SkyWhal3/STXBP1-RAG-Nemotron

🧬⚡ STXBP1-ARIA RAG Database v10.1 - NVIDIA Nemotron Embeddings

The most advanced RAG database for STXBP1 therapeutic research.

A pre-built ChromaDB vector database containing: 571,816 indexed text chunks from ~17,000 curated PubMed Central (PMC) biomedical papers + 165 base editing analysis entries, (https://huggingface.co/datasets/SkyWhal3/stxbp1-base-editing-sweep), embedded with NVIDIA's state-of-the-art Llama-Nemotron-Embed-1B-v2 model featuring 2048-dimensional embeddings.

⚡ This is the premium GPU-accelerated version — Nemotron embeddings deliver maximum semantic precision for therapeutic queries, but require a GPU with 2-4GB VRAM. For a lightweight CPU-friendly alternative, see: STXBP1-RAG-Database (BGE)

---

🆕 v10.1 Update: Base Editing Parameter Sweep Data

NEW! This version includes 165 curated entries from our exhaustive base editing compatibility analysis:

Metric	Value
Total Analyses	654,595,200
Variants Analyzed	171 pathogenic STXBP1 variants
Parameters Swept	PAM position, edit window, enzyme type
Mouse Compatibility	Wild-type mouse modeling feasibility

What's Included

Each variant entry contains:

• ✅ Compatibility Score (0-100) for adenine/cytosine base editing
• ✅ Optimal Parameters (SpCas9/SpCas9-NG, PAM position, edit window)
• ✅ Mouse Model Compatibility — Can a WT mouse model this human variant?
• ✅ Clinical Context — ClinVar classification, mutation type

Key Findings Embedded

Finding	Details
Perfect Score Variants	5 variants achieve 100/100 compatibility
K196X (c.586A>T)	Score: 95.9/100, WT mouse compatible ✅
ABE-Curable	68 variants amenable to adenine base editing
CBE-Curable	47 variants amenable to cytosine base editing

📊 Full sweep data available at: SkyWhal3/stxbp1-base-editing-sweep

---

🏆 Why Nemotron?

NVIDIA's Nemotron embedding model ranks #2 on MTEB retrieval benchmarks — distilled from their 8B flagship into an efficient 1B parameter model.

Feature	BGE (v9)	Nemotron (v10)
Embedding Dims	768	2048 ⬆️ 2.7x
Model Params	110M	1B ⬆️ 9x
MTEB Retrieval	~63	~69 ⬆️ +6 pts
Semantic Precision	Good	Excellent
Hardware	CPU OK	GPU recommended
Optional Reranker	❌	✅ Available

What 2048 Dimensions Means

Semantic Space Visualization:

768 dims (BGE)              2048 dims (Nemotron)
─────────────────           ─────────────────────
     ●                              ●
    /|\                          /  |  \  
   / | \                        /   |   \
  ●  ●  ●                      ●    ●    ●
  
  Good separation           Rich semantic space!
                            Fine-grained distinctions

Real-world impact:

• "haploinsufficiency" vs "dominant negative" → better separated
• "4-PBA chaperone" vs "AAV gene therapy" → distinct clusters
• "K196X nonsense" vs "R406H missense" → clear differentiation
• "ABE compatible" vs "prime editing required" → therapeutic routing

---

📊 Dataset Statistics

Metric	Value
Total Chunks	571,816
Source Papers	~17,000 PMC articles
Curated Entries	24 expert-written
Base Editing Entries	165 (v10.1)
Database Size	~11.5 GB
Embedding Model	nvidia/llama-nemotron-embed-1b-v2
Embedding Dimensions	2048
Model Parameters	1B
Chunk Size	~1500 chars with 200 char overlap
Index Type	ChromaDB with HNSW
Build Hardware	NVIDIA H100 80GB
Last Updated	January 6, 2026

---

🎯 Purpose

This database powers STXBP1-ARIA MAX, the premium therapeutic discovery system, enabling:

• Maximum retrieval precision for complex therapeutic queries
• Fine-grained semantic distinctions between mutation types and mechanisms
• Base editing feasibility lookups for specific variants
• Mouse model compatibility guidance for preclinical research
• Optional reranking with Nemotron cross-encoder for top-k refinement
• Literature-grounded responses with PMC citations

---

🔗 Related Resources

Resource	Link
ARIA MAX (Live Demo)	HuggingFace Space
Base Editing Sweep	Dataset
Evaluation Harness	STXBP1-Eval
CPU Version (BGE)	STXBP1-RAG-Database

---

🚀 Quick Start

Load in Python

import chromadb
from chromadb.config import Settings

# Connect to database
client = chromadb.PersistentClient(
    path="./STXBP1-RAG-Nemotron",
    settings=Settings(anonymized_telemetry=False)
)

collection = client.get_collection("stxbp1_papers")
print(f"Loaded {collection.count():,} chunks")

Query with Nemotron Embeddings

from sentence_transformers import SentenceTransformer

# Load embedding model (requires GPU for best performance)
embedder = SentenceTransformer(
    "nvidia/llama-nemotron-embed-1b-v2",
    trust_remote_code=True
)

# Embed query
query = "What base editing options exist for K196X?"
query_embedding = embedder.encode(query).tolist()

# Search
results = collection.query(
    query_embeddings=[query_embedding],
    n_results=10
)

for doc, meta in zip(results['documents'][0], results['metadatas'][0]):
    print(f"[{meta.get('pmcid', 'N/A')}] {doc[:200]}...")

---

📁 Files

File	Size	Description
chroma.sqlite3	~2 GB	Metadata + document store
{uuid}/data_level0.bin	~9 GB	HNSW vector index
{uuid}/header.bin	<1 KB	Index metadata
{uuid}/length.bin	<1 KB	Vector dimensions

---

📜 License

Apache 2.0

---

🙏 Acknowledgments

Built for the STXBP1 research community. Special thanks to:

• STXBP1 Foundation
• NVIDIA for Nemotron embeddings
• The families and researchers working toward treatments

---

Built with ❤️ for rare disease research

Last Updated: January 6, 2026