Research | Sun Lab

// 01 — focus_areas.py

Active Research Directions

// Our interdisciplinary approach combines electrophysiology, optogenetics, imaging, and computational modeling.

Prefrontal-Motor Circuits & Persistent Behavior

Investigating how dorsal medial prefrontal cortex (dmPFC) neurons projecting to motor cortex initiate and maintain persistent movement. Using single-unit extracellular recordings and opto-tagging in awake mice, we decode the contextual signals that drive continuous action.

#dmPFC #motor-cortex #opto-tagging #behavior

Long-Range Recurrent Networks & Emotion

Mapping long-range, recurrent neuronal networks that link emotion-processing regions with the somatic motor cortex. Understanding how emotional states modulate motor output through polysynaptic cortico-cortical and cortico-subcortical pathways.

#emotion-circuits #recurrent-networks #somatic-motor

Neuronal Identity Switch & Plasticity

Studying learning-induced neuronal identity switches in the superficial layers of the primary somatosensory cortex. Revealing how sensory experience reshapes the molecular and functional identity of excitatory and inhibitory neurons.

#S1-cortex #plasticity #learning #identity-switch

Epileptogenesis & Focal Cortical Dysplasia

Elucidating circuit mechanisms underlying epileptogenesis in mouse models of focal cortical malformation (FCM). Characterizing burst-suppression patterns, local field potential synchrony disruption, and spike-wave seizure dynamics across cortical layers.

#epilepsy #FCM #LFP #seizure

GABAergic Circuit Maturation

Investigating the development and experience-dependent maturation of GABAergic inhibitory interneurons in neocortical circuits. Understanding NMDA receptor NR2 subunit roles in critical period plasticity of parvalbumin and somatostatin interneurons.

#GABA #interneurons #critical-period #BDNF

Optogenetic Tool Development

Engineering novel optogenetic actuators including near-infrared activated adenylate cyclases for mammalian applications. Developing the Laserspritzer method for subcellular-resolution optogenetic investigation of synaptic integration.

#optogenetics #NIR #laserspritzer #subcellular

// 02 — ai_models.py

AI & Neural Circuit Models

// Leveraging artificial intelligence and computational modeling to decode the neural control of persistent behavior and decision-making.

_ ___ _ _ __ __ _ _ / \ |_ _|_ __ | |_ ___ __| | | \/ | ___| |_ __ | | _____ _ __ ___ / _ \ | || '_ \| __/ _ \/ _` | | |\/| |/ _ \ | '_ \| |/ / _ \ '__/ __| / ___ \ | || | | | || __/ (_| | | | | | __/ | | | | < __/ | \__ \ /_/ \_\___|_| |_|\__\___|\__,_| |_| |_|\___|_|_| |_|_|\_\___|_| |___/

We integrate cutting-edge machine learning algorithms with large-scale neural recordings to build predictive models of how neural circuits control behavior. Our AI-driven approach enables us to extract latent dynamics from high-dimensional spike train data, decode behavioral states in real time, and identify the minimal circuit motifs necessary for generating persistent motor output.

Neural Control of Persistent Behavior

Using recurrent neural networks (RNNs) and dynamical systems models fit to dmPFC→motor cortex recordings, we infer the latent state space that governs the initiation and maintenance of persistent movement. Our models reveal how contextual inputs bias the network into a self-sustained attractor state, providing a mechanistic framework for understanding why some behaviors persist while others terminate.

#RNN #attractor-dynamics #dmPFC #persistent-activity

Decision-Making for Behavior Persistence

We deploy reinforcement learning (RL) frameworks and drift-diffusion models to formalize how the brain decides whether to sustain or switch a motor action. By comparing model predictions with optogenetically perturbed neural data, we identify the specific prefrontal subpopulations that compute the "cost of persistence" versus the "value of switching," bridging computational psychiatry and systems neuroscience.

#RL #drift-diffusion #decision-making #cost-benefit

Spike Train Decoding & Population Dynamics

Applying variational autoencoders (VAEs) and Gaussian process factor analysis (GPFA) to simultaneously recorded spike trains, we uncover low-dimensional manifolds that encode behavioral intent. These AI-driven dimensionality reduction techniques allow us to predict behavioral persistence from neural population activity with high accuracy, and to map how optogenetic perturbations reshape the neural trajectory.

#VAE #GPFA #dimensionality-reduction #population-coding

Biologically Constrained Circuit Models

We construct biophysically detailed spiking neural network (SNN) models constrained by our experimental measurements of connectivity, synaptic weights, and intrinsic excitability. These models serve as "virtual labs" where we test hypotheses about E/I balance, synaptic scaling, and neuromodulatory gating — generating experimentally testable predictions about how circuit parameters control the duration and stability of persistent behavior.

#SNN #E-I-balance #synaptic-scaling #virtual-lab

Closed-Loop AI-Neurofeedback

Developing real-time decoding pipelines that use AI to read out behavioral state from neural activity and deliver precisely timed optogenetic feedback. This closed-loop paradigm allows us to causally test whether artificially maintaining or disrupting specific neural activity patterns is sufficient to prolong or terminate persistent behavior, opening new avenues for therapeutic neuromodulation.

#closed-loop #neurofeedback #real-time #neuromodulation

Cross-Species Generalization

Training transformer-based models on multi-species neural and behavioral datasets to identify conserved computational principles underlying persistent behavior. By comparing mouse dmPFC recordings with human intracranial EEG during motor tasks, we test whether the same latent dynamical motifs generalize across species — with implications for translating circuit-level findings to clinical interventions.

#transformer #cross-species #iEEG #translation

// 03 — methods.json

Techniques & Approaches

// We employ a diverse toolkit spanning molecular, cellular, circuit, and behavioral levels of analysis.

$ neurohab — Integrated Behavioral Arena

We developed the NeuroHab, an integrated behavioral arena that enables high-fidelity operant conditioning training and automated data collection in a single unified system. Operant elements such as food and water delivery as reward, conditioned stimulus, and event recording are tied together programmatically with easy-to-install open-source code to facilitate throughput and reproducibility.

// Figure 14. NeuroHab with all Mods installed. Home (left) Core (right).

$ timing_precision

All behavioral events — including food delivery, water delivery, and conditioned stimuli — are processed by internal microcontrollers and logged with <1 ms latency (typical sensor-to-log range: 56–728 μs under normal operating conditions). This precise timing is critical for integrating our system with two-photon imaging and electrophysiology equipment, enabling us to align behavior with brain activity in real time. These timestamps allow us to associate behavior with neural activity, such as calcium events and neuronal spikes, with high precision.

$ hardware_modules

The NeuroHab uses solenoid-actuated, capacitive-sensing Lickports to control lick detection and water delivery, enabling an untethered mouse to drink from an automated port similarly to a standard home-cage water bottle. For food delivery, we utilize the Kravitz Lab FED3, which detects nose pokes to automate pellet dispensing. Conditioned stimuli are delivered by dedicated modules, each containing a buzzer and an RGB LED.

$ core_architecture

All stimulus delivery is automated and recorded by the central control system, known as the Core. The NeuroHab Core coordinates all modules and records their outputs. It uses TTL pulses for communication between its two microcontrollers to orchestrate the behavioral tasks and log all event timestamps in chronological order.

#operant-conditioning #automated-behavior #open-source #low-latency #lickport #FED3

$ surgery_and_viral_injections

Employing rigorous surgical procedures and viral vector injections, we modulate neural circuits and gene expression patterns to elucidate their roles in behavior and disease progression.

Stereotaxic intracranial injection surgery

// Stereotaxic surgery and viral injection workflow

$ viral_workflow

Using advanced genetic techniques, we create mouse models targeting specific cell types and marker genes, enabling precise manipulation and observation of cellular processes.

// Brain viral expressing workflow (confocal)

$ electrophysiology

Single-unit extracellular recordings, simultaneous multiple patch-clamp, local field potential (LFP) recordings, and in vivo whole-cell recordings.

// SliceScope Pro electrophysiology rig

$ patch_clamp

Simultaneous multiple patch-clamp recording system for decoding complex neural circuits with optogenetic assistance.

// Patch Pro 1000 electrophysiology rig

$ two_photon_imaging

Multi-photon imaging and fiber photometry to capture real-time calcium activities in vivo, enabling high-resolution imaging of neural dynamics.

// Prairie Ultima IV two-photon in vivo rig

$ in_vivo_imaging

In vivo imaging setup for capturing neural activity during awake behaving experiments with head-fixed animals.

// Two-photon calcium imaging setup diagram

$ single_cell_rnaseq

Leveraging platforms like 10x Genomics, we perform single-cell RNA sequencing to dissect cellular heterogeneity and gene expression profiles within neural populations.

// 10x Genomics GEM-X scRNA-seq workflow

$ scrnaseq_pipeline

Complete pipeline from sample collection to sequencing and data analysis for high-quality single-cell profiling.

// From collection to sequencing workflow

$ behavioral_assays

Awake behaving mouse paradigms, persistent licking tasks, wheel running, sensory discrimination, and seizure monitoring during sleep.

$ circuit_tracing

AAVretro viral tracing, rabies-based monosynaptic tracing, anterograde and retrograde labeling of long-range projection neurons.

$ computational_modeling

Network simulations of E-I balance, computational models of persistent activity, and analysis of spike train dynamics and synchrony.

$ optogenetics

Channelrhodopsin-2 (ChR2) activation, opto-tagging of projection neurons, subcellular Laserspritzer stimulation, and near-infrared optogenetic tools.

$ histology_imaging

Histological techniques, immunohistochemistry, and confocal imaging to visualize cellular structures and molecular markers in tissue samples.

$ population_decoding

Decoding neural signals to infer cognitive states and behavioral patterns from population-level activity.

Research Programs