Twinscan-mars

1. Twinscan - MARS

Paul Flicek
Washington University
EBI

2. Twinscan

Twinscan incorporates a pair wise alignment
net (with an informant sequence) into its
algorithm
Twinscan “works” for human gene prediction
with almost any mammalian informant (but
mice and rats are the most useful)
Prediction characteristics are dependent on
the informant

3. Goals & Methods

Goals & Methods
Exploit the differences in gene characteristics
resulting from each individual informant
genome
Create closely-related prediction by
incorporating mixtures of pair-wise alignment
nets into Twinscan
Cluster these predictions as a way to find
multiple transcripts from a single gene

4. Pair-wise Informant Set

Mouse Informant Set
Cf
Rn
Mm
Gg
Human Target Sequence
Xt

5. Results

Successes
Successfully predicts alternatively spliced
transcripts de novo
Scales well
Approximately 1.67 transcripts per gene with the
mouse informant set
Sensitive…

6. Results

Challenges
Transcript clustering in gene dense regions (like
the manually picked ENCODE regions) can be
problematic