chromatin-state-inference

name	chromatin-state-inference
description	This skill should be used when users need to infer chromatin states from histone modification ChIP-seq data using chromHMM. It provides workflows for chromatin state segmentation, model training, state annotation.

ChromHMM Chromatin State Inference

Overview

This skill enables comprehensive chromatin state analysis using chromHMM for histone modification ChIP-seq data. ChromHMM uses a multivariate Hidden Markov Model to segment the genome into discrete chromatin states based on combinatorial patterns of histone modifications.

Main steps include:

• Refer to Inputs & Outputs to verify necessary files.
• Always prompt user if required files are missing.
• Always prompt user for genome assembly used.
• Always prompt user for the bin size for generating binarized files.
• Always prompt user for the bin size for the number of states the ChromHMM target.
• Always prompt user for the absolute path of ChromHMM JAR file.
• Run chromHMM workflow: Binarization → Learning.

---

When to use this skill

Use this skill when you need to infer chromatin states from histone modification ChIP-seq data using chromHMM.

---

Inputs & Outputs

Inputs

(1) Option 1: BED files of aligned reads

<mark1>.bed
<mark2>.bed
... # Other marks

(1) Option 2: BAM files of aligned reads

<mark1>.bam
<mark2>.bam
... # Other marks

Outputs

chromhmm_output/
  binarized/
    *.txt 
  model/
    *.txt
    ... # other files output by the ChromHMM

---

Decision Tree

Step 1: Prepare the cellmarkfile

• Prepare a .txt file (without header) containing following three columns:
  • sample name
  • marker name
  • name of the BED/BAM file

Step 2: Data Binarization

• For BAM inputs:
  Call:

  • mcp__chromhmm-tools__binarize_bam with:
  • ChromHMM_path: Path to ChromHMM JAR file, provided by user
  • genome: Provide by user (e.g. hg38)
  • input_dir: Directory containing BAM files
  • cellmarkfile: Cell mark file defining histone modifications
  • output_dir: (e.g. binarized/)
  • bin_size: Provided by user

• For BED inputs:
  Call mcp__chromhmm-tools__binarize_bed instead.

Step 3: Model Learning

Call

• mcp__chromhmm-tools__learn_model

with:

• ChromHMM_path: Path to ChromHMM JAR file, provided by user
• binarized_dir: Directory binarized file located in
• num_states: Provide by user (e.g. 15)
• output_model_dir: (e.g. model_15_states/)
• genome: Provide by user (e.g. hg38)
• num_states: Provide by user (e.g. hg38)
• threads: (e.g. 4)

Parameter Optimization

Number of States

• 8 states: Basic chromatin states
• 15 states: Standard comprehensive states
• 25 states: High-resolution states
• Optimization: Use Bayesian Information Criterion (BIC)

Bin Size

• 200bp: Standard resolution
• 100bp: High resolution (requires more memory)
• 500bp: Low resolution (faster computation)

State Interpretation

Common Chromatin States

1. Active Promoter: H3K4me3, H3K27ac
2. Weak Promoter: H3K4me3
3. Poised Promoter: H3K4me3, H3K27me3
4. Strong Enhancer: H3K27ac, H3K4me1
5. Weak Enhancer: H3K4me1
6. Insulator: CTCF
7. Transcribed: H3K36me3
8. Repressed: H3K27me3
9. Heterochromatin: Low signal across marks

Troubleshooting

• Memory errors: Reduce bin size or number of states
• Convergence problems: Increase iterations or adjust learning rate
• Uninterpretable states: Check input data quality and mark combinations
• Missing chromosomes: Verify chromosome naming consistency