Abstract
The subcellular localization of a protein is closely correlated with its function. Automatic prediction of subcellular localization based on protein sequence properties remains a challenging problem. Here, we propose a proteomic screening-based machine learning approach for interpreting differential detection of proteins in isolated organellar compartments by high-throughput mass spectrometry. The method deals with some core limitations existing in previous approaches, such as multi-compartmental ambiguity. When applied to a global-scale proteomic study, our method achieved an excellent overall accuracy of 80.5% and precision 75.1% for four major organellar compartments (cytosol, membranes, mitochondria, and nucleus). The classifiers were able to predict the subcellular localization of 2390 previously uncharacterized proteins, 1370 of which were assigned to one or more compartments with at least 80% confidence.
Original language | English (US) |
---|---|
Title of host publication | Bioinformatics |
Subtitle of host publication | A Concept-Based Introduction |
Publisher | Springer US |
Pages | 163-174 |
Number of pages | 12 |
ISBN (Print) | 9783540241669 |
DOIs | |
State | Published - 2007 |
Externally published | Yes |
Keywords
- Automatic prediction
- Machine learning
- Multi-compartment
- Protein expression profiling
- Proteomics
- Subcellular localization
ASJC Scopus subject areas
- General Computer Science