Hierarchical multi-label prediction of gene function
Bioinformatics
Robust prediction from multiple heterogeneous data sources with partial information
CIKM '10 Proceedings of the 19th ACM international conference on Information and knowledge management
| Hi-index | 0.00 |
Annotating proteins with biological functions is one of the main goals in post genomic era. Various high-throughout technologies, e.g., yeast two-hybrid systems and microarray, have provided an alternative way to protein function prediction. Despite the success obtained by high-throughout data, the errors in the data have not been handled well. In this work, a new technique for protein function prediction is presented, where a weighted functional linkage graph is generated by exploiting the existing protein-protein interaction data, complex data and gene expression data. By finding the shortest path in the functional linkage graph, the functional links among proteins can be captured. With the functional links available, the functions of unknown proteins can be predicted utilising support vector machines and the functions of those proteins that have functional links to the unknown proteins. In addition, the boosting algorithm is employed to further improve the prediction accuracy. The experiments on yeast genes show promising results and prove the efficiency of the proposed methods.