Research map / Probabilistic and scientific modeling / Dynamical systems
Bayesian Bi-clustering of Neural Spiking Activity with Latent Structures
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a nonparametric Bayesian bi-clustering model that clusters neurons into subject (population) clusters via mixture of finite mixtures over negative-binomial latent-trajectory models and clusters time into states via a sticky HDP-HMM with shared synchronous linear dynamics, fit by MCMC with Polya-Gamma augmentation.
Problem
Clustering multi-region neural spike counts both spatially (neuron populations) and temporally (experimental states) according to low-dimensional latent dynamics, when populations are unknown and over-dispersion and time-varying dynamics exist.
Approach
Jointly models neuron clusters as private latent trajectories (MFMM prior) and temporal states as synchronous local linear dynamics shared across all clusters (sticky HDP-HMM prior), with an NB observation model and an MCMC sampler using Polya-Gamma augmentation and marginalized likelihood for cluster assignments.
Key results
- On simulated data (3 subject clusters, 2 true states), the NB bi-clustering model recovers the number of subject and state clusters and true latent trajectories after rotation alignment.
- On Neuropixels data, NB bi-clustering finds about 10 subject clusters with anatomy mismatches (CA1 with VISp, LP with VISp) and about 13 state clusters, while Poisson and mixDPFA give noisier clusters.
- The paper reports qualitative results only; no quantitative accuracy comparison against baselines is given.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use NB observations instead of Poisson observations | bi-clustering on the simulated dataset · Simulation | latent-trajectory estimation bias | Helped |
| Use Poisson observations instead of NB observations | bi-clustering on the simulated dataset · Simulation | recovery of clustering structure | No clear effect |
| Use Poisson observations instead of NB observations | bi-clustering on Neuropixels data · Allen Visual Coding Neuropixels, session 719161530 | agreement of detected clusters with anatomical sites | Hurt |
| Ignore temporal state changes and use mixDPFA | neural population clustering · Allen Visual Coding Neuropixels, session 719161530 | separation of VISp and LP clusters | Hurt |
| Assume static dynamics rather than time-varying dynamics | neural population clustering · Allen Visual Coding Neuropixels, session 719161530 | unbiased clustering structure estimation | Hurt |
| Simultaneously model subject and state clustering structures | bi-clustering method · Simulation | latent-structure estimation | Helped |
Surprising findings
- Poisson bi-clustering and mixDPFA produce more mismatches with anatomy, suggesting that ignoring over-dispersion and time-varying dynamics yields spurious interactions.
Limitations
- Latent dimension p and sticky parameter m are fixed rather than inferred; real-data analysis uses p=2 and m=10 only for illustration.
- Sticky HDP-HMM restricts state durations to geometric distributions.
- Evaluation is largely qualitative on a single real session with no quantitative baseline comparison.
Experimental setup
Datasets: Allen Institute Visual Coding Neuropixels (session 719161530) simulated spiking counts
Benchmarks: simulated bi-clustering recovery Allen Neuropixels session 719161530
Metrics: posterior similarity matrix posterior histogram of cluster numbers maxPEAR latent trajectory recovery (HPD intervals) Geweke diagnostics
Baselines: Poisson bi-clustering mixDPFA
Builds on: mixPLDS mixDPFA sticky HDP-HMM mixture of finite mixtures Polya-Gamma augmentation
Closest papers by meaning
- Bayesian Clustering of Neural Spiking Activity Using a Mixture of Dynamic Poisson Factor Analyzers (NeurIPS 2022 · similarity 0.92)
- Latent Diffusion for Neural Spiking Data (NeurIPS 2024 · similarity 0.86)
- Modeling Latent Neural Dynamics with Gaussian Process Switching Linear Dynamical Systems (NeurIPS 2024 · similarity 0.85)
- From Cortical Synchronous Rhythm to Brain Inspired Learning Mechanism: An Oscillatory Spiking Neural Network with Time-Delayed Coordination (NeurIPS 2026 · similarity 0.85)
- CREIMBO: Cross-Regional Ensemble Interactions in Multi-view Brain Observations (ICLR 2025 · similarity 0.84)
- Multi-Region Markovian Gaussian Process: An Efficient Method to Discover Directional Communications Across Multiple Brain Regions (ICML 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.