NeuroXiv2

NeuroXiv2 is a multi-modal neuroscience data integration platform that unifies anatomy, morphology, spatial transcriptomics, and connectivity data into a knowledge graph (NeuroXiv-KG), and provides an AI-powered query agent (AIPOM-CoT) for intelligent exploration.

Repository Structure

NeuroXiv2/
├── NeuroXiv2-KG/       # Knowledge graph construction pipeline
├── AIPOM-CoT/          # AI agent with chain-of-thought reasoning
└── Data-Analysis/      # Cross-modal statistical analysis

• NeuroXiv2-KG — Builds NeuroXiv-KG from MERFISH spatial transcriptomics, neuron morphology reconstructions, axonal projections, and CCF brain atlas annotations, then loads all nodes and relationships into Neo4j.
• AIPOM-CoT — An adaptive, schema-guided chain-of-thought agent that translates natural-language neuroscience questions into Cypher queries, executes them against NeuroXiv-KG, and synthesises answers with statistical analysis and provenance tracking. Exposed via a FastAPI REST API.
• Data-Analysis — Scripts for cross-modal fingerprint computation, region separability analysis, within-region homogeneity tests, and multi-modal prediction (classification and clustering).

Software Dependencies

Operating System

The system has been developed and tested on macOS (Darwin). It is expected to be compatible with Linux distributions that support Docker and Python 3.9+.

Core Infrastructure

Software	Version	Purpose
Python	3.9.12	Runtime for all modules
Neo4j Community Edition	5.22.0	Graph database for NeuroXiv-KG
Docker	28.3.2	Container deployment for Neo4j

Using all the multi-modality data in NeuroXiv2, we built a knowledge graph namely NeuroXiv-KG and implemented it in Neo4j Community Edition version 5.22.0. NeuroXiv-KG is deployed in a Docker container using Docker version 28.3.2. The graph schema was designed to support unified queries across anatomy, morphology, spatial transcriptomics, and connectivity.

Python Dependencies

Database & API

Package	Purpose
neo4j	Neo4j Python driver for Bolt protocol connectivity
fastapi	REST API framework for AIPOM-CoT agent endpoint
uvicorn	ASGI server for FastAPI
pydantic	Data validation and serialization

Data Processing & Scientific Computing

Package	Purpose
numpy	Numerical computing
pandas	Tabular data manipulation
scipy	Scientific computing (statistics, spatial algorithms, distance metrics)
anndata	AnnData objects for single-cell/spatial transcriptomics data
h5py	HDF5 file I/O for large-scale datasets

Machine Learning

Package	Purpose
scikit-learn	Classification (RandomForest, LogisticRegression), clustering (KMeans), PCA, metrics
joblib	Model serialization and persistence

Medical / Neuroimaging

Package	Purpose
nrrd	NRRD volumetric image format I/O (CCF brain atlas, 25 µm resolution)
SimpleITK	Medical image processing for atlas coordinate mapping

Visualization

Package	Purpose
matplotlib	Publication-quality figure generation
seaborn	Statistical data visualization

Utilities

Package	Purpose
loguru	Structured logging
pyyaml	YAML configuration parsing
openai	OpenAI API client (GPT-4o for AIPOM-CoT reasoning)

Installation

pip install neo4j fastapi uvicorn pydantic \
            numpy pandas scipy anndata h5py \
            scikit-learn joblib \
            nrrd SimpleITK \
            matplotlib seaborn \
            loguru pyyaml openai

Configuration

AIPOM-CoT is configured via AIPOM-CoT/config.yaml. Key settings:

Parameter	Default	Description
database.neo4j_uri	bolt://localhost:7687	Neo4j Bolt endpoint
database.neo4j_user	neo4j	Database username
openai.model	gpt-4o	LLM for chain-of-thought reasoning
adaptive_planner.default_depth	deep	Query planning depth (shallow / medium / deep)
adaptive_planner.max_iterations	15	Maximum reasoning iterations

Environment variables (NEO4J_URI, NEO4J_USER, NEO4J_PASSWORD, OPENAI_API_KEY) override config file values when set.