A deep learning framework for inference of single-trial neural population dynamics from calcium imaging with subframe temporal resolution

Interpretable deep learning for deconvolutional analysis of neural signals

The widespread adoption of deep learning to model neural activity often relies on "black-box" approaches that lack an interpretable connection between neural activity and network parameters. Here, we propose using algorithm unrolling, a method for interpretable deep learning, to design the architecture of sparse deconvolutional neural networks and obtain a direct interpretation of network weights in relation to stimulus-driven single-neuron activity through a generative model. We introduce our method, deconvolutional unrolled neural learning (DUNL), and demonstrate its versatility by applying it to deconvolve single-trial local signals across multiple brain areas and recording modalities. We uncover multiplexed salience and reward prediction error signals from midbrain dopamine neurons, perform simultaneous event detection and characterization in somatosensory thalamus recordings, and characterize the heterogeneity of neural responses in the piriform cortex and across striatum during unstructured, naturalistic experiments. Our work leverages advances in interpretable deep learning to provide a mechanistic understanding of neural activity.

MARBLE: interpretable representations of neural population dynamics using geometric deep learning

The dynamics of neuron populations commonly evolve on low-dimensional manifolds. Thus, we need methods that learn the dynamical processes over neural manifolds to infer interpretable and consistent latent representations. We introduce a representation learning method, MARBLE, which decomposes on-manifold dynamics into local flow fields and maps them into a common latent space using unsupervised geometric deep learning. In simulated nonlinear dynamical systems, recurrent neural networks and experimental single-neuron recordings from primates and rodents, we discover emergent low-dimensional latent representations that parametrize high-dimensional neural dynamics during gain modulation, decision-making and changes in the internal state. These representations are consistent across neural networks and animals, enabling the robust comparison of cognitive computations. Extensive benchmarking demonstrates state-of-the-art within- and across-animal decoding accuracy of MARBLE compared to current representation learning approaches, with minimal user input. Our results suggest that a manifold structure provides a powerful inductive bias to develop decoding algorithms and assimilate data across experiments.

Spatio-temporal transformers for decoding neural movement control

Objective. Deep learning tools applied to high-resolution neurophysiological data have significantly progressed, offering enhanced decoding, real-time processing, and readability for practical applications. However, the design of artificial neural networks to analyze neural activityin vivoremains a challenge, requiring a delicate balance between efficiency in low-data regimes and the interpretability of the results.Approach. To address this challenge, we introduce a novel specialized transformer architecture to analyze single-neuron spiking activity. The model is tested on multi-electrode recordings from the dorsal premotor cortex of non-human primates performing a motor inhibition task.Main results. The proposed architecture provides an early prediction of the correct movement direction, achieving accurate results no later than 230 ms after the Go signal presentation across animals. Additionally, the model can forecast whether the movement will be generated or withheld before a stop signal, unattended, is actually presented. To further understand the internal dynamics of the model, we compute the predicted correlations between time steps and between neurons at successive layers of the architecture, with the evolution of these correlations mirrors findings from previous theoretical analyses.Significance. Overall, our framework provides a comprehensive use case for the practical implementation of deep learning tools in motor control research, highlighting both the predictive capabilities and interpretability of the proposed architecture.

Nonnegative matrix factorization for analyzing state dependent neuronal network dynamics in calcium recordings

Calcium imaging allows recording from hundreds of neurons in vivo with the ability to resolve single cell activity. Evaluating and analyzing neuronal responses, while also considering all dimensions of the data set to make specific conclusions, is extremely difficult. Often, descriptive statistics are used to analyze these forms of data. These analyses, however, remove variance by averaging the responses of single neurons across recording sessions, or across combinations of neurons, to create single quantitative metrics, losing the temporal dynamics of neuronal activity, and their responses relative to each other. Dimensionally Reduction (DR) methods serve as a good foundation for these analyses because they reduce the dimensions of the data into components, while still maintaining the variance. Nonnegative Matrix Factorization (NMF) is an especially promising DR analysis method for analyzing activity recorded in calcium imaging because of its mathematical constraints, which include positivity and linearity. We adapt NMF for our analyses and compare its performance to alternative dimensionality reduction methods on both artificial and in vivo data. We find that NMF is well-suited for analyzing calcium imaging recordings, accurately capturing the underlying dynamics of the data, and outperforming alternative methods in common use.

Analysis methods for large-scale neuronal recordings

Simultaneous recordings from hundreds or thousands of neurons are becoming routine because of innovations in instrumentation, molecular tools, and data processing software. Such recordings can be analyzed with data science methods, but it is not immediately clear what methods to use or how to adapt them for neuroscience applications. We review, categorize, and illustrate diverse analysis methods for neural population recordings and describe how these methods have been used to make progress on longstanding questions in neuroscience. We review a variety of approaches, ranging from the mathematically simple to the complex, from exploratory to hypothesis-driven, and from recently developed to more established methods. We also illustrate some of the common statistical pitfalls in analyzing large-scale neural data.

Decoding multi-limb movements from two-photon calcium imaging of neuronal activity using deep learning

Objective.Brain-machine interfaces (BMIs) aim to restore sensorimotor function to individuals suffering from neural injury and disease. A critical step in implementing a BMI is to decode movement intention from recorded neural activity patterns in sensorimotor areas. Optical imaging, including two-photon (2p) calcium imaging, is an attractive approach for recording large-scale neural activity with high spatial resolution using a minimally-invasive technique. However, relating slow two-photon calcium imaging data to fast behaviors is challenging due to the relatively low optical imaging sampling rates. Nevertheless, neural activity recorded with 2p calcium imaging has been used to decode information about stereotyped single-limb movements and to control BMIs. Here, we expand upon prior work by applying deep learning to decode multi-limb movements of running mice from 2p calcium imaging data.Approach.We developed a recurrent encoder-decoder network (LSTM-encdec) in which the output is longer than the input.Main results.LSTM-encdec could accurately decode information about all four limbs (contralateral and ipsilateral front and hind limbs) from calcium imaging data recorded in a single cortical hemisphere.Significance.Our approach provides interpretability measures to validate decoding accuracy and expands the utility of BMIs by establishing the groundwork for control of multiple limbs. Our work contributes to the advancement of neural decoding techniques and the development of next-generation optical BMIs.

Cortico-thalamic communication for action coordination in a skilled motor sequence

The coordination of forelimb and orofacial movements to compose an ethological reach-to-consume behavior likely involves neural communication across brain regions. Leveraging wide-field imaging and photo-inhibition to survey across the cortex, we identified a cortical network and a high-order motor area (MOs-c), which coordinate action progression in a mouse reach-and-withdraw-to-drink (RWD) behavior. Electrophysiology and photo-inhibition across multiple projection neuron types within the MOs-c revealed differential contributions of pyramidal tract and corticothalamic (CTMOs) output channels to action progression and hand-mouth coordination. Notably, CTMOs display sustained firing throughout RWD sequence and selectively enhance RWD-relevant activity in postsynaptic thalamus neurons, which also contribute to action coordination. CTMOs receive converging monosynaptic inputs from forelimb and orofacial sensorimotor areas and are reciprocally connected to thalamic neurons, which project back to the cortical network. Therefore, motor cortex corticothalamic channel may selectively amplify the thalamic integration of cortical and subcortical sensorimotor streams to coordinate a skilled motor sequence.

Diffusion-Based Generation of Neural Activity from Disentangled Latent Codes

Recent advances in recording technology have allowed neuroscientists to monitor activity from thousands of neurons simultaneously. Latent variable models are increasingly valuable for distilling these recordings into compact and interpretable representations. Here we propose a new approach to neural data analysis that leverages advances in conditional generative modeling to enable the unsupervised inference of disentangled behavioral variables from recorded neural activity. Our approach builds on InfoDiffusion, which augments diffusion models with a set of latent variables that capture important factors of variation in the data. We apply our model, called Generating Neural Observations Conditioned on Codes with High Information (GNOCCHI), to time series neural data and test its application to synthetic and biological recordings of neural activity during reaching. In comparison to a VAE-based sequential autoencoder, GNOCCHI learns higher-quality latent spaces that are more clearly structured and more disentangled with respect to key behavioral variables. These properties enable accurate generation of novel samples (unseen behavioral conditions) through simple linear traversal of the latent spaces produced by GNOCCHI. Our work demonstrates the potential of unsupervised, information-based models for the discovery of interpretable latent spaces from neural data, enabling researchers to generate high-quality samples from unseen conditions.

Closed-loop experiments and brain machine interfaces with multiphoton microscopy

In the field of neuroscience, the importance of constructing closed-loop experimental systems has increased in conjunction with technological advances in measuring and controlling neural activity in live animals. We provide an overview of recent technological advances in the field, focusing on closed-loop experimental systems where multiphoton microscopy-the only method capable of recording and controlling targeted population activity of neurons at a single-cell resolution in vivo-works through real-time feedback. Specifically, we present some examples of brain machine interfaces (BMIs) using in vivo two-photon calcium imaging and discuss applications of two-photon optogenetic stimulation and adaptive optics to real-time BMIs. We also consider conditions for realizing future optical BMIs at the synaptic level, and their possible roles in understanding the computational principles of the brain.

FIOLA: an accelerated pipeline for fluorescence imaging online analysis

Optical microscopy methods such as calcium and voltage imaging enable fast activity readout of large neuronal populations using light. However, the lack of corresponding advances in online algorithms has slowed progress in retrieving information about neural activity during or shortly after an experiment. This gap not only prevents the execution of real-time closed-loop experiments, but also hampers fast experiment-analysis-theory turnover for high-throughput imaging modalities. Reliable extraction of neural activity from fluorescence imaging frames at speeds compatible with indicator dynamics and imaging modalities poses a challenge. We therefore developed FIOLA, a framework for fluorescence imaging online analysis that extracts neuronal activity from calcium and voltage imaging movies at speeds one order of magnitude faster than state-of-the-art methods. FIOLA exploits algorithms optimized for parallel processing on GPUs and CPUs. We demonstrate reliable and scalable performance of FIOLA on both simulated and real calcium and voltage imaging datasets. Finally, we present an online experimental scenario to provide guidance in setting FIOLA parameters and to highlight the trade-offs of our approach.

Expressive architectures enhance interpretability of dynamics-based neural population models

Artificial neural networks that can recover latent dynamics from recorded neural activity may provide a powerful avenue for identifying and interpreting the dynamical motifs underlying biological computation. Given that neural variance alone does not uniquely determine a latent dynamical system, interpretable architectures should prioritize accurate and low-dimensional latent dynamics. In this work, we evaluated the performance of sequential autoencoders (SAEs) in recovering latent chaotic attractors from simulated neural datasets. We found that SAEs with widely-used recurrent neural network (RNN)-based dynamics were unable to infer accurate firing rates at the true latent state dimensionality, and that larger RNNs relied upon dynamical features not present in the data. On the other hand, SAEs with neural ordinary differential equation (NODE)-based dynamics inferred accurate rates at the true latent state dimensionality, while also recovering latent trajectories and fixed point structure. Ablations reveal that this is mainly because NODEs (1) allow use of higher-capacity multi-layer perceptrons (MLPs) to model the vector field and (2) predict the derivative rather than the next state. Decoupling the capacity of the dynamics model from its latent dimensionality enables NODEs to learn the requisite low-D dynamics where RNN cells fail. Additionally, the fact that the NODE predicts derivatives imposes a useful autoregressive prior on the latent states. The suboptimal interpretability of widely-used RNN-based dynamics may motivate substitution for alternative architectures, such as NODE, that enable learning of accurate dynamics in low-dimensional latent spaces.

Dimensionality reduction of calcium-imaged neuronal population activity

Calcium imaging has been widely adopted for its ability to record from large neuronal populations. To summarize the time course of neural activity, dimensionality reduction methods, which have been applied extensively to population spiking activity, may be particularly useful. However, it is unclear if the dimensionality reduction methods applied to spiking activity are appropriate for calcium imaging. We thus carried out a systematic study of design choices based on standard dimensionality reduction methods. We also developed a method to perform deconvolution and dimensionality reduction simultaneously (Calcium Imaging Linear Dynamical System, CILDS). CILDS most accurately recovered the single-trial, low-dimensional time courses from simulated calcium imaging data. CILDS also outperformed the other methods on calcium imaging recordings from larval zebrafish and mice. More broadly, this study represents a foundation for summarizing calcium imaging recordings of large neuronal populations using dimensionality reduction in diverse experimental settings.