Download ENCODE Files

When to Use

• User wants to download ENCODE data files to their local machine
• User asks to "download", "get", or "fetch" ENCODE files
• User needs specific file formats (BED, FASTQ, BAM, bigWig) from experiments
• User wants to batch download files matching search criteria
• User needs to verify file integrity after download (MD5 checksums)
• User asks about organizing downloaded files by experiment or format

Help the user download ENCODE data files to their local machine.

Download Strategy

1. Specific files by accession: Use encode_download_files with file accession IDs (e.g., "ENCFF635JIA").

2. Batch download by criteria: Use encode_batch_download to search and download in one step.

   • Always start with dry_run=True (default) to preview what will be downloaded
   • Show the user the file count, total size, and file list
   • Only proceed with dry_run=False after user confirms

3. Download organization options:

   • "flat": All files in one directory
   • "experiment": Organized by experiment accession (recommended)
   • "format": Organized by file format
   • "experiment_format": Organized by experiment, then format

Important Notes

• All downloads include MD5 verification by default (verify_md5=True)
• Ask the user for a download directory if not specified
• Warn about large downloads (>1GB total or >50 files)
• Files already downloaded will be skipped (idempotent)
• For restricted files, credentials must be configured first via encode_manage_credentials

Pitfalls & Edge Cases

1. Disk space: BAM files can be 5-50GB each; FASTQ files 1-20GB. Before any batch download, warn the user about estimated total size from the dry_run preview. A single ChIP-seq experiment can produce 10-30GB of raw data files.
2. MD5 verification failures: If MD5 verification fails, the file may be corrupted or incompletely downloaded. Always re-download rather than skipping verification. Never set verify_md5=False unless the user explicitly requests it and understands the risk.
3. Downloading too much data: Users often request BAM files when they only need peak calls or signal tracks. Suggest preferred_default=True to get ENCODE's recommended files, or filter by output_type (e.g., "IDR thresholded peaks", "fold change over control") to avoid downloading raw data unnecessarily.
4. Restricted/unreleased data: Files with status other than "released" may require ENCODE credentials. Use encode_manage_credentials(action="check") to verify credentials are configured before attempting to download restricted data.
5. Mixed assemblies in batch download: Always specify the assembly filter (e.g., "GRCh38") in batch downloads. Without it, you may download files aligned to different genome assemblies (hg19, GRCh38, mm10), making downstream analysis impossible.
6. Timeout on large files: For downloading many files or very large files, encode_batch_download handles retries and concurrent downloads better than individual encode_download_files calls. The default limit of 100 files provides a safety cap.

File Type Guide

When users request "files" without specifying a type, use this priority to suggest the right output_type:

• Peak analysis: output_type="IDR thresholded peaks" (most stringent, recommended for ChIP-seq/ATAC-seq)
• Signal visualization: file_format="bigWig", output_type="fold change over control" (for genome browser tracks)
• Gene expression: output_type="gene quantifications" (for RNA-seq TPM/FPKM tables)
• Raw data reprocessing: file_format="fastq" (only when user needs to run their own pipeline)
• Quick defaults: preferred_default=True (ENCODE's recommended files for any experiment)

What to Download for Each Analysis

Assay-Specific Recommendations

File Selection Priority

When multiple files exist for the same experiment, choose files in this priority order:

1. preferred_default=True: ENCODE curators mark recommended files. Always prefer these when available. Use encode_list_files(experiment_accession="ENCSR...", preferred_default=True) to find them.

2. Peak file hierarchy (most to least stringent):

   • IDR thresholded peaks — replicated, irreproducibility-filtered (gold standard)
   • Optimal IDR thresholded peaks — union of replicate-level peaks
   • Conservative IDR thresholded peaks — intersection of replicate-level peaks
   • Pseudoreplicated peaks — peaks from pooled pseudoreplicates
   • Replicated peaks — peaks found in both replicates (broad marks)

3. Signal track hierarchy:

   • fold change over control — normalized signal, best for comparing across experiments
   • signal p-value — statistical significance of enrichment
   • signal of unique reads — uniquely mapped read signal
   • signal of all reads — includes multi-mapped reads (noisier)

4. Assembly preference:

   • GRCh38 for human (current standard) — always use this
   • hg19 for human (legacy) — only if collaborators require it
   • mm10 for mouse (current standard)
   • Never mix assemblies within an analysis

5. Replicate preference:

   • Replicated files (combined replicates) over single-replicate files
   • Biological replicates over technical replicates
   • Isogenic replication over anisogenic

6. Status preference:

   • released — fully validated, use these
   • archived — older versions, avoid unless specifically needed
   • revoked — quality issues found, never use

Storage Estimates

Plan disk space before downloading. Use dry_run=True to get exact sizes for your query.