Learning Kernel Matrix from Gene Ontology and Annotation Data for Protein Function Prediction

Authors:
Yiming Chen;Zhoujun Li;Junwan Liu
Affiliations:
Computer School of National University of Defence and Technology, Changsha, China 410001;Computer School of Beihang University, Beijing, China 100000;Computer School of National University of Defence and Technology, Changsha, China 410001
Venue:
ISNN 2009 Proceedings of the 6th International Symposium on Neural Networks: Advances in Neural Networks - Part III
Year:
2009

Citing 7
Cited 0

Diffusion Kernels on Graphs and Other Discrete Input Spaces

ICML '02 Proceedings of the Nineteenth International Conference on Machine Learning
An Information-Theoretic Definition of Similarity

ICML '98 Proceedings of the Fifteenth International Conference on Machine Learning
Kernel Methods for Pattern Analysis

Kernel Methods for Pattern Analysis
Convex Optimization

Convex Optimization
Learning the Kernel Matrix with Semidefinite Programming

The Journal of Machine Learning Research
Learning kernels from biological networks by maximizing entropy

Bioinformatics
Large Scale Multiple Kernel Learning

The Journal of Machine Learning Research

Quantified Score

Hi-index	0.00

Visualization

Abstract

During the last few years, Kernel methods have gained considerable attention for analyzing biological data for protein function prediction. Based on biological processes annotation of Yeast and GO(gene ontology), we constructed a kernel matrix to predict protein functions. We used measurement method about semantic similarity on GO and adaptive Hausdorff distance to successfully obtain protein similarity matrix, and furthermore, transformed protein similarity matrix to a undirected graph. Then, We developed a novel method that can learn optimal diffusion kernel from graph by maximizing kernel-target alignment. Experimental results illustrate that the kernel matrix generated by our formula has larger AUC value than ordinary diffusion kernel and those proposed before. Our method can even learn a common optimal kernel matrix for multiple predict tasks at one run. Furthermore, it can also be directly used to learn from various biolobical networks.