Infernal: inference of RNA alignments

Overview

Infernal ("INFERence of RNA ALignment") is for searching DNA sequence databases for RNA structure and sequence similarities. It is an implementation of a special case of profile stochastic context-free grammars called covariance models (CMs). A CM is like a sequence profile, but it scores a combination of sequence consensus and RNA secondary structure consensus, so in many cases, it is more capable of identifying RNA homologs that conserve their secondary structure more than their primary sequence.

The latest release of Infernal is 1.0 [13 Jan 2009].

Documentation:

User's Guide [PDF, 106 pages] .
00README from the current release.

Reference:

The recommended citation for using Infernal 1.0 software is E. P. Nawrocki, D. L. Kolbe, and S. R. Eddy, Infernal 1.0: Inference of RNA alignments , Bioinformatics (2009), in press.

Download:

The current source code distribution: infernal 1.0 [tarball, 15.0 MB]
Infernal is freely available under the GNU General Public License version 3 [GPLv3].
1.0 is (at long last) the first "complete" release of Infernal.

We welcome code contributions, feature requests, and bug reports.

Note on the Rfam transition to Infernal 1.0

The current Rfam 9.1 database is build on Infernal 0.81. Rfam models are not currently compatible with Infernal 1.0.

We cannot effectively make Infernal 1.0 reverse compatible, because Infernal 1.0's model calibration step is new, essential, and computationally intensive. That is to say, we could easily have 1.0 read pre-1.0 model files, but those models would not have 1.0's capability of calculating E-values until you calibrated the models; and if you're going to calibrate the models (which takes forever), you might as well build new ones (which takes little time).

At the same time, Rfam cannot instantaneously switch to Infernal 1.0, because they use curated score thresholds, and with this significant code change, they'll need to reexamine all their thresholds. We do expect Rfam to adopt Infernal 1.0 in the next release.

In the meantime, if you want to use Infernal 1.0 with Rfam 9.1, we have posted a tarball of Rfam 9.1 models built with Infernal 1.0, which you can download from our FTP site [8.5M, tar.gz].

Alternatively, you can continue using the previous Infernal 0.81 release for Rfam work until Rfam makes the transition.

Further reference:

Covariance models were originally described in (Eddy and Durbin, 1994). They were first implemented in the COVE software package.
The tRNA identification program tRNAscan-SE is based on a tRNA CM and COVE (Lowe and Eddy, 1997).
A chapter of Biological Sequence Analysis (Durbin et al., 1998) describes CMs in more detail.
The memory-efficient divide-and-conquer CYK alignment algorithm is described in (Eddy, 2002). A new implementation, Infernal, replaced COVE.
The RSEARCH software for searching with single RNA structures (instead of with profiles built from multiple alignments) is based on Infernal's alignment engines (Klein and Eddy, 2003). RSEARCH also included MPI parallelization and E-value calculations, features that were merged back into Infernal in Infernal 1.0.
The Rfam database of RNA families is based on the Infernal software (Griffiths-Jones, 2003; Griffiths-Jones, 2005) .
The QDB query-dependent banded dynamic programming algorithm used to increase search speed is described in (Nawrocki and Eddy, 2007), along with mixture Dirichlet priors and entropy weighting.

Archived releases:

For reproducing published results, the following archived older versions may be of interest. Other releases are also archived on our FTP site, but these are the main mileposts:

Release	Date	Download	Notes
0.1	April 2002	[infernal-0.1.tar.gz, 402 KB]	Version used in the divide and conquer algorithm paper (Eddy, 2002)
0.3	July 2002	[infernal-0.3.tar.gz, 431 KB]	Version used for the first Rfam release (Griffiths-Jones et al, 2003)
0.4	August 2002	[infernal-0.4.tar.gz, 443 KB]	cmalign first appears, and was used for final work on Rfam 1.0.
0.55	April 2003	[infernal-0.55.tar.gz, 763 KB]	First version with documentation. Rfam 2.0 through Rfam 7.0 are based on 0.5x code (Griffiths-Jones et al, 2005)
0.6	November 2005	[infernal-0.6.tar.gz, 4.3 MB]	Mixture Dirichlets incorporated, under pressure from Paul Gardner and Eva Freyhult's ongoing benchmarking work.
0.7	December 2005	[infernal-0.7.tar.gz, 4.3 MB]	Entropy weighting incorporated. This version was used by Paul Gardner and Eva Freyhult in their BRaliBaseIII benchmark.
0.71	November 2006	[infernal-0.71.tar.gz, 6.0 MB]	Version used in the submitted version of the QDB algorithm manuscript (Nawrocki and Eddy, 2007)
0.72	January 2007	[infernal-0.72.tar.gz, 6.0 MB]	Version used for the published version of the QDB algorithm paper (Nawrocki and Eddy, 2007)
0.81	May 2007	[infernal-0.81.tar.gz, 6.5 MB]	Version used for current Rfam versions (9.1 and earlier).

Sean Eddy

Last modified: Fri Apr 17 15:01:23 2009