Gene Finding with HMMs

1
SLAMming the Human and mouse genomes Marina
Alexandersson Simon Cawley Lior Pachter
2
Example a human/mouse ortholog
Proliferating cell nuclear antigen (PCNA)
Human Locus
Alignment
CDS
Mouse Locus
3

• Observation
• - Finding the genes will help to find
• biologically meaningful alignments.
• Finding a good alignment will help in
• finding the genes.

4
(No Transcript)
5

• Using GPHMMs for cross-species gene finding

given a pair of syntenic sequences
predict genes by estimating hidden state sequence
Predict exon-pairs using single most likely
sequence of hidden states (Viterbi).
6
A GPHMM implementationSLAM

• SLAM components
• Splice sites (Variable length Markov models).
• Introns and Intergenic regions (2nd order Markov
  models, independent geometric lengths, CNS
  states).
• Coding sequences (3-periodic Markov models,
  generalized length distributions, protein-based
  pairHMM.)
• Input
• Pair of syntenic genomic sequences.
• Approximate alignment.
• Output
• CNS and CDS predictions in both sequences.

7
CNS
D
Y
Z
M
I
8
Approximate alignment
Currently generated by running AVID and then
relaxing
9
(No Transcript)
10
Example Rosetta Set.
Nucl.
Sn
Sp
AC
Genscan
.908
.929
.975
.951
SLAM
.981
.960
Rosetta
.935
.978
.949
11
Human genome and Mouse assembly

• RepeatMasker
• BLAT
• AVID

http//pipeline.lbl.gov/

• Build an approximate alignment
• for every sequence pair
• Realign and annotate with SLAM

12
http//bio.math.berkeley.edu/slam/mouse/