Biology subjects

Kim, J. Z.

Publications and source records attributed to Kim, J. Z..

14 recordsLinked to original sources

bioRxiv2026

Hypothesis-Driven Identification of Neural Algorithms With Dynamical Structure-Preserving Manifolds.

Neuroscience aims to uncover the algorithms by which the brain builds and manipulates rich internal representations of objects, agents, and places. Existing reverse-engineering frameworks remain ineffective for identifying these algorithms. Here, we present Dynamical Structure-Preserving Manifolds (dSPMs)---a novel multilevel reverse-engineering framework that programs, without training, recurrent neural networks to enable hypothesis-driven identification of the brain's algorithms that generate symbolic, structure-preserving representations of the world around. We apply dSPM to identify an algorithm for physical scene prediction in the dorsomedial frontal cortex (DMFC) of macaques intercepting a ball on a Pong-like virtual gameboard. dSPM better explained DMFC than alternatives, made predictions confirmed in neural data, and revealed an algorithm of DMFC that we term 'manifold-based simulation': perceived initial conditions rapidly collapse neural activity onto a low-dimensional manifold whose geometry encodes a physics-based representation of the scene, supporting far-out predictions without traditional next-step simulation. dSPM supports effective exploration of the computational foundations of biological intelligence.

neuroscience↗

bioRxiv2026

Geometric constraints and cognitive inputs jointly shape emergent brain dynamics and topology

How the brain's physical geometry gives rise to its flexible functional repertoire remains a central question in neuroscience. Here, we trained three classes of recurrent neural networks (RNNs) on a working-memory task, forming a graded hierarchy of spatial constraints: Vanilla RNNs (no spatial constraints), Masked RNNs (projection constraints limiting where information enters and leaves the network), and biophysical RNNs (bioRNNs; projection constraints and spatial embedding of the networks' connectivity using the brain's inter-regional Euclidean geometry). We assessed how well each RNN class predicted empirical fMRI activity without exposing them to it during training. Our results showed that bioRNNs were the only networks to successfully predict empirical brain activity and to organize their dynamics into a spatial pattern that recapitulated the brain's principal hierarchy (the sensorimotor-association axis). Additionally, bioRNNs' ability to predict empirical brain activity emerged along a trajectory in which geometry was laid down first, then partly traded back as the task was mastered. Importantly, brain-like topological features emerged in bioRNNs as they increased their task proficiency while maintaining their ability to predict brain activity. Taken together, our results indicate that physical geometry and cognitive inputs play distinct, complementary roles: while geometry constrains the space of possible brain dynamics, cognitive inputs determine which dynamics are expressed. They also situate topology as the scaffold through which the physically embedded brain reconciles wiring costs and computational demands.

neuroscience↗

bioRxiv2026

The hippocampus as a small-world cognitive map

When a mouse perceives a hawks shadow, it may have only seconds to decide where to run, yet the safest refuge is often neither visible nor nearby. To survive, it must search its cognitive map quickly enough to choose among many possibilities, and accurately enough to avoid dead ends and hazards along the way, all while what counts as "safe" changes as paths, refuges, and threats shift with time. This scenario highlights a core design problem: cognitive maps must preserve fine local structure for reliable action, yet remain globally searchable so that distant, useful solutions can be found efficiently in both space and time. The hippocampus is thought to support such maps, with population activity representing world states and their transitions--yet how these maps are structured to solve this design problem is not well understood. Here, we used a novel geometry-aware autoencoder to model the structure of the cognitive map from longitudinal calcium imaging from thousands of hippocampal CA1 neurons in mice learning a memory-guided navigation task. We discovered that the hippocampal population code achieves both local fidelity and global searchability through small-world network structure in the space of neural representations that leverages two complementary mechanisms. At the population level, helical (rotation-plus-drift) dynamics of neural representations relative to past experience build new maps that preserve local information about nearby positions in space and time while remaining distinguishable from earlier representations. At the cellular level, neurons with coordinated multi-field activity create sparse, long-range shortcuts between distant representations. During synchronous population events in immobility, decoded activity often jumps to distant states in space, time, and task conditions, suggesting these shortcuts are engaged during offline processing. This functional organization, with implications for both neuroscience and artificial intelligence, sheds light on how hippocampal representations may be optimized for a fundamental challenge faced by intelligent systems: efficiently searching through accurate internal models of the world.

neuroscience↗

bioRxiv2025

Inferring macroscopic intrinsic neural timescales using optimal control theory

The temporal evolution of brain activity relies on complex interactions within and between brain regions that are mediated by neurobiology and connectivity. To understand these interactions, many large-scale efforts have measured structural connectivity, neural activity, gene expression, and cognition across multiple modalities and species. However, data-driven discovery of large-scale activity models remains difficult owing to the lack of flexible quantitative frameworks for estimating the interplay between brain structure and function while preserving biophysical realism. Here, we provide such a framework by integrating network control theory (NCT) with automatic differentiation to estimate model parameters with greater biophysical realism from data. Specifically, we estimate the structural form of regional self-inhibition--a quantity that is experimentally difficult to measure--from MRI data. Next, we demonstrate that the resulting model-based self-inhibition parameters correlate significantly with regions intrinsic neural timescales (INTs), neurobiological measures of gene expression and cell-type densities, as well as behavioral measures of cognition. We demonstrate consistent results across multiple datasets and species. Finally, we demonstrate that our self-inhibition parameters enable the efficient control of brain dynamics from fewer brain regions. Taken together, our results provide a simple and flexible quantitative framework that more accurately captures the interplay between brain structure, function, and dynamics with greater biophysical realism.

neuroscience↗

bioRxiv2025

Linking neural population formatting to function

Animals capable of complex behaviors tend to have more distinct brain areas than simpler organisms, and artificial networks that perform many tasks tend to self-organize into modules (1-3). This suggests that different brain areas serve distinct functions supporting complex behavior. However, a common observation is that essentially anything that an animal senses, knows, or does can be decoded from neural activity in any brain area (4-6). If everything is everywhere, why have distinct areas? Here we show that the function of a brain area is more related to how different types of information are combined (formatted) in neural representations than merely whether that information is present. We compared two brain areas: the middle temporal area (MT), which is important for visual motion perception (7, 8), and the dorsolateral prefrontal cortex (dlPFC), which is linked to decision-making and reward expectation (9, 10)). When monkeys based decisions on a combination of motion and reward information, both types of information were present in both areas. However, they were formatted differently: in MT, they were encoded separably, while in dlPFC, they were represented jointly in ways that reflected the monkeys decision-making. A recurrent neural network (RNN) model that mirrored the information formatting in MT and dlPFC predicted that manipulating activity in these areas would differently affect decision-making. Consistent with model predictions, electrically stimulating MT biased choices midway between the visual motion stimulus and the preferred direction of the stimulated units (11), while stimulating dlPFC produced winner-take-all decisions that sometimes reflected the visual motion stimulus and sometimes reflected the preference of the stimulated units, but never in between. These results are consistent with the tantalizing possibility that a modular structure enables complex behavior by flexibly reformatting information to accomplish behavioral goals.

neuroscience↗

bioRxiv2023

Using network control theory to study the dynamics of the structural connectome

Network control theory (NCT) is a simple and powerful tool for studying how network topology informs and constrains dynamics. Compared to other structure-function coupling approaches, the strength of NCT lies in its capacity to predict the patterns of external control signals that may alter dynamics in a desired way. We have extensively developed and validated the application of NCT to the human structural connectome. Through these efforts, we have studied (i) how different aspects of connectome topology affect neural dynamics, (ii) whether NCT outputs cohere with empirical data on brain function and stimulation, and (iii) how NCT outputs vary across development and correlate with behavior and mental health symptoms. In this protocol, we introduce a framework for applying NCT to structural connectomes following two main pathways. Our primary pathway focuses on computing the control energy associated with transitioning between specific neural activity states. Our second pathway focuses on computing average controllability, which indexes nodes general capacity to control dynamics. We also provide recommendations for comparing NCT outputs against null network models. Finally, we support this protocol with a Python-based software package called network control theory for python (nctpy).

neuroscience↗

bioRxiv2022

Compression supports low-dimensional representations of behavior across neural circuits

Dimensionality reduction, a form of compression, can simplify representations of information to increase efficiency and reveal general patterns. Yet, this simplification also forfeits information, thereby reducing representational capacity. Hence, the brain may benefit from generating both compressed and uncompressed activity, and may do so in a heterogeneous manner across diverse neural circuits that represent low-level (sensory) or high-level (cognitive) stimuli. However, precisely how compression and representational capacity differ across the cortex remains unknown. Here we predict different levels of compression across regional circuits by using random walks on networks to model activity flow and to formulate rate-distortion functions, which are the basis of lossy compression. Using a large sample of youth (n = 1, 040), we test predictions in two ways: by measuring the dimensionality of spontaneous activity from sensorimotor to association cortex, and by assessing the representational capacity for 24 behaviors in neural circuits and 20 cognitive variables in recurrent neural networks. Our network theory of compression predicts the dimensionality of activity (t = 12.13, p < 0.001) and the representational capacity of biological (r = 0.53, p = 0.016) and artificial (r = 0.61, p < 0.001) networks. The model suggests how a basic form of compression is an emergent property of activity flow between distributed circuits that communicate with the rest of the network.

neuroscience↗

bioRxiv2022

Asymmetric Signaling Across the Hierarchy of Cytoarchitecture within the Human Connectome

Cortical variations in cytoarchitecture form a sensory-fugal axis that systematically shapes regional profiles of extrinsic connectivity. Additionally, this axis is thought to guide signal propagation and integration across the cortical hierarchy. While human neuroimaging work has shown that this axis constrains local properties of the human connectome, it remains unclear whether it also shapes the asymmetric signaling that arises from higher-order connectome topology. Here, we used network control theory to examine the amount of energy required to propagate dynamics across the sensory-fugal axis. Our results revealed an asymmetry in this energy indicating that bottom-up transitions were easier to complete compared to top-down transitions. Supporting analyses demonstrated that this asymmetry was underpinned by a connectome topology that is wired to support efficient bottom-up signaling. Finally, we found that this asymmetry correlated with changes in intrinsic neuronal timescales and lessened throughout youth. Our results show that cortical variation in cytoarchitecture may guide the formation of macroscopic connectome topology.

neuroscience↗

bioRxiv2021

Pathological and metabolic underpinnings of energetic inefficiency in temporal lobe epilepsy

The human brain consumes a disproportionate amount of energy to generate neural dynamics. Yet precisely how energetic processes are altered in neurological disorders remains far from understood. Here, we use network control theory to profile the brains energy landscape, describing the rich dynamical repertoire supported by the structural connectome. This approach allows us to estimate the energy required to activate a circuit, and determine which regions most support that activation. Focusing on temporal lobe epilepsy (TLE), we show that patients require more control energy to activate the limbic network than healthy volunteers, especially ipsilateral to the seizure focus. Further, greater energetic costs are largely localized to the ipsilateral temporo-limbic regions. Importantly, the energetic imbalance between ipsilateral and contralateral temporo-limbic regions is tracked by asymmetric metabolic patterns, which in turn are explained by asymmetric gray matter volume loss. In TLE, failure to meet the extra energy demands may lead to suboptimal brain dynamics and inadequate activation. Broadly, our investigation provides a theoretical framework unifying gray matter integrity, local metabolism, and energetic generation of neural dynamics.

neuroscience↗

bioRxiv2021

External drivers of BOLD signals non-stationarity

A fundamental challenge in neuroscience is to uncover the principles governing how the brain interacts with the external environment. However, assumptions about external stimuli fundamentally constrain current computational models. We show in silico that unknown external stimulation can produce error in the estimated linear time-invariant dynamical system. To address these limitations, we propose an approach to retrieve the external (unknown) input parameters and demonstrate that the estimated system parameters during external input quiescence uncover spatiotemporal profiles of external inputs over external stimulation periods more accurately. Finally, we unveil the expected (and unexpected) sensory and task-related extra-cortical input profiles using functional magnetic resonance imaging data acquired from 96 subjects (Human Connectome Project) during the resting-state and task scans. Together, we provide evidence that this embodied brain activity model offers information about the structure and dimensionality of the BOLD signals external drivers and shines light on likely external sources contributing to the BOLD signals non-stationarity.

neuroscience↗

bioRxiv2019

A dynamical systems framework to uncover the drivers of large-scale cortical activity

A fundamental challenge in neuroscience is to uncover the principles governing complex interactions between the brain and its external environment. Over the past few decades, the development of functional neuroimaging techniques and tools from graph theory, network science, and computational neuroscience have markedly expanded opportunities to study the intrinsic organization of brain activity. However, many current computational models are fundamentally limited by little to no explicit assessment of the brains interactions with external stimuli. To address this limitation, we propose a simple scheme that jointly estimates the intrinsic organization of brain activity and extrinsic stimuli. Specifically, we adopt a linear dynamical model (intrinsic activity) under unknown exogenous inputs (e.g., sensory stimuli), and jointly estimate the model parameters and exogenous inputs. First, we demonstrate the utility of this scheme by accurately estimating unknown external stimuli in a synthetic example. Next, we examine brain activity at rest and task for 99 subjects from the Human Connectome Project, and find significant task-related changes in the identified system, and task-related increases in the estimated external inputs showing high similarity to known task regressors. Finally, through detailed examination of fluctuations in the spatial distribution of the oscillatory modes of the estimated system during the resting state, we find an apparent non-stationarity in the profile of modes that span several brain regions including the visual and the dorsal attention systems. The results suggest that these brain structures display a time-varying relationship, or alternatively, receive non-stationary exogenous inputs that can lead to apparent system non-stationarities. Together, our embodied model of brain activity provides an avenue to gain deeper insight into the relationship between cortical functional dynamics and their drivers.

neuroscience↗

bioRxiv2018

Network topology of neural systems supporting avalanche dynamics predicts stimulus propagation and recovery

Many neural systems display avalanche behavior characterized by uninterrupted sequences of neuronal firing whose distributions of size and durations are heavy-tailed. Theoretical models of such systems suggest that these dynamics support optimal information transmission and storage. However, the unknown role of network structure precludes an understanding of how variations in network topology manifest in neural dynamics and either support or impinge upon information processing. Here, using a generalized spiking model, we develop a mechanistic understanding of how network topology supports information processing through network dynamics. First, we show how network topology determines network dynamics by analytically and numerically demonstrating that network topology can be designed to propagate stimulus patterns for long durations. We then identify strongly connected cycles as empirically observable network motifs that are prevalent in such networks. Next, we show that within a network, mathematical intuitions from network control theory are tightly linked with dynamics initiated by node-specific stimulation and can identify stimuli that promote long-lasting cascades. Finally, we use these network-based metrics and control-based stimuli to demonstrate that long-lasting cascade dynamics facilitate delayed recovery of stimulus patterns from network activity, as measured by mutual information. Collectively, our results provide evidence that cortical networks are structured with architectural motifs that support long-lasting propagation and recovery of a few crucial patterns of stimulation, especially those consisting of activity in highly controllable neurons. Broadly, our results imply that avalanching neural networks could contribute to cognitive faculties that require persistent activation of neuronal patterns, such as working memory or attention.

neuroscience↗

bioRxiv2018

Optimization of Energy State Transition Trajectory Supports the Development of Executive Function During Youth

Executive function develops rapidly during adolescence, and failures of executive function are associated with both risk-taking behaviors and psychopathology. However, it remains relatively unknown how structural brain networks mature during this critical period to facilitate energetically demanding transitions to activate the frontoparietal system, which is critical for executive function. In a sample of 946 human youths (ages 8-23 yr) who completed diffusion imaging as part of the Philadelphia Neurodevelopment Cohort, we capitalized upon recent advances in network control theory in order to calculate the control energy necessary to activate the frontoparietal system given the existing structural network topology. We found that the control energy required to activate the frontoparietal system declined with development. Moreover, we found that this control energy pattern contains sufficient information to make accurate predictions about individuals brain maturity. Finally, the control energy costs of the cingulate cortex were negatively correlated with executive performance, and partially mediated the development of executive performance with age. These results could not be explained by changes in general network control properties or in network modularity. Taken together, our results reveal a mechanism by which structural networks develop during adolescence to facilitate the instantiation of activation states necessary for executive function.\n\nSIGNIFICANCE STATEMENTExecutive function undergoes protracted development during youth, but it is unknown how structural brain networks mature to facilitate the activation of the frontoparietal cortex that is critical for executive processes. Here, we leverage recent advances in network control theory to establish that structural brain networks evolve in adolescence to lower the energetic cost of activating the frontoparietal system. Our results suggest a new mechanistic framework for understanding how brain network maturation supports cognition, with clear implications for disorders marked by executive dysfunction, such as ADHD and psychosis.

neuroscience↗

bioRxiv2018

Context-dependent architecture of brain state dynamics is explained by white matter connectivity and theories of network control

A diverse white matter network and finely tuned neuronal membrane properties allow the brain to transition seamlessly between cognitive states. However, it remains unclear how static structural connections guide the temporal progression of large-scale brain activity patterns in different cognitive states. Here, we deploy an unsupervised machine learning algorithm to define brain states as time point level activity patterns from functional magnetic resonance imaging data acquired during passive visual fixation (rest) and an n-back working memory task. We find that brain states are composed of interdigitated functional networks and exhibit context-dependent dynamics. Using diffusion-weighted imaging acquired from the same subjects, we show that structural connectivity constrains the temporal progression of brain states. We also combine tools from network control theory with geometrically conservative null models to demonstrate that brains are wired to support states of high activity in default mode areas, while requiring relatively low energy. Finally, we show that brain state dynamics change throughout development and explain working memory performance. Overall, these results elucidate the structural underpinnings of cognitively and developmentally relevant spatiotemporal brain dynamics.

neuroscience↗