Research map / Probabilistic and scientific modeling / Dynamical systems
Learning dynamic representations of the functional connectome in neurobiological networks
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.
Learns dynamic functional connectomes in C. elegans by computing pairwise differential affinities between calcium traces, factorizing them with non-negative tensor factorization, and applying nested weighted SBM community detection, with an experimental AWB silencing validation.
Problem
Static synaptic connectomes do not capture how neuronal communities change dynamically with behavior and stimuli; need to infer time-varying functional communities from brain-wide calcium imaging across multiple animals.
Approach
Organize pairwise non-linear differential affinities between neuron activity traces (rather than neurons) in a time x worm x pairwise-affinity tensor, factorize with NTF, and run a weighted nested SBM on each affinity factor to reveal transient functional communities.
Key results
- NWSBM achieved the highest NMI on 6 of 9 weighted LFR network types in Table 1 (e.g., 0.65 on Net 4 vs 0.47 for AS).
- Silencing AWB increased salt (160 mM NaCl) avoidance by a mean of +25%, p = 5.7e-11, with N=12 vs 12 samples of 25-50 worms.
- Tensor rank R=15 was chosen based on stability of reconstruction error across 15 random initializations.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use NWSBM instead of AS | Weighted-LFR community detection · Weighted-LFR Net 1 | NMI: 0.18 -> 0.28 | Helped |
| Use NWSBM instead of Louvain | Weighted-LFR community detection · Weighted-LFR Net 2 | NMI: 0.19 -> 0.33 | Helped |
| Use NWSBM instead of NNSED | Weighted-LFR community detection · Weighted-LFR Net 3 | NMI: 0.38 -> 0.39 | Helped |
| Use NWSBM instead of AS | Weighted-LFR community detection · Weighted-LFR Net 4 | NMI: 0.47 -> 0.65 | Helped |
| Use NWSBM instead of NNSED | Weighted-LFR community detection · Weighted-LFR Net 5 | NMI: 0.39 -> 0.39 | No clear effect |
| Use NWSBM instead of NNSED | Weighted-LFR community detection · Weighted-LFR Net 6 | NMI: 0.50 -> 0.54 | Helped |
| Use NWSBM instead of AS | Weighted-LFR community detection · Weighted-LFR Net 7 | NMI: 0.28 -> 0.25 | Hurt |
| Use NWSBM instead of Louvain | Weighted-LFR community detection · Weighted-LFR Net 8 | NMI: 1.0 -> 0.51 | Hurt |
| Use NWSBM instead of NNSED | Weighted-LFR community detection · Weighted-LFR Net 9 | NMI: 0.45 -> 0.64 | Helped |
| Silence AWB | C. elegans salt-response circuit · 160 mM NaCl drop test | salt avoidance: +25% | Helped |
| Vary the number of tensor components, selecting R=15 (ablation) | Dynamic connectome NTF · Tensor reconstruction | reconstruction-error variability | Mixed |
Surprising findings
- Silencing AWB, an aversive olfactory neuron, increased rather than decreased salt avoidance, suggesting oppositional interactions between salt sensation and olfactory aversion circuits.
Limitations
- Results and validation are specific to C. elegans sensory and inter-neurons; generalization to other organisms or domains is untested.
- Only one neuron (AWB) was experimentally validated.
- Affinities are computed from absolute derivatives, so opposite-sign coupling is treated as interaction, which may conflate inhibitory and excitatory relations.
Experimental setup
Datasets: C. elegans brain-wide calcium imaging dataset (189 sensory and inter-neurons, 21 worms) synthetic weighted LFR networks (9 types, N=125)
Benchmarks: weighted LFR benchmark (9 network types) AWB silencing salt avoidance drop test
Metrics: normalized mutual information p-value effect size
Baselines: Louvain Combo Asymptotic surprise NNSED GNNS
Builds on: hierarchical alternating least squares NTF nested weighted stochastic block model canonical polyadic decomposition
Closest papers by meaning
- Generating Hypotheses of Dynamic Causal Graphs in Neuroscience: Leveraging Generative Factor Models of Observed Time Series (ICML 2025 · similarity 0.89)
- CREIMBO: Cross-Regional Ensemble Interactions in Multi-view Brain Observations (ICLR 2025 · similarity 0.87)
- BLEND: Behavior-guided Neural Population Dynamics Modeling via Privileged Knowledge Distillation (ICLR 2025 · similarity 0.87)
- Inpainting the Neural Picture: Inferring Unrecorded Brain Area Dynamics from Multi-Animal Datasets (NeurIPS 2025 · similarity 0.86)
- $\textit{NeuroPath}$: A Neural Pathway Transformer for Joining the Dots of Human Connectomes (NeurIPS 2024 · similarity 0.86)
- Disentangling the Roles of Distinct Cell Classes with Cell-Type Dynamical Systems (NeurIPS 2024 · similarity 0.86)
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.