Download A Computational Method to Identify RNA Binding Sites in Proteins

BIOINFORMATICS AND COMPUTATIONAL BIOLOGY
A Computational Method to Identify
RNA Binding Sites in Proteins
Presented by
Jeff Sander
Iowa State University
Rocky 2006
BIOINFORMATICS AND COMPUTATIONAL BIOLOGY
Biological Motivation
• Protein
RNA complexes
Which /amino
acids are
– mRNA, tRNA, rRNA
directly
responsible for
– miRNA, siRNA, RNAi
binding
RNA?
– RNA viruses
Terribilini et al (2006) RNA
Wang & Brown (2006) NAR
Jeong & Miyano (2006) Tras Sys Biol
AMINOACYL TRANSFER RNA SYNTHETASE
BIOINFORMATICS AND COMPUTATIONAL BIOLOGY
Sequence Based Predictions
• Dataset
–
–
–
–
Protein RNA complexes from the Protein Data Bank
Less than 30% identity and 3.5A or better resolution
147 Protein RNA complexes
14% interacting residues / 86% non-interacting
• Classifier - Naïve Bayes
ARVHNTRQQGATLAFLTLRQQASLIQ
• Results:
CC: 0.33
Acc: 0.86
Sp+: 0.46
Sen+: 0.36
• Server: RNABindR
http://bindr.gdcb.iastate.edu/RNABindR/
Terribilini et al (2006) RNA
BIOINFORMATICS AND COMPUTATIONAL BIOLOGY
Using Additional Information
A R V H N T R Q Q G A T L A F
• PSSM
A R V H N T R Q Q G A T L A F
• PSI-Blast
• Str Neighbor A R V H N T R Q Q G A T L A F
• Combination A R V H N T R Q Q G A T L A F
• Actual
A R V H N T R Q Q G A T L A F
BIOINFORMATICS AND COMPUTATIONAL BIOLOGY
Results
Method
CC
Acc
SP+
SN+
Sequence
0.33
0.86
0.46
0.36
PSSM
0.35
0.86
0.48
0.38
PSI-Blast
Method
0.35
CC
0.87
Acc
SP+
SN+
0.36
0.33
0.86
0.46
0.36
0.86
0.35
0.86
0.48
0.38
0.38
0.35
0.87
0.50
Sequence
Str Neighbors
0.35
PSSM
PSI-Blast
0.50
0.48
0.87
0.35
& PSI-Blast
PSSM & Str Neigh PSSM0.39
0.37
0.87
0.87
0.52
0.38
0.39
PSSM & Str Neigh
0.39
0.87
0.53
0.39
0.87
0.38
0.87
0.55
0.36
0.36
0.39
0.88
0.56
0.37
Str & PSI-Blast
PSSM & Str & PSI
0.38
Str & PSI-Blast
PSSM & Str & PSI
0.39
0.88
0.52
0.86
0.36
PSSM & PSI-Blast Str Neighbors
0.37
0.53
0.55
0.56
0.48
0.38
0.38
0.37
BIOINFORMATICS AND COMPUTATIONAL BIOLOGY
Conclusions
• Evolutionary and structural information can enhance
prediction over basic sequence
• Combining classifiers can provide enhanced
predictions over individual classifiers
Acknowledgements
•
•
•
Michael Terribilini •
Changhui Yan
•
Jae-Hyung Lee
•
Drena Dobbs
Vasant Honavar
Robert Jernigan
Funding: USDA MGET, NIH, CIAG, ISU