Module 2: Key Long-Read Data Formats: FAST5, POD5, FASTQ & SAM/BAM Specifications

1. Evolution of Raw Signal Containers: FAST5 to POD5

Unlike Illumina sequencers which output raw pixel intensities that are discarded after real-time basecalling, Oxford Nanopore sequencers preserve the raw physical measurements so users can re-basecall data as newer neural network architectures emerge.

2. The Anatomy of a POD5 File

Each read inside a POD5 file contains discrete hardware and signal metadata:

# Inspecting raw signal traces directly from POD5 using Python
import pod5

with pod5.Reader("sample_reads.pod5") as reader:
    for read in reader.reads():
        print(f"Read ID: {read.read_id}")
        print(f"Channel: {read.channel} | Well: {read.well}")
        print(f"Sample Rate: {read.sample_rate} Hz")
        print(f"Signal Length: {len(read.signal)} raw pA measurements")
        
        # Raw picoampere calculation from 16-bit integer ADC values:
        # pA = (raw_adc + offset) * range / digitisation
        signal_pA = read.signal_pa
        print(f"Mean current: {signal_pA.mean():.2f} pA, StdDev: {signal_pA.std():.2f} pA")
        break

3. SAM/BAM Tags Unique to Long Reads

When long reads are aligned or basecalled with modifications, specific SAM tags convey epigenetic and structural information:

• MM:Z: (Base Modifications): Encodes the positions of modified bases (e.g. C+m? for 5mC) relative to canonical bases.
• ML:B:C (Modification Probabilities): Array of unsigned 8-bit integers ($0 - 255$) expressing the Phred-scaled probability that each indicated base is modified ($P = \text{val} / 256$).
• SA:Z: (Supplementary Alignment): Crucial for structural variant split-read analysis. Lists chimeric alignment segments: rname,pos,strand,CIGAR,mapQ,NM;.
• HP:i: and PS:i: (Haplotype & Phase Set): Generated by phasing tools like WhatsHap or LongPhase to assign individual reads to maternal ($1$) or paternal ($2$) chromosomes within a phase set block.