Glossary
Technical terms used throughout the documentation.
Metrics
| Term |
Definition |
| DP |
Total depth — all mapped, quality-filtered reads overlapping the variant anchor position, including REF, ALT, and 'neither'. DP ≥ RD + AD. |
| VAF |
Variant Allele Frequency — AD / (RD + AD) |
| Strand Bias |
Fisher's exact test for read direction imbalance |
| AD |
Alternate allele depth (supporting reads) |
| RD |
Reference allele depth |
| Term |
Definition |
| VCF |
Variant Call Format — standard variant file |
| MAF |
Mutation Annotation Format — annotation-rich variant file |
| BAM |
Binary Alignment Map — compressed alignment file |
| BAI |
BAM Index — enables random access to BAM |
| FASTA |
Reference genome sequence file |
| FAI |
FASTA Index — enables random access to FASTA |
Quality Scores
| Term |
Definition |
| MAPQ |
Mapping Quality — confidence in read alignment |
| BASEQ |
Base Quality — confidence in base call |
cfDNA Terms
| Term |
Definition |
| cfDNA |
Cell-free DNA — circulating DNA in plasma |
| ctDNA |
Circulating tumor DNA — tumor-derived cfDNA |
| Duplex |
Reads from both strands of original molecule |
Alignment Backends
| Term |
Definition |
| PairHMM |
Default Phase 3 alignment backend for all modes. Uses pair hidden Markov model with base quality probabilities for probabilistic alignment scoring. Selected via --alignment-backend pairhmm. |
| Smith-Waterman (SW) |
Alternative Phase 3 alignment backend. Uses edit-distance scoring to align reads against REF and ALT haplotypes. Confident call requires ≥2 score margin. Selected via --alignment-backend sw. |
| LLR |
Log-Likelihood Ratio — PairHMM confidence metric. LLR = ln(P(read|ALT)) - ln(P(read|REF)). Default threshold: 2.3 (≈ ln(10), i.e., 10:1 odds). |
Fragment Metrics
| Term |
Definition |
| DPF |
Fragment depth — total unique fragments (including discarded) |
| RDF |
Reference fragment count — fragments resolved to REF |
| ADF |
Alternate fragment count — fragments resolved to ALT |
| Fragment consensus |
Quality-weighted method to resolve R1/R2 disagreements within a fragment |
Validation Status
| Status |
Meaning |
| PASS |
Variant validated against reference, ready for counting |
| PASS_WARN_REF_CORRECTED |
REF allele ≥90% match; corrected to FASTA REF |
| PASS_WARN_HOMOPOLYMER_DECOMP |
Variant spans a homopolymer; dual-counted with corrected allele (corrected won) |
| PASS_MULTI_ALLELIC |
Variant overlaps another variant at the same locus; sibling ALT exclusion active |
| REF_MISMATCH |
REF allele does not match the reference genome at the stated position |
| FETCH_FAILED |
Could not fetch the reference region (chromosome not found, etc.) |
RNA Terms
| Term |
Definition |
| Sense strand |
mRNA coding strand; matches gene transcription direction |
| Antisense strand |
Non-coding strand; template for transcription |
| A-to-I editing |
Post-transcriptional RNA modification by ADAR enzymes; adenosine → inosine (read as guanosine) |
| NH tag |
BAM tag indicating number of reported alignments for a read |
| Splice junction |
CIGAR N operation representing an intron skip in RNA-seq reads |
| REDIportal |
Database of known human A-to-I RNA editing sites |
Engine Terms
| Term |
Definition |
| BAQ |
Base Alignment Quality — heuristic quality downgrade near indels to prevent false calls |
| UMI |
Unique Molecular Identifier — barcode for molecule-level deduplication |
| Genomic binning |
Variant grouping by chromosome for cache-efficient counting (count_bam_binned) |
| Parity test |
Test verifying count_bam and count_bam_binned produce identical results |
| Complex Del+SNV |
A deletion-format variant (len(REF) > len(ALT) == 1) where the anchor base also substitutes (alt[0] ≠ ref[0]). Examples: GC→T, AG→T. Routed to check_complex instead of check_deletion. |
| has_nearby_length_match |
Flag set in check_deletion when a windowed Del matches in length (≥5bp) but fails S3 sequence validation. Triggers Phase 3 SW fallback to handle BWA left-alignment shifts. |
| left-alignment shift |
BWA repositions an indel's anchor to the leftmost equivalent position. For partial-repeat regions, this places the anchor several bases left of where the CIGAR D operation appears in reads. |
| is_worth_realignment |
Predicate in check_complex returning true when a read's CIGAR contains indel evidence in the variant window. If false, M-block anchor coverage check classifies clean REF reads for deletion-direction variants. |
abbreviations