Abstract

We analyze the oviIN inputs as a subconnectome to reveal parallel input modules (multi-circuit hub). We replicate modularity findings and add PyTorch embedding/similarity analysis plus an interactive Streamlit browser.

Methods

Subconnectome construction

• Include all presynaptic partners to oviIN (right hemisphere).
• Build a weighted undirected graph (synapse counts, bilateral merged where appropriate).
• No low thresholding on first pass to avoid isolating weak-but-informative edges.

Community detection

• Classical modularity at coarse resolution, reporting 6–8 stable modules.
• Compare alignment with global connectome partitions (low Jaccard expected).
• Per-module ROI composition + strong/reciprocal pairs.

PyTorch embeddings

• Node features: degree, ROI counts, connectivity stats; optional morphology summaries.
• Train small encoder to learn similarity; evaluate within vs. between modules.

Demo

• Streamlit app to browse modules, toggle ROI overlays, and inspect strong inputs.
• Cached queries for snappy UX; packaged with Docker.

Results

• oviIN inputs partition into multiple densely intra-connected modules with distinct ROI footprints.
• Within-module input similarity significantly higher than between modules.
• Embeddings preserve module neighborhoods and highlight reciprocal strong inputs.

Discussion

Findings reinforce oviIN as a multi-circuit integrator: distinct upstream pathways converge with dendritic‑region specificity. The pipeline provides a baseline for hypothesis testing and follow‑up physiology/behavior studies.

Data & Reproducibility

• Env: conda env create -f environment.yml (or pip install -r requirements.txt)
• Docker: docker build -t oviin-modules . → docker run -p 8501:8501 oviin-modules
• CI: See badge/workflow in repo
• App: streamlit run app.py (module browser)

References

• Connectome & oviIN sources; modularity references; PyTorch docs.