Shibiao Wan, Man-Wai Mak, Hong Kong Polytechnic University, Hong Kong.

• 1 Introduction1.1 Proteins and Their Subcellular Locations1.2 Why Computationally Predicting Protein Subcellular Localization?1.3 Organization of The Thesis2 Literature Review2.1 Sequence-Based Methods2.2 Knowledge-Based Methods2.3 Limitations of Existing Methods3 Legitimacy of Using Gene Ontology Information3.1 Direct Table Lookup?3.2 Only Using Cellular Component GO Terms?3.3 Equivalent to Homologous Transfer?3.4 More Reasons for Using GO Information4 Single-Location Protein Subcellular Localization4.1 GOASVM: Extracting GO from Gene Ontology Annotation Database4.2 FusionSVM: Fusion of Gene Ontology and Homology-Based Features 4.3 Summary5 From Single-Location to Multi-Location 5.1 Significance of Multi-Location Proteins5.2 Multi-Label Classification 5.3 mGOASVM: A Predictor for Both Single- and Multi-Location Proteins5.4 AD-SVM: An Adaptive-decision Multi-Label Predictor5.5 mPLR-Loc: A Multi-Label Predictor Based on Penalized Logistic- Regression 5.6 Summary 6 Mining Deeper on GO for Protein Subcellular Localization6.1 Related Work 6.2 SS-Loc: Using Semantic Similarity Over GO6.3 HybridGO-Loc: Hybridizing GO Frequency and Semantic SimilarityFeatures6.4 Summary7 Ensemble Random Projection for Large-Scale Predictions7.1 Related Work 7.2 RP-SVM: A Multi-Label Classifier with Ensemble Random Projection7.3 R3P-Loc: A Predictor Based on Ridge Regression and Random Projection7.4 Summary 8 Experimental Setup8.1 Prediction of Single-Label Proteins8.2 Prediction of Multi-Label Proteins8.3 Statistical Evaluation Methods8.4 Summary9 Results and Analysis9.1 Performance of GOASVM9.2 Performance of FusionSVM9.3 Performance of mGOASVM9.4 Performance of AD-SVM9.5 Performance of mPLR-Loc9.6 Performance of SS-Loc9.7 Performance of HybridGO-Loc 9.8 Performance of Performance of RP-SVM9.9 Performance of R3P-Loc9.10 Comprehensive Comparison of Proposed Predictors 9.11 Summary10 Discussions10.1 Analysis of Single-label Predictors10.2 Advantages of mGOASVM10.3 Analysis for HybridGO-Loc10.4 Analysis for RP-SVM10.5 Comparing the Proposed Multi-Label Predictors10.6 Summary11 ConclusionsA Web-Servers for Protein Subcellular Localization B Proof of No Bias in LOOCVBibliography

Shibiao Wan, Man-Wai Mak, Hong Kong Polytechnic University, Hong Kong.

• 1 Introduction1.1 Proteins and Their Subcellular Locations1.2 Why Computationally Predicting Protein Subcellular Localization?1.3 Organization of The Thesis2 Literature Review2.1 Sequence-Based Methods2.2 Knowledge-Based Methods2.3 Limitations of Existing Methods3 Legitimacy of Using Gene Ontology Information3.1 Direct Table Lookup?3.2 Only Using Cellular Component GO Terms?3.3 Equivalent to Homologous Transfer?3.4 More Reasons for Using GO Information4 Single-Location Protein Subcellular Localization4.1 GOASVM: Extracting GO from Gene Ontology Annotation Database4.2 FusionSVM: Fusion of Gene Ontology and Homology-Based Features 4.3 Summary5 From Single-Location to Multi-Location 5.1 Significance of Multi-Location Proteins5.2 Multi-Label Classification 5.3 mGOASVM: A Predictor for Both Single- and Multi-Location Proteins5.4 AD-SVM: An Adaptive-decision Multi-Label Predictor5.5 mPLR-Loc: A Multi-Label Predictor Based on Penalized Logistic- Regression 5.6 Summary 6 Mining Deeper on GO for Protein Subcellular Localization6.1 Related Work 6.2 SS-Loc: Using Semantic Similarity Over GO6.3 HybridGO-Loc: Hybridizing GO Frequency and Semantic SimilarityFeatures6.4 Summary7 Ensemble Random Projection for Large-Scale Predictions7.1 Related Work 7.2 RP-SVM: A Multi-Label Classifier with Ensemble Random Projection7.3 R3P-Loc: A Predictor Based on Ridge Regression and Random Projection7.4 Summary 8 Experimental Setup8.1 Prediction of Single-Label Proteins8.2 Prediction of Multi-Label Proteins8.3 Statistical Evaluation Methods8.4 Summary9 Results and Analysis9.1 Performance of GOASVM9.2 Performance of FusionSVM9.3 Performance of mGOASVM9.4 Performance of AD-SVM9.5 Performance of mPLR-Loc9.6 Performance of SS-Loc9.7 Performance of HybridGO-Loc 9.8 Performance of Performance of RP-SVM9.9 Performance of R3P-Loc9.10 Comprehensive Comparison of Proposed Predictors 9.11 Summary10 Discussions10.1 Analysis of Single-label Predictors10.2 Advantages of mGOASVM10.3 Analysis for HybridGO-Loc10.4 Analysis for RP-SVM10.5 Comparing the Proposed Multi-Label Predictors10.6 Summary11 ConclusionsA Web-Servers for Protein Subcellular Localization B Proof of No Bias in LOOCVBibliography

Shibiao Wan, Man-Wai Mak, Hong Kong Polytechnic University, Hong Kong.

• 1 Introduction1.1 Proteins and Their Subcellular Locations1.2 Why Computationally Predicting Protein Subcellular Localization?1.3 Organization of The Thesis2 Literature Review2.1 Sequence-Based Methods2.2 Knowledge-Based Methods2.3 Limitations of Existing Methods3 Legitimacy of Using Gene Ontology Information3.1 Direct Table Lookup?3.2 Only Using Cellular Component GO Terms?3.3 Equivalent to Homologous Transfer?3.4 More Reasons for Using GO Information4 Single-Location Protein Subcellular Localization4.1 GOASVM: Extracting GO from Gene Ontology Annotation Database4.2 FusionSVM: Fusion of Gene Ontology and Homology-Based Features 4.3 Summary5 From Single-Location to Multi-Location 5.1 Significance of Multi-Location Proteins5.2 Multi-Label Classification 5.3 mGOASVM: A Predictor for Both Single- and Multi-Location Proteins5.4 AD-SVM: An Adaptive-decision Multi-Label Predictor5.5 mPLR-Loc: A Multi-Label Predictor Based on Penalized Logistic- Regression 5.6 Summary 6 Mining Deeper on GO for Protein Subcellular Localization6.1 Related Work 6.2 SS-Loc: Using Semantic Similarity Over GO6.3 HybridGO-Loc: Hybridizing GO Frequency and Semantic SimilarityFeatures6.4 Summary7 Ensemble Random Projection for Large-Scale Predictions7.1 Related Work 7.2 RP-SVM: A Multi-Label Classifier with Ensemble Random Projection7.3 R3P-Loc: A Predictor Based on Ridge Regression and Random Projection7.4 Summary 8 Experimental Setup8.1 Prediction of Single-Label Proteins8.2 Prediction of Multi-Label Proteins8.3 Statistical Evaluation Methods8.4 Summary9 Results and Analysis9.1 Performance of GOASVM9.2 Performance of FusionSVM9.3 Performance of mGOASVM9.4 Performance of AD-SVM9.5 Performance of mPLR-Loc9.6 Performance of SS-Loc9.7 Performance of HybridGO-Loc 9.8 Performance of Performance of RP-SVM9.9 Performance of R3P-Loc9.10 Comprehensive Comparison of Proposed Predictors 9.11 Summary10 Discussions10.1 Analysis of Single-label Predictors10.2 Advantages of mGOASVM10.3 Analysis for HybridGO-Loc10.4 Analysis for RP-SVM10.5 Comparing the Proposed Multi-Label Predictors10.6 Summary11 ConclusionsA Web-Servers for Protein Subcellular Localization B Proof of No Bias in LOOCVBibliography