CallGenes

Script: callgenes.sh Package: prok Class: CallGenes.java

Finds orfs and calls genes in unspliced prokaryotes. This includes bacteria, archaea, viruses, and mitochondria. Can also predict 16S, 18S, 23S, 5S, 5.8S, LSU, ITS, and tRNAs.

Basic Usage

callgenes.sh in=contigs.fa out=calls.gff outa=aminos.faa out16S=16S.fa

The only required parameter is in. All other parameters are optional and have default values.

Parameters

CallGenes provides gene calling capabilities with parameters organized by function.

File parameters

in=<file>: A fasta file; the only required parameter. Can also use infna, fnain, fna, or ref.
out=<file>: Output GFF file for gene annotations.
outa=<file>: Amino acid output file (FASTA format). Also accepts outamino, aminoout, outaa, aaout, or amino.
out16s=<file>: 16S rRNA output file (FASTA format). Also accepts 16sout.
out18s=<file>: 18S rRNA output file (FASTA format). Also accepts 18sout.
outits=<file>: ITS1, ITS2, and combined ITS sequence output (FASTA format). Implies its=t. Also accepts itsout.
model=<file>: A PGM file or comma-delimited list. If unspecified, a default model will be used. Can use auto or default to use the built-in model. Also accepts pgm or gm.
stats=stderr: Stats output destination (may be stderr, stdout, a file, or null). Also accepts outstats.
hist=null: Gene length histogram output file. Also accepts outhist, lengthhist, lhist, or genehist.
compareto=: Optional reference GFF file to compare with the gene calls. 'auto' will name it based on the input file name.
ingff=: Input GFF file for additional gene annotations. Also accepts gffin.

Formatting parameters

json=false: Print stats in JSON format. Also accepts json_out.
binlen=21: Histogram bin length. Also accepts binlength or histdiv.
bins=1000: Maximum number of histogram bins.
pz=f: (printzero) Print histogram lines with zero count.
ordered=false: Maintain output order matching input order (may reduce performance).
trd=f: (trimreaddescription) Set to true to trim read headers after the first whitespace. Necessary for IGV.

Genetic-code parameters

transl_table=: Explicit NCBI table 4, 11, or 25 for CDS calling and translation. Omitted preserves historical ATG/GTG/TTG starts and translation. Explicit 11 uses its full NCBI initiation set. The selected code also governs start/stop candidates during multipass training.
codefile=: Alternative complete TSV: codon, amino_acid, start header; 64 unique DNA codons, one canonical residue or *, start 0/1. Mutually exclusive with transl_table. Real terminal stops are omitted from proteins; truncated ends retain complete codons. Initial M requires a known complete allowed start. Synthetic contig-edge starts retain their elongation residue. Existing PGMs are not retrained by selecting a code; accuracy with a reassigned code requires separate evaluation. Explicit codes currently reject recode/detranslate modes.

Gene calling parameters

minlen=60: Don't call genes shorter than this length. Also accepts minlength. Default 60 in shell, but Java uses 80.
maxoverlapss=80: Maximum overlap allowed between genes on the same strand. Also accepts overlapss, overlapsamestrand, moss, or maxOverlapSameStrand.
maxoverlapos=110: Maximum overlap allowed between genes on opposite strands. Also accepts overlapos, overlapoppositestrand, moos, or maxOverlapOppositeStrand.
minStartScore=-0.10: Minimum score required for start codons.
minStopScore=-0.5: Minimum score required for stop codons.
minInnerScore=0.02: Minimum score for inner kmers. Also accepts minKmerScore.
minOrfScore=50: Minimum overall ORF score. Higher values increase specificity but reduce sensitivity. Also accepts minScore.
minAvgScore=0.08: Minimum average kmer score across the ORF.
breakLimit: Maximum number of consecutive low-scoring kmers allowed before breaking an ORF.
clearcutoffs=false: Clear all score cutoffs to very permissive values (sets breakLimit to 9999 and minOrfScore, minAvgScore, minInnerScore, minStopScore, and minStartScore to -9999). Also accepts clearfilters.

Processing parameters

merge=f: For paired reads, merge overlapping reads before calling genes.
ecco=f: Error-correct overlapping paired reads without merging.
passes=1: Number of passes for iterative model refinement. Also accepts 2pass or twopass for 2 passes.
translate=true: Set amino acid output mode to translation (default).
detranslate=f: Output canonical nucleotide sequences instead of amino acids. Also accepts retranslate.
recode=f: Re-encode nucleotide sequences over called genes using canonical codons, leaving non-coding regions unchanged.
plus=true: Process the plus strand.
minus=true: Process the minus strand.

rRNA detection parameters

setbias16s: Set detection bias for 16S rRNA genes.
setbias18s: Set detection bias for 18S rRNA genes.
setbias23s: Set detection bias for 23S rRNA genes.
setbias5s: Set detection bias for 5S rRNA genes.
setbiastRNA: Set detection bias for tRNA genes.
setbiasCDS: Set detection bias for protein-coding sequences.
min16SIdentity: Minimum identity threshold for 16S rRNA detection. Also accepts min16SId.
min18SIdentity: Minimum identity threshold for 18S rRNA detection. Also accepts min18SId.
min23SIdentity: Minimum identity threshold for 23S rRNA detection. Also accepts min23SId.
min5SIdentity: Minimum identity threshold for 5S rRNA detection. Also accepts min5SId.

rRNA alignment parameters

rrnafallback=f: For 16S, 18S, 23S, and 5S: after the universal rRNA consensus fails alignment, try later consensuses in shipped-file order. Default false preserves historical universal-only behavior.
align16s=: Enable 16S sequence alignment. Also accepts load16SSequence.
align18s=: Enable 18S sequence alignment. Also accepts load18SSequence.
align23s=: Enable 23S sequence alignment. Also accepts load23SSequence.
align5s=: Enable 5S sequence alignment. Also accepts load5SSequence.
16sstartslop: Start position tolerance for 16S detection. Also accepts ssustartslop.
16sstopslop: Stop position tolerance for 16S detection. Also accepts ssustopslop.
23sstartslop: Start position tolerance for 23S detection. Also accepts lsustartslop.
23sstopslop: Stop position tolerance for 23S detection. Also accepts lsustopslop.
5sstartslop: Start position tolerance for 5S detection.
5sstopslop: Stop position tolerance for 5S detection.

Kmer parameters

load16skmers: Load kmers for 16S detection. Also accepts load18skmers or loadssukmers.
load23skmers: Load kmers for 23S detection. Also accepts load28skmers or loadlsukmers.
load5skmers: Load kmers for 5S detection.
loadtrnakmers: Load kmers for tRNA detection.
longkmers: Enable all long kmer loading for rRNA and tRNA detection.
klong16s: Kmer length for 16S detection. Also accepts klong18s or klongssu.
klong23s: Kmer length for 23S detection. Also accepts klong28s or klonglsu.
klong5s: Kmer length for 5S detection.
klongtrna: Kmer length for tRNA detection.

Advanced scoring parameters

e1: ORF scoring parameter e1 (frame-specific penalty).
e2: ORF scoring parameter e2 (frame-specific penalty).
e3: ORF scoring parameter e3 (frame-specific penalty).
f1: ORF scoring parameter f1 (frame-specific bonus).
f2: ORF scoring parameter f2 (frame-specific bonus).
f3: ORF scoring parameter f3 (frame-specific bonus).
p0: Gene calling parameter p0 (probability adjustment).
p1: Gene calling parameter p1 (probability adjustment).
p2: Gene calling parameter p2 (probability adjustment).
p3: Gene calling parameter p3 (probability adjustment).
p4: Gene calling parameter p4 (probability adjustment).
p5: Gene calling parameter p5 (probability adjustment).
p6: Gene calling parameter p6 (probability adjustment).
q1: Gene calling parameter q1 (quality adjustment).
q2: Gene calling parameter q2 (quality adjustment).
q3: Gene calling parameter q3 (quality adjustment).
q4: Gene calling parameter q4 (quality adjustment).
q5: Gene calling parameter q5 (quality adjustment).
lookback: Lookback distance for both directions during gene calling.
lookbackplus: Lookback distance in the plus direction.
lookbackminus: Lookback distance in the minus direction.

Taxonomy parameters

taxonomy=t: Use QuickClade to classify the input and select a phylum-specific gene model for improved accuracy. Intended for isolates (one organism per file): a metagenome would be misclassified to a single phylum, so set taxonomy=f for mixed samples. Requires a QuickClade server; if it is unreachable, callgenes falls back to the general model with a warning (it never fails on this account). Also accepts tax.
percontig=f: Classify each contig separately (for metagenomes). Default is per-file (classify once for all contigs).
taxaddress=: QuickClade server address. Default: refseq. Also accepts taxserver.
local=f: Use QuickClade's LOCAL reference database (large multi-GB files under resources/) instead of the network server. This is the REVERSE of QuickClade's own default (which favors local, falling back to the server only if local files are missing) -- callgenes stays lightweight by default, and local=t opts INTO the heavy local files if present (falls back to the server, with a warning, if they're not). Also accepts localclade.
server=t: Alias for local=f (server=f is the same as local=t). With taxonomy=t and percontig=f, the detected domain, phylum, and which path (local/server) served the classification are written as "##Domain"/"##Phylum"/"##ClassificationSource" GFF header lines, and are available to an in-process Java caller via CallGenes.lastDetectedDomain()/lastDetectedPhylum()/lastClassificationSource(). Also accepts serverclade.

Experimental CM verification (default off; tRNA and 5S only)

cm=f: Opt into exact local-CYK best-interval search for small families around the candidate extent; retain caller endpoints and path scores. Each candidate gets its own best interval with overlap at least half the shorter of the candidate and CM hit. Existing candidate and borderline filters remain unchanged. Small-family defaults: +/-12 search, conditional tRNA-Sec, retry the full existing window, then add +/-64 on rejection. With euk18s=t, 18S still uses the same fast alignment as cm=f; no 18S covariance model is loaded or attached.
cmlocal=: Explicit t selects local-model CYK for the enabled small families; explicit f selects global scoring and disables omitted retries. Omitted: local for small families. Requires cm=t when true. Changes model begin/end scoring, not candidate windows or caller endpoints. Small families use exact interval search. Cutoffs are unchanged provisional model GA values, not calibrated operating points for local scoring. Diagnostic routes gain a _local marker.
cmlocalwide=: Defaults to true for enabled small-family local scoring; f disables. For small families, retry a failed local decision over the full existing padded window. Requires cm=t cmlocal=t. Spliced tRNA candidates also search the unspliced genomic ORF plus its existing padding; any mature-product pass is retained. No added padding, endpoint changes or 18S changes. Uses half-of-the-shorter overlap. Wide diagnostic routes contain _wide_local; all attempts retain their costs, with one selectedModel row across both windows.
cmlocalextend=: Defaults to 64 when local/wide retries are enabled, otherwise 0. Explicit 0 disables extension. After all local-window attempts reject a small-family candidate, retry once with this many additional bases per side of the full genomic window, clipped to the sequence boundaries. Requires cm=t cmlocal=t cmlocalwide=t. Existing passes, candidate endpoints, association and model cutoffs remain unchanged. Extended diagnostics use _extended_local and retain all costs.
cmtrnaextendedgreedy=t: Nonoverlapping local-hit association only for the extended tRNA retry. When omitted, enabled only with cm=t, local/wide scoring and positive cmlocalextend. Explicit f retains the earlier constrained-span control. With cmdiag, CM_GREEDY_HIT stderr rows retain each extracted hit.
cmsmallgreedy=: Defaults to true with cm=t and local small-family scoring. Explicit f disables this rule; cmtrnaextendedgreedy still independently controls the extended tRNA retry. Greedy association covers all tRNA/5S local windows, including spliced mature products and every retry. Association uses the sequence supplied to the CM; diagnostics retain genomic coordinates. Includes extended retries even if the narrower cmtrnaextendedgreedy flag is false. Requires cm=t cmlocal=t; Global CM mode is unchanged; disabled with cm=f.
cmmodeldir=: Directory containing RF00005.cm and RF01852.cm (tRNA and tRNA-Sec) and RF00001.cm (5S). RF01960 is not used by this caller. Any model passing its own cutoff admits a candidate; diagnostics select the highest-scoring passing model among those evaluated. Without this, look in resources/. Models are loaded only for enabled families and cm=t. Model GA cutoffs are provisional, not calibrated for this scoring mode. IUPAC residues use equal-weight averages of canonical emission bit scores; gaps are invalid.
cmwindowslack=12: Limit the search to the candidate extent plus this many bases per side, intersected with the original padded window. Applies to small-family CM verification; -1 restores the full window.
cmsecfallback=t: Evaluate RF01852 only if RF00005 fails its own cutoff. Set false to evaluate both tRNA models for every candidate.
cmmaxcells=100m: Maximum DP cells per model/window (one float score plus one int traceback choice per cell) for small-family CM verification. This is not a whole-JVM memory bound and does not apply to 18S.
cmdiag=: Optional TSV of every scored candidate, including rejected ones, source route, caller/window/CM extents, score and compute cost. All model attempts are retained; selectedModel marks one row per candidate. Unrestricted best hits diagnose neighboring loci. These candidates may still lose final gene-path competition. Exact 18S rows have an exact_extent route suffix, NA unrestricted search fields and zero interval ties; a finite parse consumes its supplied extent and does not refine the caller endpoints.

tRNA detection parameters

scavengeonly=t: Call tRNAs with the kmer-guided scavenger only (finds tRNAs at conserved long-kmer positions). This is the shipped default; set scavengeonly=f to use the PGM candidate generator instead.
scavenge=f: Run the scavenger in addition to the PGM candidate generator (augment mode) rather than replacing it.
trnaintron=t: Intron-aware pass: splice candidate spans before verification to recover intron-containing (mainly archaeal) tRNAs. Not used while scavengeonly is on.
mintrnakhits=1: Min conserved tRNA long-kmers a candidate must carry before it is aligned; a cheap pre-filter that rejects non-tRNA windows. 0 disables the pre-filter. Also accepts mintrnakmerhits.

tRNA alignment parameters

trnaalign=t: Align predicted tRNAs to a consensus library to verify and annotate them. Reduces false positives and adds anticodon/amino-acid annotation. Uses built-in library by default; override with trnalib= and trnamodel=. Also accepts aligntrna.
trnalib=<file>: Custom tRNA consensus library (fasta).
trnamodel=<file>: Custom tRNA HBM model file.
indexk=7: Kmer length for the library shortlist index. Longer is more selective (fewer models aligned per candidate); shorter is more permissive. 7 is the shipped default. Also accepts shortlistk.
indextopn=60: Max library models aligned per candidate (search breadth).
indexminhits=12: Min shared index-kmers for a model to enter the shortlist. Fixed fallback; used only when adaptiveminhits=f.
adaptiveminhits=t: Adapt the shortlist cutoff per candidate instead of the fixed indexminhits (the shipped default). Cutoff = ceil(max(adaptfloor, adapttopfrac*maxSharedKmers, adaptqfrac*queryKmers)). Also accepts adaptiveshortlist.
adaptfloor=11: Absolute floor for the adaptive cutoff (the constant term).
adapttopfrac=0.48: Adaptive cutoff as a fraction of the best model's shared-kmer count.
adaptqfrac=0.072: Adaptive cutoff as a fraction of the candidate's kmer count.
patience=20: Stop aligning after this many models without improvement once a passing hit has been found (with earlyexit).
earlyexit=t: Enable the patience-based early exit.
idpass=0.75: Min alignment identity to accept a tRNA.
hbmpass=0.75: Min HBM model score to accept a borderline tRNA.
acextract=t: Extract the anticodon directly from each verified tRNA's structure (anticodon loop position projected through the alignment). Adds an anticodon: attribute to the GFF. Also accepts anticodonextract.
acvalidate=12: Min structural score (anticodon stem + U33 + purine-37, max 15) to trust an extracted anticodon; failures fall back to model-name annotation. Also accepts acthresh.
acmargin=3: Min score margin between the best and runner-up anticodon register; ambiguous positions fall back to the model name.
trimtrna=t: Trim verified tRNA boundaries to the consensus alignment extent, removing scanner slop for accurate coordinates. Also accepts trnatrim.
maxtrna=120: (Experimental) Raise the tRNA candidate length cap, enabling relaxed length scoring for over-length candidates such as intron-containing archaeal tRNAs. Measured neutral: unspliced candidates still fail alignment verification. Also accepts maxtrnalen.

Conserved ncRNA engineering: 18S (euk18s)

euk18s=f: Enable the resource-bound development 18S recipe: seven cleaned models, 385 seeds, distinct-2 support, voted windows and rolling MSA.
euk18sconsensus=: Consensus FASTA; requires matching euk18smodelcutoffs= table. Default euk18S_consensus_sequence.fa.
euk18smodelcutoffs=: Model cutoff table for the 18S consensus. Default euk18S_model_cutoffs.tsv.
euk18sseeds=: Forward K17 FASTA; omitted uses euk18S_17mers.fa.gz.
euk18srawends=alignment: Raw endpoint source: alignment (default) or vote; omitted offsets use euk18S_seed_offsets.tsv.
euk18svoteoffsets=: Seed-offset table override. Default euk18S_seed_offsets.tsv.
euk18svoteseedsha80=<20hex>: Expected 20-hex seed-set checksum for the vote-offset table. When this is unset and no offset table is overridden, the expected value defaults to b468f4f1faaf9dd6e56f.
euk18saligner=pacbio: Alternative quantum retains the earlier primary aligner. PacBio MSA primary with sqrt-discounted deletion-run identity. Keeps model cutoffs and optional local rescue unchanged. Requires alignment ends and endpoint NN off. Small-family cm=t may run concurrently. Native out-of-window alignments are rejected without clamping. With ncrnadiag, also runs and counts a diagnostic Quantum alignment; supplemental NCRNA_MODEL_PACBIO stderr rows retain both identities, native score/endpoints/padding, separate costs and matrix payload. Diagnostics are off unless requested.
euk18smsadel4=-2: Development-only PacBio penalty at deletion residues 21-80. Requires euk18saligner=pacbio.
euk18smsaengine=rolling: Uses six score rows plus compact traceback, retaining PacBio scores and paths. Requires euk18saligner=pacbio. Explicit native selects the full matrices. Rolling matrix-payload diagnostics count rows plus ancestry; neither engine's payload is a whole-JVM memory requirement. Default rolling.
euk18smsadel5=-1: Long-deletion charge every fourth residue beyond 80. Requires euk18saligner=pacbio.
euk18smsains4=-8: Long query-insertion extension penalty (genomic inserts use DEL). Requires euk18saligner=pacbio.
euk18smsains=-205: Insertion-open penalty. Requires euk18saligner=pacbio.
euk18smsadel=-292: Deletion-open penalty. Requires euk18saligner=pacbio.
euk18smsasubscale=0.66: Scale SUB/SUBR/SUB2/SUB3 together, rounding magnitudes to nearest integer with half ties upward. Range (0,1]. All cost flags require euk18saligner=pacbio. Initial penalty domain is [-292,0]. Default substitution scale is 66%; explicit 1 restores the former baseline. Other penalties retain native constants.
euk18svotewindows=t: Use voted windows (default t).
euk18svotebeforepadding=t: Use repaired-core votes before window padding (default t). Requires voted windows, a non-negative window pad, and a positive collapse fraction.
euk18swindowpad=3109: Fallback window radius. Mapping uses K9.
euk18sflank=100: Retained explicit engineering control (default 100). In the Java source it sets the fallback window only when the window pad is negative, which euk18swindowpad does not accept, and it also enters an integer-overflow check.
euk18sminlen=916: Minimum candidate length in bases (at least 17; default 916). Joined discovery requires 916.
euk18smaxlen=4000: Maximum candidate length in bases (default 4000). Must be at least euk18sminlen.
euk18sseedminhits=2: Minimum seed hits per window (default 2).
euk18sseeddistinct=t: Count distinct seed keys (default t, distinct-2 support). Set f to count seed occurrences instead.
euk18scollapsefrac=0.9: Overlap fraction used to collapse candidate windows (0 to 1; default 0.9).
euk18smodelclip=rescue: Model-end clipping. rescue retries primary rejections only, using euk18srescueid. Setting t or f selects direct clipping or none. Requires endpoint NN off and raw alignment ends. The Java source initializes this option to rescue.
euk18srescueid=0.66: Rescue identity threshold, used with euk18smodelclip=rescue (default 0.66).
euk18sindexminhits=96: Minimum shared K9 mapping hits for a model to enter the shortlist (default 96).
euk18stopn=<uncapped>: Model-count cap per window (topN). Omitted means uncapped. Requires euk18sindexmargin=-1, because the default score-margin shortlist cannot be combined with a topN cap.
euk18sindexmargin=121: Nonnegative values replace topN with an inclusive shared-K9 score gap below the best model in each window. Keeps all positive scores >= best-minus-margin and >= indexminhits; no model-count cap. Query words are deduplicated; model occurrences retain multiplicity. Do not also specify euk18stopn. -1 keeps the existing topN policy. First accepted alignment exits.
euk18sindexfrac=0.8: B in [0,1]; also require shared K9 >= ceil(B * best). Combined cutoff is max(indexminhits, best-indexmargin, ceil(B*best)); omit the margin term when indexmargin=-1. B=0 preserves prior behavior. Nonzero B uses all models, so do not also set euk18stopn.
euk18svoteslack=60: Vote slack in bases (default 60).
euk18svotemaxsd=5: Maximum standard deviation of voted endpoints (default 5).
euk18sendpoint=f: Optional resource-sized refinement; requires explicit euk18sendpointtables= and euk18sendpointnets= directories.
euk18sendpointtables=: Directory containing the 18S endpoint tables. Required with euk18sendpoint=t.
euk18sendpointnets=: Directory containing the 18S endpoint networks. Required with euk18sendpoint=t.
euk18sjoinedproposals=: Experimental replay of frozen joined-window geometry. Requires euk18s=t, rolling PacBio, raw alignment ends, and minlen=916. Ordinary candidate rules remain unchanged; joined spans use the same identity table and final caller path.
euk18slivepositions=: Experimental live joined discovery from the shared 17-mer hit stream and an exact-consensus-bound position resource. Mutually exclusive with saved proposal replay; same fast-only controls. A joined call must realize a deletion of at least 256 bases and retain at least 916 aligned core bases.
euk18sjoinedsupport=1: Select side-support 1 or 2 for either joined source.
euk18sjoined=t: Explicit control for joined discovery; t requires rolling PacBio. Omitted Quantum/native-MSA calls use ordinary windows only. With t and no explicit position path, load euk18S_seed_positions.tsv. Use f in both backends for a matched aligner-only comparison.

Conserved ncRNA engineering: eukaryotic 5S (euk5s)

euk5s=f: Independently enable eukaryotic 5S with two seed hits per window. Needs neither ncrna=t nor rrna17=; leaves legacy flags unchanged. All calls compete in the shared DP by score and overlap penalties. Shortlist all models (currently 7); load model-specific cutoffs (currently .68/.68) and refine endpoints with packaged networks.
euk5sconsensus=: Explicit euk5S consensus FASTA for a frozen resource comparison. Requires euk5s=t or an euk5S rrna17 profile. Seed overrides use ncrnafamily=euk5S ncrnakmers=; other families are unchanged.
euk5sendpoint=t: Refine enabled euk5S calls; f retains raw alignment endpoints.
euk5sendpointtables=: Override the directory containing TABLES.tsv and tables. Defaults resolve resources/euk5S_endpoint/. Missing or incompatible resources fail before calling; start and stop may use different k.
euk5sendpointnets=: Override the directory containing MANIFEST.tsv and networks/. Defaults resolve resources/euk5S_endpoint/.
euk5smodelcutoffs=: Override resources/euk5S_model_cutoffs.tsv; f uses scalars. Each model needs family/model/idpass/idborderline TSV columns. Explicit model tables take precedence over scalar identity flags. Targeted scalar overrides bypass the default table, retaining .68 for either unset scalar. euk5s=f loads none of these resources.

Conserved ncRNA engineering: profiles and family enablement

ncrna=f: Enable generic conserved-ncRNA families (off by default). Alias: generalncrna=.
rrna17=f: Enable a 17-mer rRNA profile: 16s loads only the new 16S caller; prok loads 16S/23S/prokaryotic-5S; euk loads 18S/eukaryotic-5S. euk18s and euk5s load either eukaryotic family alone. Families share one seed scan per contig and strand. rrna17=16s requires call16s=f and preserves other calling flags. Other profiles require 16s=f 23s=f 5s=f 18s=f to prevent duplicate calls. Alias: rrna17mer=.
rrna16sseedminhits=2: Minimum seed occurrences per 16S candidate window. The packaged resource contains 285 forward 17-mers; verification uses the best K9 model, minimum identity 0.66, and complete length 1400. Requires rrna17=16s or rrna17=prok.
rrna16swindow=containment: Preserve containing candidate windows during collapse. Set legacy to use the older intersection policy. Requires rrna17=16s or rrna17=prok.
rrna16spartialminlen=1200: Rescue failed 16S windows at physical contig ends when alignment proves model clipping; 0 disables this rescue. Accepted range 0-1400. Partial calls keep raw endpoints and emit partial=10, partial=01, or partial=11 for missing genomic left, right, or both ends. Internal fragments retain the complete-gene length requirement.
rrna16sendpoint=t: Refine enabled 16S endpoints with packaged model-specific networks. Set f for raw alignment ends. Mitochondrial 16S and rescued physical-edge partials always retain raw endpoints.
rrna16sendpointtables=: Override the directory containing TABLES.tsv and the tables. Requires rrna17=16s or rrna17=prok.
rrna16sendpointnets=: Override the 16S .bbnets file or selected-network directory. Resources load only when 16S is enabled; missing or incompatible resources fail before calling.
rrna17prok5shbm=t: Enable prok5S HBM rescue for alignments with identity from 0.60 through 0.80. Set false for a matched detection ablation; requires rrna17=prok and leaves the full profile on.
rrna23sendpoint=off: 23S endpoint refinement inside rrna17=prok is disabled. The discarded g28/g16 candidate-scoring networks are not shipped; explicitly requesting either fails if its resources are missing.
v2refresh=f: Recheck claimed loci before each ncRNA window when enabled.
tmrna=f: Add the experimental tmRNA family; requires ncrna=t (or generalncrna=t). Endpoint refinement follows the global ncrnaboundarynet flag below.
sixs=f: Add the experimental paired 6S/SsrS families (RF00013 and RF01685); requires ncrna=t (or generalncrna=t). Aliases: ssrs=, 6s=. Both families must load together, with complete consensus, HBM, and kmer resources and index-aligned consensus/HBM entries. If ncrnaboundarynet=t, complete index-aligned boundary resources are also required for every loaded family; missing or misaligned resources fail loudly.
r58lsu=f: Add the bundled eukaryotic 5.8S and LSU families. This implies ncrna=t; the family remains off unless r58lsu=t is specified. Explicit resource overrides below remain available. The global ncrnaboundarynet gate remains incompatible with this pair; use the R58-specific endpoint gate below.
r58ncrnaboundarynet=f: Refine only R58 endpoints with its boundary network and 9-mer tables. Requires r58lsu=t; LSU remains boundary-NN-off. Also accepts r58boundarynn.
r58ncrnaboundary3primeonly=f: Hold the aligned R58 5' coordinate fixed and refine only the 3' endpoint. Requires r58ncrnaboundarynet=t.
r58ncrnaboundarymarginstop=0: Require this confidence improvement before moving the R58 3' endpoint. The validated candidate value is 0.01.
r58ncrnaboundarymarginstart=0: Corresponding 5' confidence margin. The current validated 3'-only configuration leaves this unused.
its=f: Derive ITS1, ITS2, and combined ITS annotations from compatible 18S, 5.8S, and LSU calls. outits=<file> also enables this.
s18=f: Add the experimental generic 18S development family (pilot); requires ncrna=t and explicit s18kmers/s18consensus/s18models. Distinct from 18s=t, which enables the legacy PGM-path 18S caller. This is a sweep harness, off by default, not a release caller.
ncrnaboundarynet=f: Refine generic ncRNA endpoints with the family-specific boundary networks. Requires ncrna=t (or generalncrna=t). When enabled, every loaded family must have complete, index-aligned boundary resources; missing or misaligned resources fail loudly.
ncrnafamily=: Select rnasep, srp_small, srp_large, tmrna, sixs_rf00013, sixs_rf01685, r58, lsu, s18, or euk5S as the generic sweep overrides below; requires ncrna=t (or generalncrna=t). A tmrna target also requires tmrna=t; a sixs_* target also requires sixs=t; an r58 or lsu target also requires r58lsu=t; an s18 target requires s18=t. euk5S instead requires euk5s=t, rrna17=euk or rrna17=euk5s; it does not require ncrna=t. Accepted aliases include rnase-p, rnase_p, srp-small, srpsmall, srp-large, srplarge, tm-rna, tm_rna, and ssra (case-insensitive). Other resource-available families remain loaded and compete normally; this is not an isolation switch.
ncrnakmers=: Explicit conserved-kmer fasta for the selected ncrnafamily=; requires that family's enabled profile and exactly one target.
ncrnaseedminhits=2: euk5S conserved 17-mer occurrences required per window. Standalone euk5s=t requires at least 2; rrna17 profiles retain explicit one-hit engineering controls for baseline comparisons.

Conserved ncRNA engineering: index, length, and identity controls

ncrnaindexk=: Targeted family mapping k (1-15); seed k remains unchanged. Larger values allocate exponentially larger indices; 6/7 are tuning choices.
ncrnaindextopn=: Maximum models shortlisted per candidate (>=1).
ncrnaindexminhits=: Shared mapping-kmer cutoff (>=0). A multi-model index keeps its best model below this cutoff unless strict filtering is on. Zero bypasses mapping only for a one-model library.
ncrnastrictindexcutoff=f: Reject euk5S windows below the mapping cutoff.
ncrnarankedfallback=f: Stop after the first identity-passing euk5S alignment whose span meets the requested maximum length.
ncrnamaxlen=2147483647: Maximum inclusive euk5S alignment/call span (>=90). Does not cap seed-window size. Overlong models are skipped before acceptance, allowing later shortlisted models to pass. Seed minhits, strict filtering, ranked fallback and maxlen require ncrnafamily=euk5S. Index controls support every listed ncrnafamily target. Omitted values preserve packaged behavior; no calibrated cutoff is implied. euk5S has no HBM: ncrnahbmpass is rejected for this target.
ncrnaadaptiveminhits=: Enable adaptive mapping cutoff for the selected family.
ncrnaadaptfloor=: Minimum adaptive cutoff (finite, >=0).
ncrnaadapttopfrac=: Best-model shared-count multiplier (finite, >=0).
ncrnaadaptqfrac=: Unique query-kmer multiplier (finite, >=0). Aliases: ncrnamapk, ncrnafixedminhits, ncrnaadaptiveshortlist. Prefix any index/adaptive flag above with r58 or lsu for an independent family override (e.g. r58ncrnaindexk=7). These require r58lsu=t and take precedence over targeted overrides.
r58ncrnaminlen=140: R58 minimum candidate length (1-1000); alias r58minlen. Only R58 changes; omitted retains 140. Requires r58lsu=t.
ncrnaidpass=: Alignment identity that accepts without HBM rescue.
ncrnaidborderline=: Lowest identity eligible for HBM rescue.
ncrnahbmpass=: HBM score needed to rescue a borderline alignment.
ncrnascorea=: Constant A in A+B*length*identity^2.
ncrnascoreb=: Multiplier B in A+B*length*identity^2.

Conserved ncRNA engineering: family resources and diagnostics

ncrnadiag=: Optional stage-diagnostic TSV path; alias r58diag. Reports all enabled ncRNA families, before final path selection. Omitted produces no diagnostics and adds no alignments.
rnasepkmers=: Explicit RNase-P conserved-kmer fasta; requires ncrna=t.
srpsmallkmers=: Explicit SRP-small conserved-kmer fasta; requires ncrna=t.
srplargekmers=: Explicit SRP-large conserved-kmer fasta; requires ncrna=t.
tmrnakmers=: Explicit tmRNA conserved-kmer fasta; requires ncrna=t and tmrna=t.
sixsrf00013kmers=: Explicit RF00013 16-mer fasta; requires ncrna=t and sixs=t.
sixsrf01685kmers=: Explicit RF01685 17-mer fasta; requires ncrna=t and sixs=t.
r58kmers=: Optional R58 17-mer fasta override; requires r58lsu=t.
lsukmers=: Optional LSU 17-mer fasta override; requires r58lsu=t.
r58consensus=: Optional R58 consensus FASTA override; requires r58lsu=t.
r58models=: Optional R58 HBM model override; requires r58lsu=t.
lsuconsensus=: Optional LSU consensus FASTA override; requires r58lsu=t.
lsumodels=: Optional LSU HBM model override; requires r58lsu=t.
r58boundarynet=: Optional R58 boundary-network override; requires r58ncrnaboundarynet=t.
r58boundarystarttable=: Optional R58 start 9-mer table override; requires the R58-specific boundary gate.
r58boundarystoptable=: Optional R58 stop 9-mer table override; requires the R58-specific boundary gate.
s18kmers=: Explicit 18S development 17-mer fasta; requires ncrna=t and s18=t.
s18consensus=: Explicit 18S development consensus FASTA; requires s18=t.
s18models=: Explicit 18S development HBM models; requires s18=t.
tmrnaconsensus=: Explicit tmRNA consensus fasta; requires ncrna=t and tmrna=t.
tmrnamodels=: Explicit aligned tmRNA HBM models; requires ncrna=t and tmrna=t. Generic sweep overrides require exactly one ncrnafamily=; family-specific overrides do not select or isolate a family.

Conserved ncRNA engineering: family scores and window padding

rnasepscorea=: Per-family A for RNase P (Java default 0). Requires ncrna=t.
rnasepscoreb=: Per-family B for RNase P (Java default 1.6). Requires ncrna=t.
srpsmallscorea=: Per-family A for SRP-small (Java default 0). Requires ncrna=t.
srpsmallscoreb=: Per-family B for SRP-small (Java default 1.8). Requires ncrna=t.
srplargescorea=: Per-family A for SRP-large (Java default 0). Requires ncrna=t.
srplargescoreb=: Per-family B for SRP-large (Java default 1.0). Requires ncrna=t.
tmrnascorea=: Explicit tmRNA A value (provisional shipped default 5). Requires ncrna=t and tmrna=t. Family score overrides do not require ncrnafamily=, may be combined in one run, and take precedence over generic ones.
tmrnascoreb=: Explicit tmRNA B value (provisional shipped default 3). Requires ncrna=t and tmrna=t. Family score overrides do not require ncrnafamily=, may be combined in one run, and take precedence over generic ones.
ncrnacollapsefrac=: Overlap fraction used to collapse candidate windows. The per-family default is 0.85 for most generic families and 0.75 for euk5S.
ncrnawindowpad=: Candidate padding around conserved-kmer hits. Also accepts ncrnapad.
rnaseppad=: RNase-P-only candidate padding override; requires ncrna=t.
srpsmallpad=: SRP-small-only candidate padding override; requires ncrna=t.
srplargepad=: SRP-large-only candidate padding override; requires ncrna=t.
tmrnapad=: tmRNA-only padding override (provisional shipped default 370); requires ncrna=t and tmrna=t.
sixsrf00013pad=: RF00013-only padding override (provisional shipped default 250); requires ncrna=t and sixs=t.
sixsrf01685pad=: RF01685-only padding override (provisional shipped default 125); requires ncrna=t and sixs=t. When omitted, shipped defaults are rnasep=320, srp_small=75, srp_large=250, tmrna=370, RF00013=250, and RF01685=125. These flags override only candidate-window padding; run one family and one varied dimension at a time.

Advanced tRNA candidate-generation thresholds (rarely changed)

trnaregion=20: Region-open score cutoff. Also accepts trnacutoff1.
trnacand=36: Composite candidate score cutoff (tRNA is very sensitive here). Also accepts trnacutoff2.
trnastart=2.4: Start point-model score cutoff. Also accepts trnacutoff3.
trnastop=1.5: Stop point-model score cutoff. Also accepts trnacutoff4.
trnainner=2.2: Average inner-kmer score cutoff. Also accepts trnacutoff5.

Statistics parameters

verbose=false: Print detailed progress information.
extended=false: Print extended statistics. Also accepts extendedstats or verbosestats.

Examples

Basic Gene Calling

callgenes.sh in=genome.fasta out=genes.gff

Calls genes in a bacterial genome and outputs GFF annotations.

Full Analysis

callgenes.sh in=contigs.fa out=calls.gff outa=proteins.faa out16S=16S.fa stats=stats.txt

Calls genes, extracts proteins, identifies 16S rRNA, and saves statistics.

High Sensitivity Settings

callgenes.sh in=metagenome.fa out=genes.gff minlen=30 minOrfScore=30 clearcutoffs=true

Uses more permissive settings for calling genes in metagenomic data.

Multi-pass Analysis

callgenes.sh in=genome.fa out=genes.gff passes=2 extended=true

Uses iterative refinement with 2 passes and detailed statistics output.

rRNA and tRNA Detection

callgenes.sh in=genome.fa out=all.gff out16S=16S.fa out18S=18S.fa longkmers=true

Detection of protein-coding genes and RNA genes with enhanced kmer matching.

Algorithm Details

Gene Calling Strategy

CallGenes implements a probabilistic gene calling algorithm using the GeneCaller.callGenes() method with GeneModel for k-mer frequency scoring. The algorithm combines multiple evidence sources through six reading frame analysis:

K-mer-based scoring: Uses probabilistic gene models (PGM files) with StatsContainer and FrameStats classes to score potential ORFs based on k-mer frequency patterns
Start/stop codon analysis: Evaluates start and stop signals using minStartScore and minStopScore thresholds in context-dependent scoring
Frame-specific penalties: Applies different scoring based on reading frame to account for codon usage bias
Overlap resolution: Intelligently handles overlapping ORFs with configurable overlap limits

Dual Strategy for Sequence Processing

The algorithm employs different strategies based on sequence type:

Single sequences: Direct ORF calling with strand-specific processing
Paired reads: Optional merging with BBMerge or error correction with ECCO before gene calling

RNA Gene Detection

CallGenes includes specialized modules for non-coding RNA detection:

rRNA detection: Supports 16S, 18S, 23S, and 5S rRNA identification using kmer matching and sequence alignment
tRNA detection: Identifies transfer RNA genes with adjustable sensitivity parameters
Configurable thresholds: Identity and bias parameters allow fine-tuning for different organisms

Multi-pass Refinement

The multi-pass feature enables iterative model improvement:

Initial gene calling with default or provided model
Model training on initial predictions
Refined gene calling with improved model
Optional additional passes for further refinement

Performance Characteristics

Memory usage: Scales with genome size and model complexity; typically 6GB default heap
Threading: Automatic parallelization across available CPU cores
I/O optimization: Supports compressed input files and ordered/unordered output modes
Scalability: Handles single genomes to large metagenomic assemblies

Output Formats

CallGenes produces multiple output formats:

GFF3: Standard genome annotation format with detailed gene features
FASTA: Amino acid sequences for protein-coding genes
Statistics: Performance metrics in text or JSON format
Histograms: Gene length distributions for quality assessment

Support

For questions and support:

Email: bbushnell@lbl.gov
Documentation: bbmap.org