gabrielmoreira Upstream Skills Index
This directory contains bioinformatics upstream analysis skills for agent-assisted data processing. Each skill provides workflow guidance, code templates, and best practices for specific analysis types.
Available Skills
| Skill | File | Description | Key Use Cases |
|---|---|---|---|
| ATAC-seq | atac.md | Bulk ATAC-seq analysis pipeline | Chromatin accessibility profiling, peak calling, motif analysis |
| RNA-seq | rna.md | Bulk RNA-seq analysis pipeline | Transcriptome profiling, gene expression quantification |
| scATAC-seq | scatac.md | Single-cell ATAC-seq analysis | Single-cell chromatin accessibility, cellranger-atac processing |
| scRNA-seq | scrna.md | Single-cell RNA-seq analysis | Single-cell transcriptomics, cell type annotation with LLM |
| Spatial | spatial.md | Spatial transcriptomics analysis | Visium HD bin-to-cell conversion, spatial cell segmentation |
Skill Summaries
ATAC-seq (atac.md)
Purpose: Complete bulk ATAC-seq data processing from raw FASTQ files to peak calls and downstream analysis.
Key Workflows:
- Quality control with FastQC
- Adapter trimming with Trim Galore
- Genome alignment with Bowtie2/BWA
- BAM filtering and duplicate removal
- Peak calling with MACS2/Genrich
- Coverage track generation
- Motif analysis with HOMER
Tools Used: FastQC, Trim Galore, Bowtie2, BWA, samtools, Picard, MACS2, Genrich, deepTools, HOMER
RNA-seq (rna.md)
Purpose: Complete bulk RNA-seq data processing from raw FASTQ files to gene expression quantification.
Key Workflows:
- Quality control with FastQC
- Adapter trimming
- Genome alignment with STAR/HISAT2
- Expression quantification with featureCounts
- BAM processing and QC
- Coverage track generation
Tools Used: FastQC, Trim Galore, STAR, HISAT2, featureCounts, samtools, MultiQC
scATAC-seq (scatac.md)
Purpose: Single-cell ATAC-seq data processing using 10X Genomics cellranger-atac with downstream analysis.
Key Workflows:
- Smart cellranger-atac detection and installation
- 10X FASTQ file validation and renaming
- Reference genome setup
- cellranger-atac count execution
- Quality control filtering
- Dimensionality reduction (LSI/PCA/UMAP)
- Graph-based clustering
- Peak annotation
Tools Used: cellranger-atac, scanpy, custom downstream analysis tools
scRNA-seq (scrna.md)
Purpose: Comprehensive single-cell RNA-seq analysis with omicverse integration and LLM-powered cell type annotation.
Key Workflows:
- Data loading and inspection
- Quality control with omicverse
- Preprocessing (normalization, HVG selection)
- PCA and batch correction
- Clustering and UMAP visualization
- Marker gene discovery (from data and description)
- AUCell scoring for cell type markers
- LLM-powered cell type annotation
Tools Used: scanpy, omicverse, AUCell, LLM annotation
Special Features:
- Persistent Python state management
- Context-aware marker generation
- Interactive LLM cell type annotation
- Structured results output
Spatial Transcriptomics (spatial.md)
Purpose: Spatial transcriptomics analysis for Visium HD data with bin-to-cell conversion using omicverse.
Key Workflows:
- Visium 10X data loading
- H&E-based cell segmentation (Cellpose)
- Label expansion
- GEX-based cell segmentation
- Secondary label salvaging
- Bin-to-cell conversion
Tools Used: omicverse, Cellpose, scanpy
Special Features:
- Flexible execution (template-guided or direct implementation)
- Multi-modal cell segmentation (H&E + GEX)
- Intelligent label combination
Common Patterns Across Skills
TodoList Management
All skills use a consistent TodoList pattern:
show_todos()- View current task listadd_todo("task")- Add new taskmark_task_done("description")- Complete current taskexecute_current_task()- Get guidance for next action
Phase-based Execution
- Phase 0: Setup and validation
- Phase 1: TODO creation (strict de-duplication)
- Phase 2: Execute with tracking (loop)
- Phase 3: Adaptive refinement
Python Environment Rules (for Python-based skills)
- Persistent state across calls
- Memory-first approach (avoid redundant I/O)
- Smart variable checking with try/except
- Error recovery without full restart
How to Use
- Select the appropriate skill based on your data type
- Read the skill document to understand the workflow
- Follow the phase-based execution pattern
- Adapt code templates to your specific data
- Analyze results at each step before proceeding
- Use TodoList to track progress
Integration Notes
These skills are designed to work with:
- Agent-based execution systems
- TodoList management tools
- Python persistent interpreters (for Python-based skills)
- Bash command execution (for pipeline skills)
Each skill can be invoked independently or as part of a larger analysis workflow.