Current extract_cutesv_indels.py Feature Extraction (Implementation Summary)

The script builds per-variant evidence from a BAM and VCF in two passes.
First, it precomputes split-read (inter-alignment) signatures per chromosome by finding reads with multiple alignments and applying distance/overlap heuristics to adjacent segments.
Then, for each VCF variant, it scans CIGAR operations in overlapping reads to collect intra-alignment signatures (D for deletions, I for insertions).

For each variant, it writes a BED-like file with rows: CHROMOSOME, START, END, READ, TYPE where TYPE is one of:

• INTRA_DEL / INTRA_INS
• INTER_DEL / INTER_INS

It also generates:

• a signature plot (variant interval plus read-support intervals), and
• an encoded matrix image where per-base support is represented as:
  • 0 = no support
  • 1 = intra-alignment support
  • 2 = inter-alignment support

Concrete CIGAR Example (What Gets Recorded)

Assume:

• variant from VCF: chr21:1005-1025 (so start=1005, length=20)
• extension used by the script: 50 bp
• one read: readA
• read alignment start: 990
• simulated CIGAR: 10M6D20M4I10M

The script tracks a reference cursor current_read_position, starting at 990.

If --type DEL

Processing:

• 10M: cursor 990 -> 1000
• 6D: deletion spans 1000-1006; this overlaps the variant window (955-1075), so it writes:

chr21    1000    1006    readA    INTRA_DEL

• 20M: cursor 1006 -> 1026
• 4I: ignored in DEL mode
• 10M: cursor 1026 -> 1036

Result for this read in DEL mode: one INTRA_DEL record.

If --type INS

Processing:

• 10M: cursor 990 -> 1000
• 6D: cursor 1000 -> 1006 (no INS record)
• 20M: cursor 1006 -> 1026
• 4I: insertion tested at 1026-1030; overlaps variant window, so it writes:

chr21    1026    1030    readA    INTRA_INS

• 10M: cursor continues from updated value

Result for this read in INS mode: one INTRA_INS record.

In both modes, any overlapping split-read signature found in the precomputed inter-alignment pass is appended as INTER_DEL or INTER_INS rows in the same per-variant BED file.