What's new since the paper

Changes to RNAhub since the 2025 Nucleic Acids Research paper (preprint March 2025). Updated October 2026.

RNAhub 2.0: minidb mode experimental

The paper's eukaryotic databases are subclustered: one genome was kept per RNase MRP cluster, so that a search over one large concatenated database finishes in reasonable time. In minidb mode (RNAhub 2.0) each genome is stored as its own compressed file instead. The query is first searched against every genome separately, with many genomes in parallel. Only the hits (E ≤ 10) are collected into a small database for this job, and the usual three iterative nhmmer searches and R-scape then run on it. Because the expensive step is split across genomes, minidb mode can search much larger genome sets: 4,618 fungal, 1,895 invertebrate, 1,028 plant, 772 vertebrate and 231 mammalian genomes, 1.5 to 14 times more than in the paper. The pre-search uses the RepeatMasker-masked query. Several databases can be searched together.

Genome databases

Numbers are genomes unless marked otherwise; – = not available.

New ways to search

• Upload Alignment. Start from homologs you already have: plain Fasta hits (aligned automatically), aligned Fasta, or Stockholm. The alignment is used as the query profile for the iterative search. The upload is checked before the job is created.
• Covariance-model search. A Stockholm upload with a consensus structure (#=GC SS_cons) is turned into an Infernal covariance model (cmbuild, cmcalibrate) and searched with cmsearch. Pseudoknots, which a covariance model cannot represent, are set aside and restored in the final alignment.
• Re-run with other databases from any result page: the input, mode and settings are filled in for you.
• Option to skip the RepeatMasker step, and a separate E-value for the flanking search.

Results

• Known 3D structures. Before the search starts, the query and the models of its Rfam families are searched against all nucleic-acid chains in the PDB; after the search, the final alignment is too. Each hit links to RCSB with the entry title, and is aligned to the query in Rfam model coordinates, with the differing residues highlighted.
• Rfam hits as cards (significant hits first, with model coordinates and links) instead of the raw cmscan table.
• Interactive structure viewer (RFview, R-scape 2.6.16) next to the R2R drawings, and the query's predicted structure in dot-bracket.
• Interactive alignment viewer for the input, Rfam-seed and final alignments, with covarying pairs.
• Alignment summary statistics: number of sequences, effective number of sequences, % canonical pairs.
• Downloaded files are named after the job; clearer error messages that show the pipeline's own diagnosis.

Interface

• Progress shown while the query is scanned against Rfam and the PDB after submission.
• Pages adapted for phones.
• Genome counts shown for every database; new databases are marked.
• Usage statistics on this page.

Pipeline and server

• Gzipped databases can be searched directly. Large intermediate files are compressed or removed at the end of a run, so job results take much less disk space.
• More robust handling of edge cases: empty iterations, single-sequence alignments, and queries that are mostly repeats (the run stops if more than 80% of the query is masked).
• Job scheduling fixes: a job can no longer be started twice, at most three jobs run at once, and failed runs are reported as errors instead of staying "running".
• The pipeline (github.com/mmagnus/rnahub) can be run locally, including minidb mode, with updated documentation.