test/create_test_data.sh

# ── Subset full HCC1143 BAMs to chr17 for pipeline development ───────────────
# Input:  full WES BAMs (wherever you downloaded them)
# Output: hcc1143_T_clean.bam / hcc1143_N_clean.bam (chr17 only)
#
# Why chr17: TP53 is on chr17 — the known HCC1143 driver mutation
# we use as our validated anchor epitope. Subsetting to chr17 keeps
# the test data small (~50MB vs ~23GB) while retaining the mutation
# we actually care about.
# ─────────────────────────────────────────────────────────────────────────────

TUMOR_FULL=~/melanoma-pipeline/data/full/hcc1143_T_clean.bam
NORMAL_FULL=~/melanoma-pipeline/data/full/hcc1143_N_clean.bam
OUT=~/melanoma-pipeline/data/test

mkdir -p $OUT

# Subset tumor to chr17
# -b: output BAM format, -h: include header, -@ 8: 8 threads
samtools view \
    -b \
    -h \
    -@ 8 \
    $TUMOR_FULL \
    chr17 \
    -o $OUT/tumor_chr17.bam

# Index tumor subset
samtools index \
    -@ 8 \
    $OUT/tumor_chr17.bam \
    $OUT/tumor_chr17.bai

# Subset normal to chr17
samtools view \
    -b \
    -h \
    -@ 8 \
    $NORMAL_FULL \
    chr17 \
    -o $OUT/normal_chr17.bam

# Index normal subset
samtools index \
    -@ 8 \
    $OUT/normal_chr17.bam \
    $OUT/normal_chr17.bai

# Verify read counts
echo "Tumor chr17 reads:"
samtools flagstat $OUT/tumor_chr17.bam

echo "Normal chr17 reads:"
samtools flagstat $OUT/normal_chr17.bam


# ── Extract HLA reads from normal BAM for OptiType ───────────────────────────
# Always use normal BAM for HLA typing — tumor DNA can have LOH
# (Loss of Heterozygosity) at HLA loci which gives wrong alleles.
#
# Three-bucket strategy to avoid missing HLA reads due to ALT contigs:
#   Bucket 1 — primary MHC region on chr6
#   Bucket 2 — ALT contigs (reads that overflowed from primary chr6 assembly)
#   Bucket 3 — unmapped reads (too divergent to map anywhere in hg38)
# All three are merged and remapped against OptiType's HLA-specific reference.
# ─────────────────────────────────────────────────────────────────────────────

OUT_HLA=~/melanoma-pipeline/data/hla
mkdir -p $OUT_HLA

# Bucket 1 — primary MHC region on chr6
# 28510120-33480577 = full MHC locus including HLA-A, B, C and flanking genes
samtools view \
    -b \
    -h \
    -@ 8 \
    $NORMAL_FULL \
    chr6:28510120-33480577 \
    -o $OUT_HLA/hla_chr6_primary.bam

# Bucket 2 — ALT contigs
# Reads from patients with HLA alleles too divergent for the primary reference
# — most common in non-European populations. Dynamically finds all ALT/HLA
# contigs in the BAM header so this works regardless of hg38 build variant.
ALT_CONTIGS=$(samtools view -H $NORMAL_FULL \
    | grep "^@SQ" \
    | grep -iE "_alt|HLA" \
    | awk '{print $2}' \
    | sed 's/SN://')

if [ -n "$ALT_CONTIGS" ]; then
    ALT_TMP=$(mktemp -d)
    while IFS= read -r contig; do
        safe_name="${contig//\//_}"
        samtools view -b -h -@ 8 $NORMAL_FULL "$contig" -o "$ALT_TMP/${safe_name}.bam"
    done <<< "$ALT_CONTIGS"
    samtools merge -f -@ 8 $OUT_HLA/hla_alt_contigs.bam "$ALT_TMP"/*.bam
    rm -rf "$ALT_TMP"
else
    # No ALT contigs — create an empty BAM so downstream merge still works
    samtools view -b -h $NORMAL_FULL -o $OUT_HLA/hla_alt_contigs.bam /dev/null 2>/dev/null || \
        samtools view -H $NORMAL_FULL -b -o $OUT_HLA/hla_alt_contigs.bam
fi

# Bucket 3 — unmapped reads
# Small fraction of BAM but catches rare/unusual alleles that couldn't
# map anywhere in hg38 due to extreme divergence from the reference
# -f 4: flag 4 = read unmapped
samtools view \
    -b \
    -f 4 \
    -@ 8 \
    $NORMAL_FULL \
    -o $OUT_HLA/hla_unmapped.bam

# Merge all three buckets into one candidate pool
samtools merge \
    -f \
    -@ 8 \
    $OUT_HLA/hla_candidates.bam \
    $OUT_HLA/hla_chr6_primary.bam \
    $OUT_HLA/hla_alt_contigs.bam \
    $OUT_HLA/hla_unmapped.bam

# Convert merged BAM to paired FASTQ for remapping
# Sort by name first (-n) so read pairs are adjacent
samtools sort -n -@ 8 $OUT_HLA/hla_candidates.bam \
    | samtools fastq \
    -1 $OUT_HLA/hla_candidates_R1.fastq \
    -2 $OUT_HLA/hla_candidates_R2.fastq \
    -0 /dev/null \
    -s /dev/null

# Remap against OptiType's HLA-specific reference
# This is the critical step — reads that were on ALT contigs or unmapped
# now find their correct HLA allele sequence. The HLA reference contains
# sequences for thousands of known alleles across all populations.
# hla_reference_dna.fasta is bundled inside the fred2/optitype Docker image
# at /usr/local/bin/data/hla_reference_dna.fasta
# Extract it once and reuse:
#   docker run --rm fred2/optitype cat /usr/local/bin/data/hla_reference_dna.fasta \
#       > ~/melanoma-pipeline/reference/hla_reference_dna.fasta
HLA_REF=~/melanoma-pipeline/reference/hla_reference_dna.fasta

# Index HLA reference if not already done
if [ ! -f "${HLA_REF}.bwt" ]; then
    echo "Indexing HLA reference..."
    bwa index $HLA_REF
fi

bwa mem \
    -t 8 \
    $HLA_REF \
    $OUT_HLA/hla_candidates_R1.fastq \
    $OUT_HLA/hla_candidates_R2.fastq \
    | samtools sort -n -@ 8 \
    | samtools fastq \
    -F 4 \
    -1 $OUT_HLA/hla_fished_R1.fastq \
    -2 $OUT_HLA/hla_fished_R2.fastq \
    -0 /dev/null \
    -s /dev/null

# hla_fished_R1/R2.fastq are now ready to pass to OptiType (Step 3 of the pipeline)
echo "HLA read extraction complete."
echo "Fished read pairs: $(( $(wc -l < $OUT_HLA/hla_fished_R1.fastq) / 4 ))"
echo "Ready for OptiType: $OUT_HLA/hla_fished_R1.fastq + hla_fished_R2.fastq"