Identification of noncoding RNA with the ERPIN program

1
Identification of non-coding RNAwith the ERPIN
program
Daniel Gautheret INSERM ERM 206 Techniques
avancées en Génomique et Clinique Marseille André
Lambert CNRS Centre de Physique
Théorique Marseille
2
RNomics

• ncRNA genes
• tRNA
• 5S16S30S rRNA
• snoRNA (C/D H/ACA box)
• snRNA/uRNA
• tmRNA
• miRNA
• SRP RNA
• RNAse P
• RNASE MRP
• rprA,csrB,oxyS

• Introns
• Group I
• Group II

• Short Motifs
• E-loop
• GNRA loop
• UNCG loop
• AA platform

• Functional Motifs
• IRE
• IRES
• SECIS
• TAR
• Hammerhead
• Hairpin ribozyme
• Rho dept. transcription terminator
• T-box antiterminator

3
The specificity of RNA search
s
BLAST
HMM
s
s

• ncRNA is defined both by primary and secondary
  structure
•  Substitution matrices  for nucleic acids are
  terrible compared to PAM etc.

Current ncRNA annotation is terrible!
4
The trouble with existing programs

• Specific Programs (tRNA-scan)
• Descriptor-based programs
• Subjective and painful descriptor generation
• Subtle constraints easily overlooked (e.g.
  certain pairings forbidden)
• Yes/no answer no scoring

• Stochastic Context Free Grammars
• Not  practical  for large alignments or
  genome-wide searches (Eddy, 1999)
• Time cost O(N4) for sequence of length N
• Pseudoknots not allowed

h1 s1 h1 s2 h2 s3 h2 h1 55 1 h2 55
NNNNRYNNNN s1 77 NUNNNNN s2 440 s3 77
UUCNNNN
RnaMot descriptor for anticodonTYC domain of tRNA
5
RNA Structure Requires Probabilistic Descriptors

• Most nts are biased
• Many biases escape human inspection
• Probabilistic descriptors are needed to
  incorporate this

Can we use this information practically?
R. Gutells consensus bacterial 16S
6
ERPIN Profile-based RNA Search
Lambert Gautheret, JMB, 313, 1003-1011 (2001)
l10
l14
h3
h5
Target sequence
Helix score for h5-h3 computed from helix profile
Sequence (14 nt)
best score for l10 (4 gaps)
GTTCTTGCATGTTTGACGGAAC GTTCTTGCATGATTGACGGAAC GTTC
TTGCATGTTTGACGGAAC TTTCCTGCATGCTTGACGGAAC TTTAT--C
AAGTTCAT-ATAAA ATTAT--CGTGCCTTC-ATAAT ATTAT--CGTGT
CTTC-ATAAT ATTAT--CATGTTTC--ATAAT
best score for for l14 (0 gaps)
Single-strand profile (14 positions)
Training set
7
Validation

• Original paper validation on known RNA genes and
  motifs
• tRNA
• IRE (Iron Response Element)
• SECIS (Selenocysteine Insertion Sequence)

8
Handling of complex structures, search strategies
(Erpin 2 3)
tRNA Training set

• h1111111aa2222bbbbbbbbbbb2222c33333ddddddd33333eee
  eeeeeeeeeeeeeeeeeeeee44444fffffff444441111111gggg
• gtDA0260 TGC PHAGE T5
• -GGGCGAATAGTGTCAGC-GGG--AGCACACCAGACTTGCAATCTGGTA-
  ------------------G-GGAGGGTTCGAGTCCCTCTTTGTCCACCA
• gtDA0340 TGC ARCHAEGLOBUS FULG. ARCHAE
• -GGGCTCGTAGCTCAGC--GGG--AGAGCGCCGCCTTTGCGAGGCGGAG-
  ------------------GCCGCGGGTTCAAATCCCGCCGAGTCCA---
• gtDA0380 TGC HALOBACTERIUM CUT. ARCHAE
• -GGGCCCATAGCTCAGT--GGT--AGAGTGCCTCCTTTGCAAGGAGGAT-
  ------------------GCCCTGGGTTCGAATCCCAGTGGGTCCA---

• First search for TYC loop
• If found, look for all other elements
• -gt Faster search with little sensitivity loss

9
SECIS search

• Selenocysteine Insertion Sequence
• required for readthrough of UGA codon
• signature for selenoprotein genes
• Previous searches
• RNAMOT 30 hits / 100mb.
• PATSCAN energy 150 hits / 100mb
• Additional filters necessary
• Analysis of reading frame
• Experiments
• Can we do better?

SECIS
stop
UTR
CDS
10
Erpin score and false positive rates

• Y 6.6 105 e-0.3 S
• E Kmn e-? S
• (extreme value distrib.)
• Score 30 (min)
• 3.8 hits/100mb
• Score 40 (ave)
• 0.13 hits/100mb

SECIS element hits in 700 mb randomized sequences
11
SECIS Iterative Search

• Initial training set 43 SECIS sequences
• Total scanned 4 Gb
• After 5 rounds 120 sure SECIS 200 novel

12
The Erpin Download Site
13
The ncRNA annotation Project

• Create training sets for all major classes of
  ncRNA
• Collaborative projects
• Establish global ncRNA annotation resource

14
All searches parametered to scan a bacterial
genome in less than 1 minute
http//tagc.univ-mrs.fr/pub/erpin/
15
(No Transcript)
16
How is a protein-coding gene detected?

• ORF
• Pol II Promoters and CpG islands
• Codon/hexanucleotide statistics (species-specific
  biases) -gt HMM
• Availability of
• ESTs
• Large protein gene sequence databases