General Departmental Seminar Series
Artificial Neural Network Prediction of Antisense Oligodeoxynucleotide Activity
Michael Giddings, Department of Human Genetics
University of Utah
Wednesday, May 3, 2000, 11:00 am - 12:00 pm
2310 Computer Science & Statistics
Antisense oligonucleotides have uses for both research and therapeutic purposes. They provide the ability to target a gene with high specificity, reducing its in vivo expression through hybridization with its RNA transcript. A given RNA sequence presents many potential sites for antisense targeting. A significant issue in antisense oligonucleotide design is the selection of those sites where antisense targeting will be most effective. Prior computational approaches to this problem have considered hybridization energetics and/or structural issues to try to predict these sites. We have approached this problem by examining the correlation between short (e.g. tetranucleotide) sequence motifs and the activity of antisense molecules, using a database of 349 oligos reported in the literature. Upon finding that certain motifs have a high correlation with oligo efficacy, we designed a neural network system to utilize these correlation patterns for predicting oligo efficacy. Networks were developed which can predict active target sites with a success rate of over 50%. This can be compared to a success rate for finding active oligos of less than 10% when sites are selected by trial and error. The method therefore provides a five-fold reduction in the number of oligos that must be screened to find effective sites. The talk will present our approach, covering some of the network architectures and cross-validation procedures used, as well as a discussion of how this tool might be used practically in the laboratory. Given time, there may also be brief forays into the topics of proteomics and careers in bioinformatics.
Back to General Departmental Seminar Series