Processing held-out CRISPR enhancer screens for benchmarking ENCODE-rE2G

Overview

This pipeline processes validation datasets to test ENCODE_rE2G performance on predicting enhancer-gene pairs through:

1. Differential Expression Analysis: Uses SCEPTRE to identify significant enhancer-gene interactions in various CRISPRi Perturb-seq datasets
2. Power Analysis: Simulates perturbation effects at various effect sizes to create a high-confidence set of non-significant element-gene pairs
3. Data Integration: Incorporates various other held-out datasets and applies filtering based on genic features to eliminate confounders
4. Benchmarking Output: Produces final held-out dataset for enhancer-gene prediction benchmarking

The final output file used for benchmarking can be found in results/combine_val_data_and_format/Final_Validation_Dataset.tsv.gz

Repository Structure

ENCODE_Sceptre_Analysis/
├── config/
│   └── config.yml                 # Pipeline configuration parameters
├── workflow/
│   ├── Snakefile                  # Main workflow definition
│   ├── rules/                     # Snakemake rule definitions
│   │   ├── sceptre_setup.smk
│   │   ├── sceptre_power_analysis.smk
│   │   ├── create_encode_output.smk
│   │   └── combine_val_data_and_format.smk
│   ├── scripts/                   # Analysis scripts
│   │   ├── sceptre_setup/
│   │   ├── sceptre_power_analysis/
│   │   ├── encode_datasets/
│   │   └── combine_val_data_and_format/
│   └── envs/                      # Conda environment definitions
├── resources/                     # Input data and reference files
├── results/                       # Pipeline outputs
├── Perturb_Seq_Test_Set_Preprocessing/  # Pre-processing to create pipeline inputs - README.md included in this folder
└── README.md

Requirements

Software Dependencies

• Snakemake: 7.3.2
• Conda: 24.11.3

Environment Management

Dependencies are managed automatically through conda environments defined in workflow/envs/. Snakemake will create and activate the appropriate environments for each step.

Key environments include:

• sceptre_env.yml: SCEPTRE differential expression analysis
• analyze_crispr_screen.yml: Single-cell analysis tools
• r_process_crispr_data.yml: Data processing and formatting

Input Data

Required Input Files in resources/

1. Externally processed Enhancer-Gene pairs

   • DC TAP-seq data: resources/combine_val_data_and_format/DC_TAP_Seq_data.tsv
     • File source
   • ENCODE-rE2G Training dataset: resources/combine_val_data_and_format/EPCrisprBenchmark_ensemble_data_GRCh38.tsv
   • Other test datasets: resources/create_encode_output/ENCODE/EPCrisprBenchmark/

2. Raw data input created in Perturb_Seq_Test_Set_Preprocessing - see Data Processing section

   • Klann et al. 2021: resources/sceptre_setup/Klann/
   • Morris et al. 2023: resources/sceptre_setup/Morrisv1/ & resources/sceptre_setup/Morrisv2/
   • Xie et al. 2019: resources/sceptre_setup/Xie/

Data Preprocessing

(Optional as these results are included on SYNAPSE) Before running the main pipeline, you must first preprocess the raw CRISPR screen data.

See the comprehensive guide in: Perturb_Seq_Test_Set_Preprocessing/README.md

This preprocessing step includes:

• Converting raw sequencing data to count matrices
• Filtering for high-confidence guides and creating a guide annotation file
• Generating metadata files

Running the Pipeline

Quick Start

1. Clone the repository:

git clone https://github.com/jamesgalante/ENCODE_Test_Dataset_Analysis.git
cd ENCODE_Test_Dataset_Analysis

2. Complete data preprocessing (see Perturb_Seq_Test_Set_Preprocessing/README.md)

• Optionally, download the preprocessed data from [SYNAPSE](INCLUDE LINK TO SYNAPSE)

3. Run the complete pipeline:

# For HPC with SLURM
snakemake all

# For local execution (not recommended due to size)
snakemake --use-conda --cores 8 all

Configuration Profiles

SLURM Profile (Recommended for HPC)

While these configuration parameters can be included via flags (e.g. --use-conda) and are often variable depending on the HPC setup, the profile used to create this pipeline is provided. Store this profile in a Snakemake config file (e.g., ~/.config/snakemake/slurm_profile/config.yaml):

jobs: 500
cluster: slurm
use-conda: true
notemp: true
default-resources:
    - runtime="13h"
    - mem="32G"
# Add your specific SLURM configuration:
# - slurm_account=your_account
# - slurm_partition=your_partition
# - slurm_extra="--nice"

Then run:

snakemake --profile slurm_profile all

Pipeline Stages

1. SCEPTRE Setup (sceptre_setup.smk)

• Downloads genome annotation files
• Pairs all tested elements to any gene within 1Mb
• Creates SCEPTRE input objects for each dataset

2. Power Analysis (sceptre_power_analysis.smk)

• Runs SCEPTRE differential expression analysis
• Performs power simulations at multiple effect sizes (2%, 3%, 5%, 10%, 15%, 20%, 25%, 50%)
• Estimates statistical power for detecting enhancer-gene interactions

3. ENCODE Formatting (create_encode_output.smk)

• Filters based on genic features
• Integrates other test datasets from resources/create_encode_output/ENCODE/EPCrisprBenchmark/

4. Final Integration (combine_val_data_and_format.smk)

• Integrates DC TAP-seq data
• Removes training set overlaps
• Resolves duplicates between datasets
• Produces final held-out dataset

Outputs

• Final Validation Dataset: results/combine_val_data_and_format/Final_Validation_Dataset.tsv.gz