OmicsClaw: An AI Research Partner That Remembers Your Multi-Omics Data

What is OmicsClaw?

OmicsClaw is an AI research assistant with persistent memory for multi-omics analysis. Unlike stateless tools where every session starts from zero, OmicsClaw remembers your datasets, analysis history, and preferences across sessions.

It covers 63+ skills across 6 omics domains: spatial transcriptomics, single-cell, genomics, proteomics, metabolomics, and bulk RNA-seq. Chat with it via Telegram or Feishu, or use the CLI.

Why Memory Matters

Traditional Tools	OmicsClaw
Re-upload data every session	Remembers file paths and metadata
Forget analysis history	Tracks full lineage (preprocess → cluster → DE)
Repeat parameters manually	Learns and applies your preferences
CLI-only, steep learning curve	Chat interface + CLI
Stateless execution	Persistent research partner

Example conversation with memory:

Session 1:
You: [Upload visium_brain.h5ad]
You: "Preprocess this"
Bot: Done. [Saves dataset + analysis to memory]

Session 2 (next day):
You: "Find spatial domains"
Bot: "Using your Visium brain data (5000 spots, normalized yesterday)"
     Done. [Links to parent analysis]

Quick Start

Chat Interface (Recommended)

git clone https://github.com/TianGzlab/OmicsClaw.git
cd OmicsClaw
pip install -e .
pip install -r bot/requirements.txt

# Configure
cp .env.example .env
# Edit .env: set LLM_API_KEY and bot tokens

# Start Telegram bot
python bot/telegram_bot.py

Command Line

pip install -e .

# Try a demo (no data needed)
python omicsclaw.py run spatial-preprocessing --demo

# Run with your data
python omicsclaw.py run spatial-preprocessing --input data.h5ad --output results/

Installation Tiers

pip install -e . — Core system
pip install -e ".[spatial]" — Spatial transcriptomics
pip install -e ".[singlecell]" — Single-cell omics
pip install -e ".[genomics]" — Genomics
pip install -e ".[full]" — All 63+ skills

The 6 Omics Domains

Domain	Skills	Key Capabilities
Spatial Transcriptomics	15	QC, clustering, cell typing, deconvolution, spatial statistics, cell communication, velocity
Single-Cell Omics	9	Preprocessing, doublet detection, annotation, trajectory, batch integration, GRN
Genomics	10	Variant calling, alignment, annotation, structural variants, assembly, CNV
Proteomics	8	MS QC, peptide ID, quantification, differential abundance, PTM analysis
Metabolomics	8	Peak detection, XCMS preprocessing, annotation, normalization, pathway enrichment
Bulk RNA-seq	13	FASTQ QC, alignment, DE, splicing, enrichment, deconvolution, co-expression, survival

Example: Spatial Transcriptomics Pipeline

# 1. Preprocess: QC, normalize, cluster
python omicsclaw.py run spatial-preprocessing --input data.h5ad --output output/preprocess

# 2. Identify tissue domains
python omicsclaw.py run spatial-domain-identification --input output/preprocess/processed.h5ad --output output/domains

# 3. Find spatially variable genes
python omicsclaw.py run spatial-svg-detection --input output/domains/processed.h5ad --output output/svg

# 4. Cell-cell communication
python omicsclaw.py run spatial-cell-communication --input output/preprocess/processed.h5ad --output output/communication

Smart Orchestration

The orchestrator routes natural language to the right analysis:

# Keyword mode (fast)
python omicsclaw.py run orchestrator --query "find spatially variable genes" --input data.h5ad --output output

# LLM mode (semantic understanding)
python omicsclaw.py run orchestrator --query "I want to understand which genes show spatial patterns" --routing-mode llm --output output

# Named pipelines
python omicsclaw.py run orchestrator --pipeline standard --input data.h5ad --output output

Supported Platforms

Data formats: .h5ad, .vcf, .mzML, .csv/.tsv

Spatial platforms: Visium, Xenium, MERFISH, Slide-seq

Sequencing: 10x scRNA-seq, Illumina, PacBio, LC-MS/MS

OmicsClaw: An AI Research Partner That Remembers Your Multi-Omics Data

Table of Contents