In this dissertation, the problem of learning from highly imbalanced data is studied. Imbalance data learning is of great importance and challenge in many real applications. Dealing with a minority class normally needs new concepts, observations and solutions in order to fully understand the underlying complicated models. We try to systematically review and solve this special learning task in this dissertation.We propose a new ensemble learning framework—Diversified Ensemble Classifiers for Imbal-anced Data Learning (DECIDL), based on the advantages of existing ensemble imbalanced learning strategies. Our framework combines three learning techniques: a) ensemble learning, b) artificial example generation, and c) diversity construction by reversely data re-labeling. As a meta-learner, DECIDL utilizes general supervised learning algorithms as base learners to build an ensemble committee. We create a standard benchmark data pool, which contains 30 highly skewed sets with diverse characteristics from different domains, in order to facilitate future research on imbalance data learning. We use this benchmark pool to evaluate and compare our DECIDL framework with several ensemble learning methods, namely under-bagging, over-bagging, SMOTE-bagging, and AdaBoost. Extensive experiments suggest that our DECIDL framework is comparable with other methods. The data sets, experiments and results provide a valuable knowledge base for future research on imbalance learning. We develop a simple but effective artificial example generation method for data balancing. Two new methods DBEG-ensemble and DECIDL-DBEG are then designed to improve the power of imbalance learning. Experiments show that these two methods are comparable to the state-of-the-art methods, e.g., GSVM-RU and SMOTE-bagging. Furthermore, we investigate learning on imbalanced data from a new angle—active learning. By combining active learning with the DECIDL framework, we show that the newly designed Active-DECIDL method is very effective for imbalance learning, suggesting the DECIDL framework is very robust and flexible.Lastly, we apply the proposed learning methods to a real-world bioinformatics problem—protein methylation prediction. Extensive computational results show that the DECIDL method does perform very well for the imbalanced data mining task. Importantly, the experimental results have confirmed our new contributions on this particular data learning problem.

京都大学 / 0048 / 新制・課程博士 / 博士(医科学) / 甲第21028号 / 医科博第89号 / 新制||医科||6(附属図書館) / 京都大学大学院医学研究科医科学専攻 / (主査)教授 斎藤 通紀, 教授 浅野 雅秀, 教授 玉木 敬二 / 学位規則第4条第1項該当 / Doctor of Medical Science / Kyoto University / DFAM

In this dissertation, the problem of learning from highly imbalanced data is studied. Imbalance data learning is of great importance and challenge in many real applications. Dealing with a minority class normally needs new concepts, observations and solutions in order to fully understand the underlying complicated models. We try to systematically review and solve this special learning task in this dissertation.We propose a new ensemble learning framework—Diversified Ensemble Classifiers for Imbal-anced Data Learning (DECIDL), based on the advantages of existing ensemble imbalanced learning strategies. Our framework combines three learning techniques: a) ensemble learning, b) artificial example generation, and c) diversity construction by reversely data re-labeling. As a meta-learner, DECIDL utilizes general supervised learning algorithms as base learners to build an ensemble committee. We create a standard benchmark data pool, which contains 30 highly skewed sets with diverse characteristics from different domains, in order to facilitate future research on imbalance data learning. We use this benchmark pool to evaluate and compare our DECIDL framework with several ensemble learning methods, namely under-bagging, over-bagging, SMOTE-bagging, and AdaBoost. Extensive experiments suggest that our DECIDL framework is comparable with other methods. The data sets, experiments and results provide a valuable knowledge base for future research on imbalance learning. We develop a simple but effective artificial example generation method for data balancing. Two new methods DBEG-ensemble and DECIDL-DBEG are then designed to improve the power of imbalance learning. Experiments show that these two methods are comparable to the state-of-the-art methods, e.g., GSVM-RU and SMOTE-bagging. Furthermore, we investigate learning on imbalanced data from a new angle—active learning. By combining active learning with the DECIDL framework, we show that the newly designed Active-DECIDL method is very effective for imbalance learning, suggesting the DECIDL framework is very robust and flexible.Lastly, we apply the proposed learning methods to a real-world bioinformatics problem—protein methylation prediction. Extensive computational results show that the DECIDL method does perform very well for the imbalanced data mining task. Importantly, the experimental results have confirmed our new contributions on this particular data learning problem.

T-lymphocytes are an essential cell type of the adaptive immune system. Due to their importance in immune responses and disorders, the molecular mechanisms leading to T-lymphocyte activation have been the subject of extensive research which has translated into important therapeutic developments. Early signalling events involving tyrosine phosphorylation are well characterised. However, later events involving other post-translational modifications are less well understood. Several studies have provided evidence suggesting a role for protein arginine methylation in T-lymphocyte activation. Arginine methylation is an essential post-translational modification in mammals and yet has not been extensively studied. No large scale analysis of arginine methylation sites has been performed. To gain insight into the role of protein arginine methylation in T-lymphocyte activation, the aims of this work were to: 1. Establish whether levels of arginine methylation are altered during Tlymphocyte activation 2. Use mass spectrometry based proteomics to identify arginine methylated proteins in the T-lymphocyte proteome 3. Further characterise an arginine methylated protein important to Tlymphocyte activation Arginine methylation was found to be induced after long term (>20 hours) stimulation of primary T-lymphocytes. Large increases in the main protein arginine methyltransferase, PRMT1, were also observed. Enrichment and labelling methods were developed to detect arginine methylated peptides from T-lymphocytes by mass spectrometry. This resulted in the identification of 265 unique arginine methylation sites in 141 proteins. 204 of the methylation sites were novel and 103 of the proteins had not previously been described as arginine methylated. Individual arginine methylation sites were characterised before and after activation of T-lymphocytes, with some sites showing significant changes in abundance. Among the novel arginine methylated proteins discovered were Dynamin II, WASp and WIPF1. These proteins are involved in re-organisation of the actin cytoskeleton at the immunological synapse formed between a Tlymphocyte and an antigen presenting cell. The functional consequences of the arginine methylation sites inWASp were characterised. WASp is essential for T-lymphocyte activation and some initial evidence showed that one of the arginine methylation sites is important for WASp activation.

Proteinmethylierung spielt eine immer größere Rolle in der Regulierung zellulärer Prozesse. Die Entwicklung effizienter proteomweiter Methoden zur Detektion von Methylierung auf Proteinen ist limitiert und technisch schwierig. In dieser Arbeit haben wir einen neuen Hefe-Zwei-Hybrid-Ansatz (Y2H) entwickelt, der Proteine, die miteinander wechselwirken, mit Hilfe von Sequenzierungen der zweiten Generation identifiziert (Y2H-Seq). Der neue Y2H-Seq-Ansatz wurde systematisch mit dem Y2H-Seq-Ansatz verglichen. Dafür wurde ein Bait-Set von 8 Protein-Arginin-Methyltransferasen, 17 Protein-Lysin-Methyltransferasen und 10 Demethylasen gegen 14,268 Prey-Proteine getestet. Der Y2H-Seq-Ansatz ist weniger arbeitsintensiv, hat eine höhere Sensitivität als der Standard Y2H-Matrix-Ansatz und ist deshalb besonders geeignet, um schwache Interaktionen zwischen Substraten und Protein-Methyltransferasen zu detektieren. Insgesamt wurden 523 Wechselwirkungen zwischen 22 Bait-Proteinen und 324 Prey-Pr oteinen etabliert, darunter 11 bekannte Methyltransferasen-Substrate. Netzwerkanalysen zeigen, dass Methyltransferasen bevorzugt mit Transkriptionsregulatoren, DNA- und RNA-Bindeproteinen wechselwirken. Diese Daten repräsentieren das erste proteomweite Wechselwirkungsnetzwerk über Protein-Methyltransferasen und dienen als Ressource für neue potentielle Methylierungssubstrate. In einem in vitro Methylierungsassay wurden exemplarisch mit Hilfe massenspektrometrischer Analysen die methylierten Aminosäurereste einiger Kandidatenproteine bestimmt. Von neun getesteten Proteinen waren sieben methyliert, zu denen gehören SPIN2B, DNAJA3, QKI, SAMD3, OFCC1, SYNCRIP und WDR42A. Wahrscheinlich sind viele Methylierungssubstrate im Netzwerk vorhanden. Das vorgestellte Protein-Protein-Wechselwirkungsnetzwerk zeigt, dass Proteinmethylierung sehr unterschiedliche zelluläre Prozesse beeinflusst und ermöglicht die Aufstellung neuer Hypothesen über die Regulierung Molekularer Mechanismen durch Methylierung. / Protein methylation on arginine and lysine residues is a largely unexplored posttranslational modification which regulates diverse cellular processes. The development of efficient proteome-wide approaches for detecting protein methylation is limited and technically challenging. We developed a novel workload reduced yeast-two hybrid (Y2H) approach to detect protein-protein interactions utilizing second generation sequencing. The novel Y2H-seq approach was systematically evaluated against our state of the art Y2H-matrix screening approach and used to screen 8 protein arginine methyltransferases, 17 protein lysine methyltransferases and 10 demethylases against a set of 14,268 proteins. Comparison of the two approaches revealed a higher sensitivity of the new Y2H-seq approach. The increased sampling rate of the Y2H-seq approach is advantageous when assaying transient interactions between substrates and methyltransferases. Overall 523 interactions between 22 bait proteins and 324 prey proteins were identified including 11 proteins known to be methylated. Network analysis revealed enrichment of transcription regulator activity, DNA- and RNA-binding function of proteins interacting with protein methyltransferases. The dataset represents the first proteome-wide interaction network of enzymes involved in methylation and provides a comprehensively annotated resource of potential new methylation substrates. An in vitro methylation assay coupled to mass spectrometry revealed amino acid methylation of candidate proteins. Seven of nine proteins tested were methylated including SPIN2B, DNAJA3, QKI, SAMD3, OFCC1, SYNCRIP and WDR42A indicating that the interaction network is likely to contain many putative methyltransferase substrate pairs. The presented protein-protein interaction network demonstrates that protein methylation is involved in diverse cellular processes and can inform hypothesis driven investigation into molecular mechanisms regulated through methylation.