bio-long-read-sequencing-isoseq-analysis

name	bio-long-read-sequencing-isoseq-analysis
description	Analyze PacBio Iso-Seq data for full-length isoform discovery and quantification. Use when characterizing transcript diversity or identifying novel splice variants.
tool_type	cli
primary_tool	IsoSeq3

Iso-Seq Analysis

IsoSeq3 Pipeline Overview

# Full pipeline: subreads -> HQ transcripts
# 1. CCS: Generate circular consensus sequences
# 2. Lima: Remove primers and demultiplex
# 3. Refine: Remove polyA and concatemers
# 4. Cluster: Group into isoforms
# 5. Polish: Generate high-quality consensus (optional with HiFi)

CCS Generation

# Generate CCS from subreads (skip if using HiFi reads)
ccs input.subreads.bam ccs.bam \
    --min-rq 0.9 \
    --min-passes 3 \
    --num-threads 32

# For HiFi reads, CCS is already done
# Start directly from HiFi reads

Primer Removal with Lima

# Iso-Seq specific primer removal
lima ccs.bam primers.fasta demux.bam \
    --isoseq \
    --peek-guess \
    --num-threads 16

# Output: demux.primer_5p--primer_3p.bam
# Lima reports also contain demux statistics

# Check lima report
cat demux.lima.summary

Primer File Format

>primer_5p
AAGCAGTGGTATCAACGCAGAGTACATGGG
>primer_3p
AAGCAGTGGTATCAACGCAGAGTAC

Refine Full-Length Reads

# Remove polyA tails and concatemers
isoseq3 refine demux.primer_5p--primer_3p.bam primers.fasta refined.bam \
    --require-polya \
    --min-polya-length 20

# Output: refined.bam (full-length non-chimeric reads)
# Also: refined.filter_summary.json

# Check refinement stats
cat refined.filter_summary.json | jq

Cluster Into Isoforms

# Cluster similar transcripts
isoseq3 cluster refined.bam clustered.bam \
    --verbose \
    --use-qvs \
    --num-threads 32

# Output files:
# - clustered.bam: Clustered transcripts
# - clustered.hq_transcripts.fasta: High-quality consensus
# - clustered.lq_transcripts.fasta: Low-quality consensus
# - clustered.cluster_report.csv: Cluster membership

Align to Reference

# Map HQ transcripts to reference genome
minimap2 -ax splice:hq \
    -uf \
    --secondary=no \
    reference.fa \
    clustered.hq_transcripts.fasta \
    | samtools sort -o aligned.bam

samtools index aligned.bam

# For downstream analysis
pbmm2 align reference.fa clustered.bam aligned.bam \
    --preset ISOSEQ \
    --sort

Collapse Redundant Isoforms

# Collapse mapped transcripts
isoseq3 collapse aligned.bam collapsed.gff

# Output:
# - collapsed.gff: Collapsed transcript models
# - collapsed.abundance.txt: Read counts per isoform
# - collapsed.group.txt: Isoform groupings

# Convert to GTF
gffread collapsed.gff -T -o collapsed.gtf

SQANTI3 Quality Control

# Classify isoforms against reference annotation
sqanti3_qc.py \
    clustered.hq_transcripts.fasta \
    reference.gtf \
    reference.fa \
    -o sqanti_output \
    --aligner_choice minimap2 \
    --cage_peak cage_peaks.bed \
    --polyA_motif_list polyA_motifs.txt \
    --cpus 16

# Key output files:
# - sqanti_output_classification.txt: Per-isoform metrics
# - sqanti_output_junctions.txt: Splice junction details
# - sqanti_output.params.txt: Run parameters

SQANTI3 Categories

Category	Code	Description
Full Splice Match	FSM	All junctions match reference
Incomplete Splice Match	ISM	Subset of reference junctions
Novel In Catalog	NIC	Novel combination of known junctions
Novel Not in Catalog	NNC	Contains novel junction
Antisense	AS	Overlaps gene on opposite strand
Genic	G	Within gene but no junction match
Intergenic	IR	Between genes
Fusion	FU	Spans multiple genes

SQANTI3 Filtering

# Filter artifacts using SQANTI3 rules
sqanti3_filter.py \
    sqanti_output_classification.txt \
    --isoforms clustered.hq_transcripts.fasta \
    --gtf collapsed.gtf \
    --faa predicted_proteins.faa \
    -o sqanti_filtered

# Custom filtering
python << 'EOF'
import pandas as pd

classification = pd.read_csv('sqanti_output_classification.txt', sep='\t')

# Keep FSM, ISM, NIC with evidence
keep = classification[
    (classification['structural_category'].isin(['full-splice_match', 'incomplete-splice_match', 'novel_in_catalog'])) &
    (classification['FL'] >= 2) &
    (classification['bite'] == 'FALSE')
]
keep['isoform'].to_csv('filtered_isoforms.txt', index=False, header=False)
EOF

Quantification with Pigeon

# PacBio's isoform quantification tool
pigeon classify \
    aligned.bam \
    reference.gtf \
    reference.fa \
    -o pigeon_output

# Produces count matrix and classification
pigeon report pigeon_output_classification.txt

TAMA for Annotation Merge

# Merge Iso-Seq with reference annotation
# First, convert to TAMA format
tama_format_convert.py \
    -i collapsed.gtf \
    -f gtf \
    -o isoseq.bed

# Create file list
echo -e "isoseq.bed\tcapped\t1\t1" > file_list.txt
echo -e "reference.bed\tcapped\t1\t2" >> file_list.txt

# Merge annotations
tama_merge.py \
    -f file_list.txt \
    -p merged \
    -a 50 \
    -z 50 \
    -m 10

Python Processing

import pysam
import pandas as pd

def parse_cluster_report(report_path):
    df = pd.read_csv(report_path)
    isoform_counts = df.groupby('cluster_id').size()
    return isoform_counts

def get_transcript_lengths(fasta_path):
    lengths = {}
    with pysam.FastxFile(fasta_path) as fh:
        for entry in fh:
            lengths[entry.name] = len(entry.sequence)
    return lengths

def summarize_isoseq(cluster_report, hq_fasta):
    counts = parse_cluster_report(cluster_report)
    lengths = get_transcript_lengths(hq_fasta)

    print(f'Total isoforms: {len(counts)}')
    print(f'Median support: {counts.median():.0f} reads')
    print(f'Mean length: {sum(lengths.values())/len(lengths):.0f} bp')

    return counts, lengths

Differential Isoform Usage

library(IsoformSwitchAnalyzeR)

# Import SQANTI3 results
switchList <- importIsoformExpression(
    isoformCountMatrix = 'counts.txt',
    isoformRepExpression = 'tpm.txt',
    designMatrix = design
)

# Add SQANTI3 annotations
switchList <- addORFfromFASTA(
    switchList,
    orfs = 'sqanti_corrected.fasta'
)

# Analyze switching
switchList <- isoformSwitchTestDEXSeq(switchList)

# Extract significant switches
sig_switches <- switchList$isoformSwitchAnalysis[
    switchList$isoformSwitchAnalysis$padj < 0.05,
]

Quality Metrics

Metric	Good	Acceptable	Poor
CCS passes	>3	2-3	<2
Full-length %	>80%	60-80%	<60%
Clustering rate	>90%	80-90%	<80%
FSM %	>50%	30-50%	<30%
Novel isoforms	10-30%	30-50%	>50% (suspect)

Troubleshooting

Issue	Possible Cause	Solution
Low full-length %	Primer issues	Check primer sequences
High concatemer %	Library prep	Increase SMRTbell cleanup
Few FSM	Poor reference	Use comprehensive GTF
High NNC %	Novel biology or artifacts	Validate with orthogonal data
Low clustering	High diversity	Reduce clustering stringency

Docker/Singularity

# Using PacBio Docker images
docker run -v /data:/data \
    quay.io/biocontainers/isoseq3:4.0.0--h9ee0642_0 \
    isoseq3 cluster /data/refined.bam /data/clustered.bam

# SQANTI3 in Singularity
singularity exec sqanti3.sif \
    sqanti3_qc.py input.fa ref.gtf ref.fa -o output

Related Skills

• long-read-sequencing/basecalling - ONT/PacBio basics
• rna-quantification/alignment-free-quant - Expression analysis
• genome-intervals/gtf-gff-handling - GTF/GFF handling
• differential-expression/timeseries-de - Differential isoform usage