Parameters

Full reference for every rnamotifs flag, grouped by role. See How the algorithm works for the concepts behind them.

Required

FlagMeaning
input_filePositional: the splicing-change file (Input Format), or an rMATS file with --from-rmats.
-n, --nameRun name; becomes the results sub-folder and plot title.
-g, --genomeReference assembly: hg19, hg38, mm9, mm10, mm39. The matching genome must be downloaded (genomes/).

Clustering — what counts as a motif cluster

FlagDefaultWhat it controlsWhen to change
-w, --half-window15Half-width hw (bp) for clustering motif copies; copies within ±hw join one cluster. In paper notation this is the clustering window n = 2·hw (--n is an alias: --n 30-w 15). Larger → looser, longer-range multivalency.Increase to capture dispersed motif arrays; decrease for tight local clusters. The MaRs sweep tries hw ∈ {5, 15, 25, 35} (n ∈ {10, 30, 50, 70}).
-m, --min-height4Minimum number of motif copies stacked within a window to call a cluster. Higher → stricter.Lower (2–3) for sparse motifs / small exon sets; raise for very abundant motifs.
-p, --pth0.5Percentage threshold used when binarising cluster occupancy along the map.Rarely changed; leave at default unless reproducing a specific protocol.

Region geometry — where you look

FlagDefaultWhat it controls
-e, --enrichment-window30Enrichment window ew (bp; paper notation e) over which positional enrichment is aggregated. Larger → smoother, broader signal. The MaRs sweep tries {30, 50, 100, 200, 300}.
--in-exon30How far (bp) into the exon the analysed region extends from each splice site.
--in-intron300How far (bp) into the intron the analysed region extends from each splice site.
--event-typeSESE (skipped exon) or 🚧 RI (intron retention — work in progress, see below).

hw (paper n = 2·hw) and ew (paper e) are the two parameters most worth tuning — they trade sensitivity vs specificity. There is no universal optimum; the MaRs discovery mode grid- or Bayes-searches them per RBP.

Statistics — how strict the filter is

FlagDefaultWhat it controls
-b, --bootstraps10000Bootstrap iterations for the empirical-FDR null. 1000 is the practical setting (fast; floor ≈ 10⁻³). 10000 is ~10× slower for a lower floor that rarely changes the called set.
--p-fisher0.1Fisher's-exact p-value threshold for regulated-vs-control enrichment.
--p-empirical0.00005Empirical p-value threshold from the bootstrap null. Only reachable with a large -b; with -b 1000 use a looser value such as 0.01.
-k, --kmer-size4Motif length. 4 (tetramers) is the validated default. 5/6 enumerate exponentially more, sparser motifs and generally reduce discriminability at higher cost.

Performance & plotting

FlagDefaultMeaning
-c, --cores1CPU threads (k-mer scan + bootstrap are parallelised). The workload is largely memory-bandwidth-bound, so throughput plateaus well below very high core counts. Bootstrap reproducibility holds only at a fixed --cores (per-thread RNG seeding).
--top-nallPlot only the top-N ranked tetramers (cosmetic; does not change the called set).

🚧 Optional analyses — work in progress

These run but are not yet validated in this release; treat output as experimental.

FlagDefaultMeaning
🚧 --event-type RIIntron-retention mode (different region geometry).
🚧 --structureoffRNA secondary-structure accessibility profiles for enriched motifs (requires ViennaRNA).
🚧 --structure-window31Folding window size (bp) for structure profiling.
🚧 --conservationoffPhyloP conservation profiles (requires pre-downloaded .phylop.bin, see genomes/download_phylop.sh).

rMATS input (with --from-rmats)

See Input Format → Feeding rMATS directly for --rmats-incl, --rmats-fdr, --rmats-constit, --rmats-max-constit.


Tuning cheat-sheet

SymptomTry
Nothing called (no_enriched.txt)Lower --min-height (2–3); loosen --p-fisher/--p-empirical; widen -e; ensure ≥ ~50 regulated exons; verify dIRank is {-1,0,1}, not raw ΔΨ.
Too many noisy motifsRaise --min-height; tighten --p-fisher/--p-empirical; increase -b.
Clusters too local / fragmentedIncrease -w and/or -e.
Want longer footprintsIncrease -w; only try -k 5 with a strong prior (-k 4 usually wins).
Runs too slowUse -b 1000; raise -c (diminishing returns — memory-bandwidth bound).
Don't know best hw/ew (paper n/e)Don't guess — use rnamotifs-mars --mode discovery to grid/Bayes-search.

← Back to tutorial index

← Back to tutorial overview