscience↗

bioRxiv2023

Host-cell Interactions of Engineered T cell Micropharmacies

Genetically engineered, cytotoxic, adoptive T cells localize to antigen positive cancer cells inside patients, but tumor heterogeneity and multiple immune escape mechanisms have prevented the eradication of most solid tumor types. More effective, multifunctional engineered T cells are in development to overcome the barriers to the treatment of solid tumors, but the interactions of these highly modified cells with the host are poorly understood. We previously engineered prodrug-activating enzymatic functions into chimeric antigen receptor (CAR) T cells, endowing them with an orthogonal killing mechanism to conventional T-cell cytotoxicity. These drug-delivering cells, termed Synthetic Enzyme-Armed KillER (SEAKER) cells, demonstrated efficacy in mouse lymphoma xenograft models. However, the interactions of an immunocompromised xenograft with such complex engineered T cells are distinct from those in an immunocompetent host, precluding an understanding of how these physiologic processes may affect the therapy. Here, we also expand the repertoire of SEAKER cells to target solid-tumor melanomas in syngeneic mouse models using specific targeting with TCR-engineered T cells. We demonstrate that SEAKER cells localize specifically to tumors, and activate bioactive prodrugs, despite host immune responses. We additionally show that TCR-engineered SEAKER cells are efficacious in immunocompetent hosts, demonstrating that the SEAKER platform is applicable to many adoptive cell therapies.

immunology↗

bioRxiv2022

Chronic brain functional ultrasound imaging in freely moving rodents performing cognitive tasks

Functional ultrasound imaging (fUS) is an emerging imaging technique that indirectly measures neural activity via changes in blood volume. To date it has not been used to image chronically during cognitive tasks in freely moving animals. Performing those experiments faces a number of exceptional challenges: performing large durable craniotomies with chronic implants, designing behavioural experiments matching the hemodynamic timescale, stabilizing the ultrasound probe during freely moving behavior, accurately assessing motion artifacts and validating that the animal can perform cognitive tasks at high performance while tethered. In this study, we provide validated solutions for those technical challenges. In addition, we present standardized step-by-step repro-ducible protocols, procedures and data processing pipelines that open up the opportunity to perform fUS in freely moving rodents performing complex cognitive tasks. Moreover, we present proof-of-concept analysis of brain dynamics during a decision making task. We obtain stable recordings from which we can robustly decode task variables from fUS data over multiple months. Moreover, we find that brain wide imaging through hemodynamic response is nonlinearly related to cognitive variables, such as task difficulty, as compared to sensory responses previously explored.

neuroscience↗

bioRxiv2020

Optimal Adaptive Electrode Selection to Maximize Simultaneously Recorded Neuron Yield

Neural-Matrix style, high-density electrode arrays for brain-machine interfaces (BMIs) and neuroscientific research require the use of multiplexing: Each recording channel can be routed to one of several electrode sites on the array. This capability allows the user to flexibly distribute recording channels to the locations where the most desirable neural signals can be resolved. For example, in the Neuropixel probe, 960 electrodes can be addressed by 384 recording channels. However, currently no adaptive methods exist to use recorded neural data to optimize/customize the electrode selections per recording context. Here, we present an algorithm called classification-based selection (CBS) that optimizes the joint electrode selections for all recording channels so as to maximize isolation quality of detected neurons. We show, in experiments using Neuropixels in non-human primates, that this algorithm yields a similar number of isolated neurons as would be obtained if all electrodes were recorded simultaneously. Neuron counts were 41-85% improved over previously published electrode selection strategies. The neurons isolated from electrodes selected by CBS were a 73% match, by spike timing, to the complete set of recordable neurons around the probe. The electrodes selected by CBS exhibited higher average per-recording-channel signal-to-noise ratio. CBS, and selection optimization in general, could play an important role in development of neurotechnologies for BMI, as signal bandwidth becomes an increasingly limiting factor. Code and experimental data have been made available1.

neuroscience↗

bioRxiv2020

Low-Dose CDK4/6 Inhibitors Induce Presentation of Pathway Specific MHC ligands as Targets for Cancer Immunotherapy

PurposeCyclin dependent kinase 4/6 inhibitors (CDK4/6i) lead to cell-cycle arrest but also demonstrate antineoplastic activity through triggering T cell-mediated immunity. One of the potential mechanisms responsible for this immunological effect might be qualitative and quantitative changes in human leukocyte antigen (HLA) ligands on the cell surface after treatment with CDK4/6i. These changes may increase the immunogenicity of breast cancer cells offering potential synergies for combinations with cancer immunotherapies. Experimental DesignWe investigated the ability of two CDK4/6 inhibitors (CDK4/6i), Abemaciclib and Palbociclib, to alter the immunopeptidome at subclinical, non-toxic, levels in different breast cancer cell lines. Biochemical isolation of HLA ligands, identification by mass spectrometry and subsequent network analysis after drug treatment were used to characterize the changes in the immunopeptidome. The mechanisms for altered CDK4/6 presentation were explored. ResultsLow-dose treatment with 100nM of Abemaciclib and Palbociclib led to upregulation of cell surface HLA levels and induced hundreds of HLA ligands in breast cancer cell lines. These new ligands were significantly and most strongly enriched for peptides derived from proteins involved in the "G1/S phase transition of cell cycle" pathway and included among others, HLA ligands from CDK4, CDK6, Cyclin D1 and Cyclin E1. An increase in transcript, protein, and subsequent ubiquitination for Cyclin D1, which could lead to enhanced degradation of the target protein, was identified as a potential mechanism for the altered presentation of peptides. ConclusionsCDK4/6i treatment gave rise to drug-induced antigens through cell cycle disruption and increased antigen presentation. Interestingly, these induced HLA ligands are often sourced from the proteins of the CDK4/CDK6/CCND1 complex or more downstream interaction partners, providing evidence that inhibition of a distinct cellular pathway leads to increased presentation of the proteins involved. These findings suggested CDK4/6i provided a tool for highly selective induction of HLA ligands that may be targeted by T cell-based immunotherapeutics. Translational RelevanceThese data demonstrated that low-dose treatment of breast cancer cells with CDK4/6 inhibitors, Abemaciclib and Palbociclib, induced marked changes in presentation of HLA ligands, especially from proteins involved in the G1/S phase transition, the phase in which these drugs arrest the cells. Enhanced ubiquitination and degradation was identified as a mechanism for the altered presentation for one of the relevant proteins. The induced HLA ligands may provide ideal specific targets for combination immunotherapies. The data show for the first time that selective inhibition of a distinct pathway can lead to specific presentation of HLA ligands in breast cancer cells. This work supports the rationale for testing the combination of low-dose CDK4/6i with immunotherapeutic agents, such as immune checkpoint blockade antibodies or T-cell-based approaches specifically directed against one of the induced HLA ligands.

cancer biology↗

bioRxiv2018

Sparse-Coding Variational Auto-Encoders

The sparse coding model posits that the visual system has evolved to efficiently code natural stimuli using a sparse set of features from an overcomplete dictionary. The original sparse coding model suffered from two key limitations, however: (1) computing the neural response to an image patch required minimizing a nonlinear objective function via recurrent dynamics; (2) fitting relied on approximate inference methods that ignored uncertainty. Although subsequent work has developed several methods to overcome these obstacles, we propose a novel solution inspired by the variational auto-encoder (VAE) framework. We introduce the sparse-coding variational auto-encoder (SVAE), which augments the sparse coding model with a probabilistic recognition model parametrized by a deep neural network. This recognition model provides a neurally plausible feedforward implementation for the mapping from image patches to neural activities, and enables a principled method for fitting the sparse coding model to data via maximization of the evidence lower bound (ELBO). The SVAE differs from standard VAEs in three key respects: the latent representation is overcomplete (there are more latent dimensions than image pixels), the prior is sparse or heavy-tailed instead of Gaussian, and the decoder network is a linear projection instead of a deep network. We fit the SVAE to natural image data under different assumed prior distributions, and show that it obtains higher test performance than previous fitting methods. Finally, we examine the response properties of the recognition network and show that it captures important nonlinear properties of neurons in the early visual pathway.

neuroscience↗