usage: intronIC [-h] [--version] [--quiet] [--debug] [--config CONFIG_PATH]
                [--generate-config] [-n SPECIES_NAME] [-o OUTPUT_DIR]
                [-g GENOME] [-a ANNOTATION] [-b BED] [-q SEQUENCE_FILE]
                [--model MODEL] [--pretrained-model PRETRAINED_MODEL]
                [--normalizer-mode {human,adaptive,auto}]
                [--species-prior PRIOR] [--load-normalizer PATH]
                [--save-normalizer]
                [-f {cds,exon,both}] [--min-intron-len MIN_INTRON_LEN] [-i]
                [-v] [-d] [--flank-len FLANK_LEN] [--no-nc-ss-adjustment]
                [-t THRESHOLD] [--no-nc] [--pseudocount PSEUDOCOUNT]
                [--no-ignore-nc-dnts] [--five-score-coords START END]
                [--bp-region-coords START END]
                [--three-score-coords START END]
                [-p PROCESSES] [--cv-processes CV_PROCESSES] [--streaming]
                [--clean-names] [--no-clean-names] [-u] [--no-abbreviate]
                [--abbreviate-filenames] [--no-headers] [--seed SEED]
                {train,classify,extract,test} ...

intronIC: Intron classification and extraction tool

positional arguments:
  {train,classify,extract,test}
                        Command to run (default: classify if not specified)
    train               Train a classifier on reference data only (no
                        genome/annotation needed)
    classify            Extract and classify introns from genome/annotation
    extract             Extract intron sequences without classification
    test                Run a quick installation test with bundled test data

options:
  -h, --help            show this help message and exit
  --version             show program's version number and exit
  --quiet               Suppress non-essential output
  --debug               Enable debug logging
  --config CONFIG_PATH  Path to configuration file (auto-loads from standard
                        paths if not specified)
  --generate-config     Generate configuration file template and exit
  -n SPECIES_NAME, --species-name SPECIES_NAME, --species_name SPECIES_NAME
                        Species name for output files (e.g., homo_sapiens)
  -o OUTPUT_DIR, --output-dir OUTPUT_DIR, --output_dir OUTPUT_DIR
                        Output directory (default: current directory)
  --normalizer-mode {human,adaptive,auto}, --normalizer_mode {human,adaptive,auto}
                        Normalizer mode for pretrained model classification
                        (default: auto): human: Use scaler from training
                        species (recommended for U12-absent genomes) adaptive:
                        Refit scaler on experimental data (experimental, may
                        cause FPs in U12-free species) auto: Use human if
                        available in model, otherwise adaptive
  --species-prior PRIOR
                        Expected U12 prior for target species (0 to 1).
                        Adjusts classification probabilities via Bayes rule to
                        account for different U12 base rates. By default
                        (None), no adjustment is applied - classifier uses raw
                        probabilities from the model trained on human data
                        without making assumptions about target species.
                        Recommended values when adjustment is desired: -
                        0.005: Human-like species (similar to training data) -
                        1e-6: U12-absent species (C. elegans, many fungi) -
                        1e-4: U12-poor species Lower values reduce false
                        positives in U12-free lineages.
  --load-normalizer PATH
                        Load a saved normalizer (from a previous run, or any
                        compatible scaler pickle) and use it instead of
                        fitting one. Honored in both streaming and in-memory
                        modes. When passed, this overrides whatever scaler
                        the model bundle ships with and skips the per-input
                        adaptive fit. Typical use: pass --save-normalizer on
                        a first full-genome run, then --load-normalizer
                        <path>.normalizer.pkl on subsequent runs over subsets
                        of the same genome to keep z-scores consistent.
  --save-normalizer     Save the fitted normalizer for future runs (adaptive
                        mode only). Use this on your first full-genome run for
                        a species to establish a reference normalization.
                        Future runs can use --load-normalizer to reuse this
                        normalization. Saved to <output_prefix>.normalizer.pkl

input selection:
  Choose one mode: (1) -g + -a for annotation, (2) -g + -b for BED, or (3) -q for sequences

  -g GENOME, --genome GENOME
                        Path to genome FASTA file (required with -a or -b)
  -a ANNOTATION, --annotation ANNOTATION
                        Path to GFF3/GTF annotation file (requires -g)
  -b BED, --bed BED     Path to BED file with intron coordinates (requires -g)
  -q SEQUENCE_FILE, --sequence-file SEQUENCE_FILE, --sequence_file SEQUENCE_FILE
                        Path to pre-extracted intron sequences (.iic format)

model options:
  --model MODEL         Path to pretrained model (.model.pkl)
  --pretrained-model PRETRAINED_MODEL, --pretrained_model PRETRAINED_MODEL
                        (Deprecated: use --model) Path to pretrained model

extraction parameters:
  -f {cds,exon,both}, --feature {cds,exon,both}
                        Feature type to extract from (default: both)
  --min-intron-len MIN_INTRON_LEN, --min_intron_len MIN_INTRON_LEN
                        Minimum intron length (default: 30)
  -i, --allow-multiple-isoforms, --allow_multiple_isoforms
                        Include non-longest isoforms
  -v, --no-intron-overlap, --no_intron_overlap
                        Exclude overlapping introns
  -d, --include-duplicates, --include_duplicates
                        Include duplicate coordinate introns
  --flank-len FLANK_LEN, --flank_len FLANK_LEN
                        Exonic flank length (default: 100)
  --no-nc-ss-adjustment, --no_nc_ss_adjustment
                        Disable U12 boundary correction

scoring parameters:
  -t THRESHOLD, --threshold THRESHOLD
                        U12 probability threshold 0-100 (default: 90)
  --no-nc, --no_nc      Exclude non-canonical introns from scoring
  --pseudocount PSEUDOCOUNT
                        PWM pseudocount (default: 0.0001)
  --no-ignore-nc-dnts, --no_ignore_nc_dnts
                        Include terminal dinucleotides in non-canonical
                        scoring
  --five-score-coords START END, --five_score_coords START END
                        5' splice site region (default: -3 9)
  --bp-region-coords START END, --bp_region_coords START END
                        Branch point region (default: -55 -5)
  --three-score-coords START END, --three_score_coords START END
                        3' splice site region (default: -6 4)

performance options:
  -p PROCESSES, --processes PROCESSES
                        Parallel processes for scoring (default: 1)
  --cv-processes CV_PROCESSES, --cv_processes CV_PROCESSES
                        Processes for cross-validation (training only, default: same as -p)
  --in-memory, --no-streaming
                        Use in-memory mode: load full genome into memory. Higher memory
                        usage but may be slightly faster for very small genomes.
  --streaming           Use streaming mode (default): stores sequences in temp on-disk
                        storage, keeps only scoring motifs in memory. Roughly halves
                        peak RSS vs --in-memory (e.g., ~5 GB vs ~10 GB for full human at
                        -p 6). Bit-identical to in-memory.

output options:
  --clean-names, --clean_names
                        Remove "transcript:" and "gene:" prefixes (default:
                        True)
  --no-clean-names, --no_clean_names
                        Keep ID prefixes
  -u, --uninformative-naming, --uninformative_naming
                        Use simple naming scheme
  --no-abbreviate, --no_abbreviate, --na
                        Use full species name in outputs
  --abbreviate-filenames, --abbreviate_filenames, --afn
                        Abbreviate species name in filenames
  --no-headers, --no_headers
                        Omit column headers from output files (default:
                        include headers)

advanced options:
  --seed SEED           Random seed (default: 42)

Examples:
  # Test installation with bundled data
  intronIC test -p 4

  # Train a model on reference data (no genome or annotation required)
  intronIC train -n homo_sapiens

  # Classify introns with pretrained model (uses streaming mode by default)
  intronIC classify -g genome.fa -a annotation.gff -n species --model species.model.pkl

  # Extract intron sequences without classification
  intronIC extract -g genome.fa -a annotation.gff -n species

  # Parallel processing for faster analysis (streaming mode scales efficiently)
  intronIC classify -g genome.fa -a annotation.gff -n species -p 8

  # Use in-memory mode for very small genomes
  intronIC classify -g genome.fa -a annotation.gff -n species --no-streaming

  # Backward compatible (no subcommand = classify)
  intronIC -g genome.fa -a annotation.gff -n species --model species.model.pkl