Simulation-based inference for efficient identification of generative models in computational connectomics.

Recent advances in connectomics research enable the acquisition of increasing amounts of data about the connectivity patterns of neurons. How can we use this wealth of data to efficiently derive and test hypotheses about the principles underlying these patterns? A common approach is to simulate neuronal networks using a hypothesized wiring rule in a generative model and to compare the resulting synthetic data with empirical data. However, most wiring rules have at least some free parameters, and identifying parameters that reproduce empirical data can be challenging as it often requires manual parameter tuning. Here, we propose to use simulation-based Bayesian inference (SBI) to address this challenge. Rather than optimizing a fixed wiring rule to fit the empirical data, SBI considers many parametrizations of a rule and performs Bayesian inference to identify the parameters that are compatible with the data. It uses simulated data from multiple candidate wiring rule parameters and relies on machine learning methods to estimate a probability distribution (the 'posterior distribution over parameters conditioned on the data') that characterizes all data-compatible parameters. We demonstrate how to apply SBI in computational connectomics by inferring the parameters of wiring rules in an in silico model of the rat barrel cortex, given in vivo connectivity measurements. SBI identifies a wide range of wiring rule parameters that reproduce the measurements. We show how access to the posterior distribution over all data-compatible parameters allows us to analyze their relationship, revealing biologically plausible parameter interactions and enabling experimentally testable predictions. We further show how SBI can be applied to wiring rules at different spatial scales to quantitatively rule out invalid wiring hypotheses. Our approach is applicable to a wide range of generative models used in connectomics, providing a quantitative and efficient way to constrain model parameters with empirical connectivity data.

Flexible and efficient simulation-based inference for models of decision-making.

Inferring parameters of computational models that capture experimental data are a central task in cognitive neuroscience. Bayesian statistical inference methods usually require the ability to evaluate the likelihood of the model-however, for many models of interest in cognitive neuroscience, the associated likelihoods cannot be computed efficiently. Simulation-based inference (SBI) offers a solution to this problem by only requiring access to simulations produced by the model. Previously, Fengler et al. introduced likelihood approximation networks (LANs, Fengler et al., 2021) which make it possible to apply SBI to models of decision-making, but require billions of simulations for training. Here, we provide a new SBI method that is substantially more simulation efficient. Our approach, mixed neural likelihood estimation (MNLE), trains neural density estimators on model simulations to emulate the simulator, and is designed to capture both the continuous (e.g., reaction times) and discrete (choices) data of decision-making models. The likelihoods of the emulator can then be used to perform Bayesian parameter inference on experimental data using standard approximate inference methods like Markov Chain Monte Carlo sampling. We demonstrate MNLE on two variants of the drift-diffusion model and show that it is substantially more efficient than LANs: MNLE achieves similar likelihood accuracy with six orders of magnitude fewer training simulations, and is significantly more accurate than LANs when both are trained with the same budget. Our approach enables researchers to perform SBI on custom-tailored models of decision-making, leading to fast iteration of model design for scientific discovery.

NitroSynapsin ameliorates hypersynchronous neural network activity in Alzheimer hiPSC models.

Beginning at early stages, human Alzheimer's disease (AD) brains manifest hyperexcitability, contributing to subsequent extensive synapse loss, which has been linked to cognitive dysfunction. No current therapy for AD is disease-modifying. Part of the problem with AD drug discovery is that transgenic mouse models have been poor predictors of potential human treatment. While it is undoubtedly important to test drugs in these animal models, additional evidence for drug efficacy in a human context might improve our chances of success. Accordingly, in order to test drugs in a human context, we have developed a platform of physiological assays using patch-clamp electrophysiology, calcium imaging, and multielectrode array (MEA) experiments on human (h)iPSC-derived 2D cortical neuronal cultures and 3D cerebral organoids. We compare hiPSCs bearing familial AD mutations vs. their wild-type (WT) isogenic controls in order to characterize the aberrant electrical activity in such a human context. Here, we show that these AD neuronal cultures and organoids manifest increased spontaneous action potentials, slow oscillatory events (~1 Hz), and hypersynchronous network activity. Importantly, the dual-allosteric NMDAR antagonist NitroSynapsin, but not the FDA-approved drug memantine, abrogated this hyperactivity. We propose a novel model of synaptic plasticity in which aberrant neural networks are rebalanced by NitroSynapsin. We propose that hiPSC models may be useful for screening drugs to treat hyperexcitability and related synaptic damage in AD.

Neuronal timescales are functionally dynamic and shaped by cortical microarchitecture.

Complex cognitive functions such as working memory and decision-making require information maintenance over seconds to years, from transient sensory stimuli to long-term contextual cues. While theoretical accounts predict the emergence of a corresponding hierarchy of neuronal timescales, direct electrophysiological evidence across the human cortex is lacking. Here, we infer neuronal timescales from invasive intracranial recordings. Timescales increase along the principal sensorimotor-to-association axis across the entire human cortex, and scale with single-unit timescales within macaques. Cortex-wide transcriptomic analysis shows direct alignment between timescales and expression of excitation- and inhibition-related genes, as well as genes specific to voltage-gated transmembrane ion transporters. Finally, neuronal timescales are functionally dynamic: prefrontal cortex timescales expand during working memory maintenance and predict individual performance, while cortex-wide timescales compress with aging. Thus, neuronal timescales follow cytoarchitectonic gradients across the human cortex and are relevant for cognition in both short and long terms, bridging microcircuit physiology with macroscale dynamics and behavior.

The human brain can both quickly react to a fleeting sight, like a changing traffic light, and slowly integrate complex information to form a long-term plan. To mirror these requirements, how long a neuron can be activated for ­ its 'timescale' ­ varies greatly between cells. A range of timescales has been identified in animal brains, by measuring single neurons at a few different locations. However, a comprehensive study of this property in humans has been hindered by technical and ethical concerns. Without this knowledge, it is difficult to understand the factors that may shape different timescales, and how these can change in response to environmental demands. To investigate this question, Gao et al. used a new computational method to analyse publicly available datasets and calculate neuronal timescales across the human brain. The data were produced using a technique called invasive electrocorticography, where electrodes placed directly on the brain record the total activity of many neurons. This allowed Gao et al. to examine the relationship between timescales and brain anatomy, gene expression, and cognition. The analysis revealed a continuous gradient of neuronal timescales between areas that require neurons to react quickly and those relying on long-term activity. 'Under the hood', these timescales were associated with a number of biological processes, such as the activity of genes that shape the nature of the connections between neurons and the amount of proteins that let different charged particles in and out of cells. In addition, the timescales could be flexible: they could lengthen when areas specialised in working memory were actively maintaining information, or shorten with age across many areas of the brain. Ultimately, the technique and findings reported by Gao et al. could have useful applications in the clinic, using neuronal timescale to better understand brain disorders and pinpoint their underlying causes.

Parameterizing neural power spectra into periodic and aperiodic components.

Electrophysiological signals exhibit both periodic and aperiodic properties. Periodic oscillations have been linked to numerous physiological, cognitive, behavioral and disease states. Emerging evidence demonstrates that the aperiodic component has putative physiological interpretations and that it dynamically changes with age, task demands and cognitive states. Electrophysiological neural activity is typically analyzed using canonically defined frequency bands, without consideration of the aperiodic (1/f-like) component. We show that standard analytic approaches can conflate periodic parameters (center frequency, power, bandwidth) with aperiodic ones (offset, exponent), compromising physiological interpretations. To overcome these limitations, we introduce an algorithm to parameterize neural power spectra as a combination of an aperiodic component and putative periodic oscillatory peaks. This algorithm requires no a priori specification of frequency bands. We validate this algorithm on simulated data, and demonstrate how it can be used in applications ranging from analyzing age-related changes in working memory to large-scale data exploration and analysis.

For the Sciences They Are A-Changin': A Response to Commentaries on Núñez et al.'s (2019) "What Happened to Cognitive Science?"

A recent issue of Topics in Cognitive Science featured 11 thoughtful commentaries responding to our article "What happened to cognitive science?" (Núñez et al., 2019). Here, we identify several themes that arose in those commentaries and respond to each. Crucial to understanding our original article is the fundamental distinction between multidisciplinary and interdisciplinary endeavors: Cognitive science began (and has stayed) as multidisciplinary but has failed to move on to form a cohesive interdisciplinary field. We clarify and elaborate our original argument and reiterate the importance of a data-driven evaluation of the current status of the field, which exhibits a marked disciplinary imbalance, a lack of a coherent conceptual core, and a striking absence of a consistent curriculum in the institutions that grant degrees in this domain. Half a century after the creation of cognitive science, it may now be a good time to revisit goals and visions for how to best approach the ever-fascinating scientific study of the mind(s).

Complex Oscillatory Waves Emerging from Cortical Organoids Model Early Human Brain Network Development.

Structural and transcriptional changes during early brain maturation follow fixed developmental programs defined by genetics. However, whether this is true for functional network activity remains unknown, primarily due to experimental inaccessibility of the initial stages of the living human brain. Here, we developed human cortical organoids that dynamically change cellular populations during maturation and exhibited consistent increases in electrical activity over the span of several months. The spontaneous network formation displayed periodic and regular oscillatory events that were dependent on glutamatergic and GABAergic signaling. The oscillatory activity transitioned to more spatiotemporally irregular patterns, and synchronous network events resembled features similar to those observed in preterm human electroencephalography. These results show that the development of structured network activity in a human neocortex model may follow stable genetic programming. Our approach provides opportunities for investigating and manipulating the role of network activity in the developing human cortex.

What happened to cognitive science?

More than a half-century ago, the 'cognitive revolution', with the influential tenet 'cognition is computation', launched the investigation of the mind through a multidisciplinary endeavour called cognitive science. Despite significant diversity of views regarding its definition and intended scope, this new science, explicitly named in the singular, was meant to have a cohesive subject matter, complementary methods and integrated theories. Multiple signs, however, suggest that over time the prospect of an integrated cohesive science has not materialized. Here we investigate the status of the field in a data-informed manner, focusing on four indicators, two bibliometric and two socio-institutional. These indicators consistently show that the devised multi-disciplinary program failed to transition to a mature inter-disciplinary coherent field. Bibliometrically, the field has been largely subsumed by (cognitive) psychology, and educationally, it exhibits a striking lack of curricular consensus, raising questions about the future of the cognitive science enterprise.

Setd5 haploinsufficiency alters neuronal network connectivity and leads to autistic-like behaviors in mice.

SETD5, a gene linked to intellectual disability (ID) and autism spectrum disorder (ASD), is a member of the SET-domain family and encodes a putative histone methyltransferase (HMT). To date, the mechanism by which SETD5 haploinsufficiency causes ASD/ID remains an unanswered question. Setd5 is the highly conserved mouse homolog, and although the Setd5 null mouse is embryonic lethal, the heterozygote is viable. Morphological tracing and multielectrode array was used on cultured cortical neurons. MRI was conducted of adult mouse brains and immunohistochemistry of juvenile mouse brains. RNA-Seq was used to investigate gene expression in the developing cortex. Behavioral assays were conducted on adult mice. Setd5+/- cortical neurons displayed significantly reduced synaptic density and neuritic outgrowth in vitro, with corresponding decreases in network activity and synchrony by electrophysiology. A specific subpopulation of fetal Setd5+/- cortical neurons showed altered gene expression of neurodevelopment-related genes. Setd5+/- animals manifested several autism-like behaviors, including hyperactivity, cognitive deficit, and altered social interactions. Anatomical differences were observed in Setd5+/- adult brains, accompanied by a deficit of deep-layer cortical neurons in the developing brain. Our data converge on a picture of abnormal neurodevelopment driven by Setd5 haploinsufficiency, consistent with a highly penetrant risk factor.

Inferring synaptic excitation/inhibition balance from field potentials.

Neural circuits sit in a dynamic balance between excitation (E) and inhibition (I). Fluctuations in E:I balance have been shown to influence neural computation, working memory, and information flow, while more drastic shifts and aberrant E:I patterns are implicated in numerous neurological and psychiatric disorders. Current methods for measuring E:I dynamics require invasive procedures that are difficult to perform in behaving animals, and nearly impossible in humans. This has limited the ability to examine the full impact that E:I shifts have in cognition and disease. In this study, we develop a computational model to show that E:I changes can be estimated from the power law exponent (slope) of the electrophysiological power spectrum. Predictions from the model are validated in published data from two species (rats and macaques). We find that reducing E:I ratio via the administration of general anesthetic in macaques results in steeper power spectra, tracking conscious state over time. This causal result is supported by inference from known anatomical E:I changes across the depth of rat hippocampus, as well as oscillatory theta-modulated dynamic shifts in E:I. Our results provide evidence that E:I ratio may be inferred from electrophysiological recordings at many spatial scales, ranging from the local field potential to surface electrocorticography. This simple method for estimating E:I ratio-one that can be applied retrospectively to existing data-removes a major hurdle in understanding a currently difficult to measure, yet fundamental, aspect of neural computation.

Interpreting the electrophysiological power spectrum.

Recent experimental findings suggest that there may be rich physiological information embedded within the power spectrum of neurophysiological recordings, which, in addition to power in specific oscillatory frequencies, can be extracted with the appropriate model. This article reviews previous empirical and modeling results, as well as the canonical power law model that is often used to describe the power spectrum. In addition, a modified power law model with parameters estimating synaptic and spiking contributions is proposed.