Download Comparative Annotation of Viral Genomes with Non

Comparative Annotation of Viral Genomes
with Non-Conserved Gene Structure
Saskia de Groot and Jotun Hein
Department of Statistics, University of Oxford
Motivation
Results
Viral genome annotation is a complex task:
• Overlapping and nested reading frames
• Atypical sequence evolution
• Non-conserved gene structure
HIV2 vs. HIV2
84-89% Sensitivity
97-99.9% Specificity
⇒ Current comparative HMM methodologies can’t cope.
Aim
• Introduce a pair hidden Markov model to annotate two aligned
homologous genomes simultaneously
• Analyse HIV1 and HIV2 – two sequences related, but with
non-homologous gene structure
HIV1 vs. HIV2
84% Sensitivity
98.5% Specificity
• Incorporate prior knowledge by annotating one sequence
conditional on the other
Methods
Introduce pair HMM specific
to overlapping reading frames.
3 reading frames ⇒
23 x23 = 64 states
Define three different types of
start transition probability α, β, γ
depending on coding state
HIV1 | HIV2
98.7% Sensitivity
99.5% Specificity
Use evolutionary model specific
to overlapping reading frames –
substitutions are accepted by a
selection factor f.
Use EM with Forward-Backward and Newton-Raphson for
parameter estimation & Viterbi to get annotation.
Conclusion
Future Work
• Shown validity of overlapping pair HMM approach
• Improve model by adding varying selection levels
• Demonstrated amount of information contained in
conservation of gene structure
• Incorporate a viral genome aligner for de novo gene
annotation
• Provided successful method for annotating new viral strains
• Build viral genome & evolution simulator to test hypotheses