"It finds nothing" / no_enriched.txt. ~90% of the time this is the dIRank pitfall: real ΔΨ values in (-1, 1) are read
as control because the regulated threshold is |dIRank| ≥ 1. Use categorical {-1, 0, +1}, or --from-rmats. Otherwise: lower --min-height (2–3), loosen --p-fisher/--p-empirical, widen -e, and check you have enough exons. See Input Format.
How many exons do I need? Roughly ≥ 50 regulated exons per direction; fewer gives unstable maps.
Why 512 motifs, not 256? The tetramer set includes degenerate IUPAC motifs (e.g. Y, W, S), not only the 256 exact 4-mers.
Empirical p never gets small. The floor is 1/bootstraps-limited. Raise -b (e.g. 10000) for a lower floor, but expect the called set to change little. With -b 1000, set --p-empirical to a reachable value such as 0.01.
"Genome not found." Run the matching genomes/<assembly>.download.sh first (hg19, hg38, mm9, mm10, mm39).
Results differ slightly between runs. The bootstrap RNG is seeded per thread, so results are reproducible only at a fixed --cores. Change the core count and borderline calls can shift. Keep --cores constant for reproducible comparisons.
It's slow. Use --bootstraps 1000 for development. Raising --cores helps only up to a point — the workload is largely memory-bandwidth-bound, so throughput plateaus well below high core counts.
Too many noisy motifs. Raise --min-height, tighten --p-fisher/--p-empirical, and/or increase -b.
ViennaRNA / PhyloP errors with --structure / --conservation. These are 🚧 work-in-progress features. --structure needs ViennaRNA installed; --conservation needs pre-downloaded .phylop.bin files (genomes/download_phylop.sh). Treat their output as experimental for now.