Bio Splicing Qc

"Assess RNA-seq quality for splicing analysis" -> Evaluate junction saturation, junction novelty rate, splice site strength, and read coverage.

• Python/CLI: junction_saturation.py, junction_annotation.py (RSeQC)
• Python: maxentpy for splice site scoring, pysam for junction read counting

# RSeQC junction saturation (check sequencing depth)
# Note: -s flag removed in RSeQC v3.0
junction_saturation.py \
    -i sample.bam \
    -r annotation.bed \
    -o sample_junc_sat

import subprocess
import matplotlib.pyplot as plt
import pandas as pd

# Run junction saturation for multiple samples
samples = ['sample1.bam', 'sample2.bam', 'sample3.bam']

for sample in samples:
    subprocess.run([
        'junction_saturation.py',
        '-i', sample,
        '-r', 'annotation.bed',
        '-o', sample.replace('.bam', '_junc_sat')
    ], check=True)

# Parse results and check for plateau
# Plateau indicates sufficient depth for splicing analysis
# If curves still rising, may need more sequencing depth

Junction Annotation

Goal: Classify observed junctions as known, partially novel, or completely novel relative to annotation.

Approach: Run RSeQC junction annotation and compute the ratio of known to novel junctions as a data quality indicator.

# Classify junctions as known, partial novel, or complete novel
junction_annotation.py \
    -i sample.bam \
    -r annotation.bed \
    -o sample_junc_annot

import pandas as pd

# Analyze junction annotation results
junc_stats = pd.read_csv('sample_junc_annot.junction.xls', sep='\t')

# Calculate junction type proportions
total = junc_stats['total_splicing_events'].sum()
known = junc_stats[junc_stats['annotation'] == 'known']['total_splicing_events'].sum()
novel = total - known

print(f'Known junctions: {known/total:.1%}')
print(f'Novel junctions: {novel/total:.1%}')

# High novel junction rate may indicate:
# - Incomplete annotation
# - Mapping artifacts
# - Interesting biology (cancer, tissue-specific)

Splice Site Strength Scoring

Goal: Score donor and acceptor splice sites to identify weak or cryptic splice sites.

Approach: Use MaxEntScan (via maxentpy) to compute information-theoretic scores for 5' and 3' splice site sequences.

# MaxEntScan scoring via maxentpy
# 5'ss (donor): typical score 8-10 bits
# 3'ss (acceptor): typical score 8-12 bits

from maxentpy import maxent
from maxentpy.maxent import score5, score3

# Score 5' splice site (9bp: 3 exon + 6 intron)
donor_seq = 'CAGGTAAGT'  # Consensus: CAG|GTAAGT
score_5ss = score5(donor_seq)
print(f"5'ss score: {score_5ss:.2f}")

# Score 3' splice site (23bp: 20 intron + 3 exon)
acceptor_seq = 'TTTTTTTTTTTTTTTTTTTTCAG'
score_3ss = score3(acceptor_seq)
print(f"3'ss score: {score_3ss:.2f}")

# Weak splice sites (score < 5) may indicate:
# - Alternative/cryptic splice sites
# - Annotation errors
# - Regulatory splice sites

Junction Read Coverage

Goal: Profile the distribution of junction-spanning read counts across all splice sites in a BAM file.

Approach: Parse CIGAR strings for N operations (splice junctions) using pysam and tally reads per junction.

import pysam
import pandas as pd

def count_junction_reads(bam_path, min_overhang=8):
    '''Count junction-spanning reads per splice site.'''
    bam = pysam.AlignmentFile(bam_path, 'rb')
    junction_counts = {}

    for read in bam.fetch():
        if read.is_unmapped:
            continue

        # Check CIGAR for splice junctions (N operation)
        ref_pos = read.reference_start
        for op, length in read.cigartuples:
            if op == 3:  # N = splice junction
                junction = (read.reference_name, ref_pos, ref_pos + length)
                junction_counts[junction] = junction_counts.get(junction, 0) + 1
            if op in [0, 2, 3]:  # M, D, N consume reference
                ref_pos += length

    bam.close()
    return junction_counts

# Analyze coverage distribution
junctions = count_junction_reads('sample.bam')
counts = list(junctions.values())

print(f'Total junctions: {len(junctions)}')
print(f'Junctions >= 10 reads: {sum(1 for c in counts if c >= 10)}')
print(f'Junctions >= 20 reads: {sum(1 for c in counts if c >= 20)}')

Quality Thresholds

Troubleshooting Low Detection

Related Skills

• splicing-quantification - Quantify after QC passes
• read-alignment/star-alignment - Alignment quality affects junctions
• read-qc/quality-reports - General sequencing QC