大规模蛋白质功能预测的高效算法研究

大规模蛋白质功能的快速、准确预测对于揭示生命奥秘，阐明疾病发生发展机制，研制新药等意义重大。随着蛋白质序列数量加速增长，以及基因本体被广泛地应用于蛋白质功能标注，本项目的主要目标是研究大规模蛋白质功能预测的高效且可靠的新算法，提高预测精度，为进一步的生物学实验验证提供更精准的指导，从而更好地服务于广大生物医学研究人员。本课题以蛋白质的功能预测为研究对象，在机器学习的理论框架指导下，首先设计基于表示学习的算法获得蛋白质的合适表示弥补传统特征抽取方法的不足；然后设计基于多标记学习的算法有效处理高达4万多个基因本体标签的预测；最后设计基于排序学习的算法有机融合多源异构信息、不同特征表示和预测方法，以提高预测精度。拟解决的科学问题包括：（1）如何结合蛋白质分子特点设计合适的特征表示；（2）如何设计考虑不同基因本体标签之间相关性的高效预测算法；（3）如何有效融合多源信息、不同特征表示和预测算法。

The rapid and accurate prediction of large-scale protein function is very important to the understanding the secret of the life, elucidating the mechanism of disease, and the drug discovery. With the accelerating growth of protein sequences, and the wide application of Gene Ontology in annotating protein function, the main objective of this project is to develop effective and reliable new algorithms for large-scale protein function prediction, to improve the accuracy of prediction, which can guide the biologists in biochemical experimental verification, and thus provide a better service for the biomedical researchers. The main idea is to develop an accurate algorithm based on machine learning. Firstly, we try to obtain a better representation of protein based on representation learning; Secondly, we will design an efficient algorithm to handle more than 40,000 Gene Ontology terms by multi-label learning; Finally, based on learning to rank, we will try to improve the prediction performance by integrating heterogeneous information sources, different types of representation, and predicting methods. Here are three key scientific problems to solve: (1) how to design a better representation by considering the characteristics of proteins; (2) how to design an efficient predication algorithm by considering the dependency of Gene Ontology terms; 3) How to effectively integrate heterogeneous information sources, different representations and prediction methods.