Quality control for pooled CRISPR screens. Covers library representation, read distribution, replicate correlation, and essential gene recovery. Use when assessing screen quality before hit calling or diagnosing poor screen performance.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
Quality control for pooled CRISPR screens. Covers library representation, read distribution, replicate correlation, and essential gene recovery. Use when assessing screen quality before hit calling or diagnosing poor screen performance.
tool_type
python
primary_tool
pandas
CRISPR Screen Quality Control
Load Count Data
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
# Load MAGeCK count output
counts = pd.read_csv('screen.count.txt', sep='\t', index_col=0)
genes = counts['Gene']
count_matrix = counts.drop('Gene', axis=1)
print(f'sgRNAs: {len(count_matrix)}')
print(f'Genes: ')
()
{genes.nunique()}
print
f'Samples: {count_matrix.columns.tolist()}'
Library Representation
# Zero-count sgRNAs per sample
zero_counts = (count_matrix == 0).sum()
zero_pct = zero_counts / len(count_matrix) * 100print('Zero-count sgRNAs per sample:')
for sample, pct in zero_pct.items():
status = 'OK'if pct < 1else'WARNING'if pct < 5else'FAIL'print(f' {sample}: {pct:.2f}% [{status}]')
# Low-count sgRNAs (<30 reads)
low_counts = (count_matrix < 30).sum()
low_pct = low_counts / len(count_matrix) * 100print('\nLow-count sgRNAs (<30 reads):')
for sample, pct in low_pct.items():
print(f' {sample}: {pct:.2f}%')
Read Distribution (Gini Index)
defgini_index(x):
'''Calculate Gini index (0=perfect equality, 1=complete inequality)'''
x = np.sort(x[x > 0])
n = len(x)
cumx = np.cumsum(x)
return (n + 1 - 2 * np.sum(cumx) / cumx[-1]) / n
gini_values = count_matrix.apply(gini_index)
print('\nGini index per sample (lower is better, <0.2 ideal):')
for sample, gini in gini_values.items():
status = 'OK'if gini < 0.2else'WARNING'if gini < 0.3else'FAIL'print(f' {sample}: {gini:.3f} [{status}]')
# Correlation matrix
log_counts = np.log10(count_matrix + 1)
corr_matrix = log_counts.corr()
plt.figure(figsize=(8, 6))
sns.heatmap(corr_matrix, annot=True, cmap='RdYlBu_r', vmin=0.5, vmax=1,
square=True, fmt='.2f')
plt.title('Replicate Correlation (log10 counts)')
plt.tight_layout()
plt.savefig('qc_correlation.png', dpi=150)
# Check replicate pairsprint('\nReplicate correlations:')
for i, col1 inenumerate(count_matrix.columns):
for col2 in count_matrix.columns[i+1:]:
r = corr_matrix.loc[col1, col2]
status = 'OK'if r > 0.8else'WARNING'if r > 0.6else'FAIL'print(f' {col1} vs {col2}: r={r:.3f} [{status}]')
Essential Gene Recovery
# Load known essential genes (e.g., from Hart et al. or DepMap)
essential_genes = set(pd.read_csv('essential_genes.txt', header=None)[0])
nonessential_genes = set(pd.read_csv('nonessential_genes.txt', header=None)[0])
# Load MAGeCK results
results = pd.read_csv('screen.gene_summary.txt', sep='\t')
# Check recovery in T0 vs later timepoint
present_essential = results[results['id'].isin(essential_genes)]
present_nonessential = results[results['id'].isin(nonessential_genes)]
# ROC-like analysisfrom sklearn.metrics import roc_auc_score
y_true = results['id'].isin(essential_genes).astype(int)
y_score = -results['neg|score'] # More negative = more essentialif y_true.sum() > 0:
auc = roc_auc_score(y_true, y_score)
print(f'\nEssential gene recovery AUC: {auc:.3f}')
status = 'EXCELLENT'if auc > 0.9else'GOOD'if auc > 0.8else'FAIR'if auc > 0.7else'POOR'print(f'Status: {status}')
sgRNA Performance
# sgRNAs per gene
sgrnas_per_gene = genes.value_counts()
print(f'\nsgRNAs per gene: mean={sgrnas_per_gene.mean():.1f}, min={sgrnas_per_gene.min()}, max={sgrnas_per_gene.max()}')
# Check for genes with few sgRNAs
few_sgrnas = sgrnas_per_gene[sgrnas_per_gene < 3]
iflen(few_sgrnas) > 0:
print(f'WARNING: {len(few_sgrnas)} genes have <3 sgRNAs')
Sample Normalization Check
# Total reads per sample
total_reads = count_matrix.sum()
print('\nTotal reads per sample:')
for sample, total in total_reads.items():
print(f' {sample}: {total:,}')
# Check for major imbalances
cv = total_reads.std() / total_reads.mean()
print(f'\nCoefficient of variation: {cv:.3f}')
if cv > 0.5:
print('WARNING: Large variation in sequencing depth')
