CRISPR Screen Analyzer

Analyze pooled CRISPR screening data to identify essential genes, drug resistance/sensitivity candidates, and screen quality metrics. Supports Robust Rank Aggregation (RRA) analysis, quality control assessment, and hit identification for functional genomics studies.

Key Capabilities:

• - Quality Control Assessment: Calculate Gini index, read depth, and dropout metrics to evaluate screen quality
• Log Fold Change Calculation: Compute sgRNA-level fold changes between treatment and control conditions
• Statistical Analysis: Perform Robust Rank Aggregation (RRA) to identify significantly enriched or depleted sgRNAs
• Hit Identification: Apply FDR and fold change thresholds to identify candidate genes
• Multi-Sample Support: Process multiple replicates and treatment conditions simultaneously

---

When to Use

✅ Use this skill when:

• - Analyzing genome-wide viability screens to identify essential genes required for cell survival
• Performing drug resistance screens to find genes whose knockout confers resistance
• Conducting drug sensitivity screens to identify synthetic lethal interactions
• Performing quality control assessment of CRISPR screen data before downstream analysis
• Comparing multiple treatment conditions (e.g., drug vs DMSO, hypoxia vs normoxia)
• Validating screen quality before publication or further experimental validation
• Generating hit lists for secondary screens or validation experiments

❌ Do NOT use when:

• - Analyzing single-cell CRISPR data (Perturb-seq, CROP-seq) → Use specialized single-cell analysis tools
• Working with arrayed CRISPR screens (well-by-well format) → Use standard differential expression analysis
• Performing CRISPR activation (CRISPRa) or interference (CRISPRi) screens → May need adjusted normalization
• Requiring Bayesian or MAGeCK statistical analysis → This tool uses RRA; use MAGeCK for alternative algorithms
• Analyzing small custom libraries (<1000 sgRNAs) → Statistical power may be insufficient
• Time-course CRISPR screens → Requires specialized trajectory analysis methods

Related Skills:

• - 上游 (Upstream): crispr-grna-designer, INLINECODE1
• 下游 (Downstream): go-kegg-enrichment, pathway-visualization, INLINECODE4

---

Integration with Other Skills

Upstream Skills:

• - crispr-grna-designer: Design sgRNA libraries before screening; validate library composition
• INLINECODE6: Assess sequencing quality before CRISPR screen analysis
• INLINECODE7: Verify sgRNA alignment rates and mapping quality

Downstream Skills:

• - go-kegg-enrichment: Perform pathway enrichment on identified hit genes
• INLINECODE9: Visualize hits in pathway contexts
• INLINECODE10: Design follow-up experiments for candidate genes
• INLINECODE11: Compare screen results with known essential gene databases

Complete Workflow:

Library Design (crispr-grna-designer) → Transduction → Sequencing → fastqc-report-interpreter → crispr-screen-analyzer → go-kegg-enrichment → Hit Validation

---

Core Capabilities

1. Quality Control Metrics Calculation

Assess CRISPR screen quality using established metrics including Gini index, read depth, and sgRNA dropout rates.

CODEBLOCK1

QC Metrics Explained:

Metric	Target Range	Interpretation
Gini Index	<0.3	Measures library evenness; lower = more uniform
Total Reads

>10M per sample | Sufficient depth for statistical power |
| Zero-count sgRNAs | <5% | Acceptable dropout; higher indicates library loss |
| Read Distribution | Log-normal | Should follow expected distribution |

Best Practices:

• - ✅ Check Gini index first: Values >0.4 indicate potential library bias or bottleneck
• ✅ Compare replicates: QC metrics should be consistent across replicates
• ✅ Assess time points: Later time points typically show higher dropout
• ✅ Validate early: Poor QC may require screen repetition

Common Issues and Solutions:

Issue: High Gini index (>0.4)

• - Symptom: Uneven sgRNA representation suggesting library bottleneck
• Solution: Check MOI (multiplicity of infection); verify puromycin selection; consider repeating screen

Issue: Excessive zero-count sgRNAs (>10%)

• - Symptom: Many sgRNAs not detected in final samples
• Causes: Low sequencing depth, library degradation, or strong selection
• Solution: Increase sequencing depth; verify library quality at transduction

2. Log Fold Change Calculation

Calculate log2 fold changes between treatment and control conditions to identify enriched or depleted sgRNAs.

CODEBLOCK2

LFC Calculation:
CODEBLOCK3

Interpretation:

LFC Range	Interpretation	Biological Meaning
LFC < -2	Strong depletion	Essential gene or drug sensitivity
LFC -2 to -1

Moderate depletion | Moderate effect |
| LFC -1 to 1 | No change | No significant effect |
| LFC 1 to 2 | Moderate enrichment | Moderate resistance |
| LFC > 2 | Strong enrichment | Resistance gene or suppressor |

Best Practices:

• - ✅ Use pseudocount of 1 to avoid log(0) issues
• ✅ Average replicates to reduce technical variance
• ✅ Visualize distribution to identify batch effects or outliers
• ✅ Check positive controls (known essential genes should have negative LFC)

Common Issues and Solutions:

Issue: Skewed LFC distribution

• - Symptom: Mean LFC significantly different from 0
• Causes: Library size differences, batch effects, or strong selection
• Solution: Apply TMM or DESeq2 normalization; check for batch effects

Issue: Extreme outliers

• - Symptom: Few sgRNAs with very large LFC values
• Solution: Winsorize extreme values; verify these are not technical artifacts

3. Robust Rank Aggregation (RRA) Statistical Analysis

Perform statistical analysis to identify significantly enriched or depleted sgRNAs using z-score and FDR correction.

CODEBLOCK4

RRA Analysis Steps:

1. 1. Z-score calculation: INLINECODE12
2. P-value calculation: Two-tailed normal test
3. FDR correction: Benjamini-Hochberg procedure

Statistical Output:

Column	Description	Usage
INLINECODE13	sgRNA identifier	Mapping to genes
INLINECODE14

Log fold change | Effect size |
| pvalue | Raw p-value | Statistical significance |
| fdr | Adjusted p-value (FDR) | Multiple testing correction |

Best Practices:

• - ✅ Use FDR < 0.05 as standard significance threshold
• ✅ Consider FDR < 0.01 for high-confidence hits
• ✅ Combine p-value and LFC for hit prioritization
• ✅ Validate top hits experimentally before publication

Common Issues and Solutions:

Issue: No significant hits despite visible effects

• - Symptom: Biological effects present but no FDR-significant results
• Causes: High variance, insufficient replicates, or weak effects
• Solution: Increase replicate number; use more permissive FDR threshold; use gene-level aggregation

Issue: Too many significant hits

• - Symptom: Hundreds or thousands of FDR-significant sgRNAs
• Causes: Low variance, strong selection, or batch effects
• Solution: Apply more stringent FDR threshold; add LFC cutoff; filter by effect size

4. Hit Identification with Thresholds

Apply statistical and biological thresholds to identify candidate genes for follow-up validation.

CODEBLOCK5

Hit Classification:

Category	Criteria	Biological Interpretation
Essential	FDR<0.05, LFC<-1	Required for cell viability
Drug Sensitive

FDR<0.05, LFC<-1 | Synthetic lethal with treatment |
| Drug Resistant | FDR<0.05, LFC>1 | Confers resistance to treatment |
| Suppressor | FDR<0.05, LFC>1 | Suppresses phenotype of interest |

Best Practices:

• - ✅ Use consistent thresholds across related screens for comparability
• ✅ Require multiple sgRNAs per gene for confidence (≥2 recommended)
• ✅ Validate with orthogonal methods (siRNA, rescue experiments)
• ✅ Compare with known essential genes as positive controls

Common Issues and Solutions:

Issue: Single sgRNA hits

• - Symptom: Only one sgRNA per gene significant
• Solution: Require ≥2 significant sgRNAs per gene; check for off-target effects

Issue: Off-target effects dominating

• - Symptom: Known essential genes not identified; unexpected hits prominent
• Solution: Use second-generation libraries with improved specificity; validate with rescue

5. Gene-Level Aggregation

Aggregate sgRNA-level results to gene-level statistics for biological interpretation.

CODEBLOCK6

Gene Aggregation Methods:

Method	Description	Best For
Mean LFC	Average across sgRNAs	General hit calling
Best FDR

Most significant sgRNA | Conservative approach |
| Second-best | Second most significant | Reduces outlier effects |
| STARS/RRA | Rank-based aggregation | Standard CRISPR analysis |

Best Practices:

• - ✅ Require ≥3 sgRNAs per gene for reliable gene-level calling
• ✅ Use mean LFC for primary analysis; best FDR for validation
• ✅ Check sgRNA concordance - all should show same direction
• ✅ Remove genes with conflicting sgRNAs from hit list

Common Issues and Solutions:

Issue: Discordant sgRNAs for same gene

• - Symptom: Some sgRNAs positive, others negative for same gene
• Causes: Off-target effects, library errors, or complex biology
• Solution: Exclude genes with discordant sgRNAs; investigate specific cases

6. Multi-Condition Comparison

Compare CRISPR screen results across multiple treatment conditions or time points.

CODEBLOCK7

Multi-Condition Analysis:

Comparison Type	Question Addressed	Interpretation
Drug vs Control	What genes mediate drug response?	Resistance/sensitivity mechanisms
Condition A vs B

Differential genetic dependencies | Context-specific essentiality |
| Time-course | How does genetic dependency change? | Temporal dynamics |
| Cell line comparison | Cell-type specific dependencies | Lineage-specific vulnerabilities |

Best Practices:

• - ✅ Use same control across multiple treatments for comparability
• ✅ Check correlation between replicates and conditions
• ✅ Look for condition-specific hits for mechanism insights
• ✅ Validate common hits as robust findings

Common Issues and Solutions:

Issue: High variability between replicates

• - Symptom: Low correlation between replicates of same condition
• Solution: Increase replicate number; check for technical batch effects

---

Complete Workflow Example

From count matrix to hit identification:

CODEBLOCK8

Python API Usage:

CODEBLOCK9

Expected Output Files:

CODEBLOCK10

---

Common Patterns

Pattern 1: Viability Screen (Essential Gene Identification)

Scenario: Identify genes essential for cell survival by comparing T0 (transduction) vs T14 (14 days post-transduction).

CODEBLOCK11

Workflow:

1. 1. Collect cells at T0 (immediately after transduction)
2. Maintain parallel culture for 14 days (T14)
3. Harvest T14 cells when control cells reach confluence
4. Sequence both T0 and T14 samples
5. Analyze depletion of sgRNAs at T14 relative to T0
6. Identify genes with significantly depleted sgRNAs (essential genes)
7. Validate top hits with individual sgRNA validation

Output Example:
CODEBLOCK12

Pattern 2: Drug Resistance Screen

Scenario: Identify genes whose knockout confers resistance to a cytotoxic drug (e.g., vemurafenib in BRAF-mutant melanoma).

CODEBLOCK13

Workflow:

1. 1. Transduce cells with genome-wide sgRNA library
2. Split into drug-treated and DMSO control groups
3. Treat with drug at appropriate concentration (IC70-IC90)
4. Maintain for 2-3 weeks until control cells die
5. Harvest resistant colonies from drug-treated group
6. Compare sgRNA representation: Drug vs DMSO
7. Identify enriched sgRNAs (resistance genes)
8. Validate resistance with individual sgRNAs and drug dose-response

Output Example:
CODEBLOCK14

Pattern 3: Drug Sensitivity/Synthetic Lethality Screen

Scenario: Identify genes that, when knocked out, sensitize cells to drug treatment (synthetic lethal interactions).

CODEBLOCK15

Workflow:

1. 1. Transduce cells with sgRNA library
2. Treat with sub-lethal drug concentration (IC30)
3. Maintain for 2 weeks under drug selection
4. Compare sgRNA representation: Drug-treated vs control
5. Identify depleted sgRNAs (synthetic lethal/sensitizer genes)
6. Validate with individual sgRNAs and combination assays
7. Compare with genetic dependency maps (DepMap)

Output Example:
CODEBLOCK16

Pattern 4: Comparative Screen (Cell Line vs Cell Line)

Scenario: Compare genetic dependencies between two cell lines to identify lineage-specific vulnerabilities.

CODEBLOCK17

Workflow:

1. 1. Perform viability screens in multiple cell lines in parallel
2. Normalize each screen independently
3. Compare gene-level essentiality scores across lines
4. Identify genes essential in one lineage but not another
5. Validate lineage-specific dependencies
6. Explore therapeutic relevance (tumor-type specific targets)

Output Example:

Comparative Screen: Melanoma vs Lung Cancer
  Melanoma-specific essential: 127 genes
  Lung-specific essential: 203 genes
  Common essential: 1,847 genes
  
Top Melanoma-Specific Dependencies:
  MITF:   LFC diff = -4.5 (essential in melanoma, not lung)
  SOX10:  LFC diff = -3.8
  TYR:    LFC diff = -3.2
  
Top Lung-Specific Dependencies:
  NKX2-1: LFC diff = -3.9
  TP63:   LFC diff = -3.1
  
Therapeutic Implications:
  - Lineage-specific targets identified
  - Potential for tumor-type selective therapy

---

Quality Checklist

Pre-Analysis Checks:

• - [ ] CRITICAL: Verify library composition matches expected sgRNA list
• [ ] Check sequencing depth (>10M reads per sample recommended)
• [ ] Confirm sample annotations match count matrix columns
• [ ] Verify control and treatment sample assignments are correct
• [ ] Check for batch effects (different sequencing runs, library preps)
• [ ] Review positive control performance (known essential genes)
• [ ] Confirm negative controls show no significant effects
• [ ] Validate replicate consistency (correlation >0.7 expected)

During Analysis:

• - [ ] Calculate and review QC metrics (Gini, read depth, dropout)
• [ ] CRITICAL: Check Gini index <0.4 for library quality
• [ ] Examine LFC distribution for normality and outliers
• [ ] Verify positive controls are significantly depleted (viability screens)
• [ ] Check for batch effects using PCA or correlation heatmaps
• [ ] Apply appropriate statistical thresholds (FDR < 0.05 standard)
• [ ] Require multiple sgRNAs per gene for hit calling (≥2 recommended)
• [ ] Compare hit lists with published data for similar screens

Post-Analysis Verification:

• - [ ] CRITICAL: Validate top hits show concordance across sgRNAs
• [ ] Check known positive controls are recovered
• [ ] Assess negative control performance (should not be significant)
• [ ] Compare replicate correlation for hits vs non-hits
• [ ] Review hit gene functions for biological plausibility
• [ ] Check for potential off-target effects (seed sequence analysis)
• [ ] Verify hit numbers are reasonable (10s-100s, not 1000s)
• [ ] Generate visualization (MA plots, volcano plots, heatmaps)

Before Validation or Publication:

• - [ ] CRITICAL: Validate top 5-10 hits with individual sgRNAs
• [ ] Perform rescue experiments to confirm on-target effects
• [ ] Compare with orthogonal datasets (DepMap, published screens)
• [ ] Check for cell line-specific vs pan-essential classification
• [ ] Assess therapeutic relevance of identified hits
• [ ] Plan secondary screens if primary screen quality issues found
• [ ] Document all parameters and thresholds used
• [ ] Prepare data for public deposition (if applicable)

---

Common Pitfalls

Experimental Design Issues:

• - ❌ Insufficient sequencing depth → Poor statistical power, missed hits

- ✅ Minimum 10M reads per sample; 20M+ for complex libraries

• - ❌ Library bottleneck → Gini index >0.4, skewed representation

- ✅ Maintain MOI <0.3; use sufficient cell numbers (500-1000x library coverage)

• - ❌ Inadequate replicates → High variance, irreproducible results

- ✅ Use ≥3 biological replicates per condition

• - ❌ Wrong time point → Too early (no selection) or too late (extensive dropout)

- ✅ Optimize time point based on doubling time and selection pressure

Analysis Issues:

• - ❌ Ignoring QC metrics → Analyzing poor quality data

- ✅ Always review Gini index, read depth, and dropout before analysis

• - ❌ Incorrect sample assignment → Control/treatment mix-up

- ✅ Double-check sample annotation file; validate with positive controls

• - ❌ Single sgRNA hits → Potential off-target effects

- ✅ Require ≥2 significant sgRNAs per gene; check concordance

• - ❌ Over-reliance on p-values → Many false positives with large library

- ✅ Use FDR correction; add LFC threshold; validate experimentally

Interpretation Issues:

• - ❌ Ignoring cell number effects → Different growth rates confound results

- ✅ Normalize for cell doublings; use appropriate controls

• - ❌ Off-target effects dominating → False positive hits

- ✅ Use improved libraries (e.g., Brunello, Brie); validate with rescue

• - ❌ Pan-essential vs selective → Misclassifying broadly essential genes

- ✅ Compare with DepMap data; use differential analysis for specificity

• - ❌ Not validating hits → Publishing false positives

- ✅ Validate top hits with individual sgRNAs; perform rescue experiments

Technical Issues:

• - ❌ Batch effects → Confounding by library prep or sequencing batch

- ✅ Randomize samples across batches; include batch in statistical model

• - ❌ Contamination → Cross-sample contamination affects quantification

- ✅ Use unique molecular identifiers (UMIs); check for index hopping

• - ❌ Reference genome mismatch → sgRNAs not mapping correctly

- ✅ Use same genome version as library design; check sgRNA sequences

• - ❌ Incomplete annotation → sgRNAs missing gene mapping

- ✅ Verify library annotation file is complete and current

---

Troubleshooting

Problem: No significant hits despite strong biological effect

• - Symptoms: Clear phenotype but no FDR-significant sgRNAs
• Causes:

- High variance between replicates
- Insufficient sequencing depth
- Weak effect sizes
- Stringent statistical thresholds

• - Solutions:

- Increase replicate number
- Increase sequencing depth
- Use more permissive FDR threshold (0.1)
- Consider gene-level aggregation

Problem: Too many significant hits (1000s)

• - Symptoms: Excessive number of hits, many likely false positives
• Causes:

- Low variance (overdispersion underestimated)
- Strong selection pressure
- Library quality issues
- Noisy data

• - Solutions:

- Use more stringent FDR threshold (0.01)
- Increase LFC threshold (1.5 or 2.0)
- Filter by sgRNA concordance
- Review QC metrics and repeat if poor quality

Problem: High Gini index (>0.4)

• - Symptoms: Library representation highly skewed
• Causes:

- Library bottleneck at transduction
- Insufficient cell numbers
- High MOI leading to multiple integrations

• - Solutions:

- Use lower MOI (<0.3)
- Increase cell numbers (500-1000x library size)
- Improve transduction efficiency
- Consider repeating screen

Problem: Known essential genes not identified

• - Symptoms: Positive controls (RPL30, RPS19) not significantly depleted
• Causes:

- Insufficient selection time
- Library quality issues
- Analysis errors

• - Solutions:

- Extend time point for viability screens
- Check library composition and representation
- Verify analysis parameters (control vs treatment assignment)

Problem: Discordant sgRNAs for same gene

• - Symptoms: Only 1-2 of 5 sgRNAs significant for hit genes
• Causes:

- Off-target effects
- Variable sgRNA efficiency
- Library design issues

• - Solutions:

- Require ≥3 significant sgRNAs for gene-level hits
- Check sgRNA sequences for off-target potential
- Use improved second-generation libraries
- Validate with independent sgRNAs

Problem: Batch effects between replicates

• - Symptoms: Low correlation between replicates of same condition
• Causes:

- Different library prep batches
- Different sequencing runs
- Technical variation

• - Solutions:

- Include batch as covariate in analysis
- Use ComBat or similar batch correction
- Re-sequence inconsistent replicates
- Randomize samples across batches in future

Problem: Negative controls showing significant effects

• - Symptoms: Non-targeting controls (NTC) or safe-targeting sgRNAs in hit list
• Causes:

- Technical artifacts
- Random chance with large library
- Library design issues

• - Solutions:

- Review NTC performance; should not be systematically enriched/depleted
- If systematic, investigate technical issues
- Use NTC distribution to set empirical thresholds

---

References

Available in references/ directory:

• - (No reference files currently available for this skill)

External Resources:

• - AddGene CRISPR Libraries: https://www.addgene.org/crispr/libraries/
• DepMap Portal: https://depmap.org/portal/
• MAGeCK Documentation: https://sourceforge.net/p/mageck/wiki/Home/
• BAGEL Algorithm: https://github.com/hart-lab/bagel
• CRISPR Screen Analysis Best Practices: https://pubmed.ncbi.nlm.nih.gov/29651053/

---

Scripts

Located in scripts/ directory:

• - main.py - CRISPR screen analysis engine with QC, RRA, and hit identification

---

Common CRISPR Screen Types

Screen Type	Comparison	Expected Hits	Typical Duration
Viability	T14 vs T0	Essential genes depleted	10-14 days
Drug Resistance

Drug vs DMSO | Resistance genes enriched | 14-21 days | | Drug Sensitivity | Drug vs DMSO | Sensitizers depleted | 14-21 days | | Comparative | Cell A vs Cell B | Lineage-specific dependencies | 10-14 days | | Sensitizer | Drug A+B vs Drug A | Combination targets | 10-14 days |

Parameters

Parameter	Type	Default	Required	Description
INLINECODE20, INLINECODE21	string	-	Yes	sgRNA count matrix file
INLINECODE22, INLINECODE23

string | - | Yes | Sample annotation file | | --control | string | - | No | Control samples (comma-separated) | | --treatment, -t | string | - | No | Treatment samples (comma-separated) | | --output, -o | string | - | No | Output directory | | --fdr | float | 0.05 | No | FDR threshold |

Usage

Basic Usage

CODEBLOCK19

Risk Assessment

Risk Indicator	Assessment	Level
Code Execution	Python script executed locally	Low
Network Access

No external API calls | Low | | File System Access | Read count files, write results | Low | | Data Exposure | Processes genomic screening data | Medium | | PHI Risk | May contain cell line genetic info | Low |

Security Checklist

• - [x] No hardcoded credentials or API keys
• [x] No unauthorized file system access
• [x] Input validation for file paths
• [x] Output directory restricted
• [x] Error messages sanitized
• [x] Script execution in sandboxed environment

Prerequisites

CODEBLOCK20

Evaluation Criteria

Success Metrics

• - [x] Successfully loads sgRNA count matrices
• [x] Calculates QC metrics (Gini index, zero counts)
• [x] Performs RRA analysis
• [x] Identifies significant hits with FDR control

Test Cases

1. 1. Basic Analysis: Count matrix + samplesheet → QC metrics + hit list
2. RRA Analysis: Control vs Treatment → Ranked gene list with p-values
3. QC Metrics: Count data → Gini scores, zero sgRNA counts

Lifecycle Status

• - Current Stage: Active
• Next Review Date: 2026-03-09
• Known Issues: None
• Planned Improvements:

- Add MAGeCK integration - Support for multiple analysis methods - Enhanced visualization

---

Last Updated: 2026-02-09 Skill ID: 183 Version: 2.0 (K-Dense Standard)