Integrative machine learning methods for prediction of protein-protein interactions and analysis of the dynamics of interactomes

蛋白质是构成任何生物体或生物系统的细胞机械主体的大分子，在细胞形态学和生理学、DNA合成、转录、翻译、剪接等基本和必要的生物过程中起着重要作用。然而，蛋白质不是孤立地起作用，而是通过与DNA、RNA和其他蛋白质等分子相互作用来发挥其功能。相互作用组旨在研究生命系统中蛋白质相互作用的主要方面。交互组是动态的，因为交互和最终的功能是以时间和空间的方式表现出来的。为了了解生物系统中复杂的细胞机制，有必要在分子水平上研究这些相互作用的性质和特异性以及参与其中的动力学，其中蛋白质-蛋白质相互作用（PPIs）的预测起着重要作用。**虽然PPIs的预测已经从许多不同的角度进行了研究，以解决不同的问题，但研究的主要方面包括：界面位点（where）、蛋白质在复合体中的排列（how - aka docking）、蛋白质复合体的类型（what）、分子相互作用事件（if）和时空趋势（dynamics）。该建议主要关注ppi的类型、相互作用事件和时间方面的预测，更具体地说，是设计机器学习方法，从高通量数据中预测和分析ppi，了解这些相互作用的动态方面及其与基因组和转录特征的关系。本研究的主要目标是：(1)开发新的机器学习方法来预测高通量ppi，其中包括基于结构和序列的信息；(2)分析和阐明与高通量ppi相关的动力学的主要特性；(3)将新一代测序技术的转录组学数据与相互作用组学相结合，用于检测前列腺癌和乳腺癌的生物标志物和理解转录组学机制。**预测相互作用并揭示PPIs的重要特性和动态方面将有助于理解细胞和分子机制，改善疾病诊断和治疗以及药物开发。这将为生物学、生物化学和生命科学的研究人员提供有价值的信息。此外，转录组学的综合方法在前列腺癌等应用中是新颖的，这是申请人的合作项目之一。与目前将生物标记物（嵌合事件或剪接）与基因联系起来的方法相比，这将带来更好的见解。然而，通过结构域或短基序与其他蛋白质相互作用的异构体或蛋白质形式是相互作用网络的一部分，也是动态的。整合转录组学和相互作用组学以产生更强大的生物标志物的优势将有利于疾病筛查、诊断、治疗和随访。癌症研究的主要目标之一是在早期阶段进行检测，在许多情况下，癌症患者被过度治疗。前列腺癌和乳腺癌是加拿大最常见的癌症类型，在过去十年中一直是癌症死亡的主要原因之一。使用机器学习技术来预测和发现可靠的生物标志物可以改善癌症诊断，特别是在早期发现方面。通过多学科合作，相互作用组学和转录组学的研究成果将有助于开发新的治疗策略和降低癌症的复发率。

Proteins are large molecules that constitute the body of cellular machinery of any living organism or biological system, playing important roles in fundamental and essential biological processes such as cellular morphology and physiology, DNA synthesis, transcription, translation, splicing and many others. Proteins, however, do not act in isolation but perform their functions by interacting with molecules such as DNA, RNA, and other proteins. The interactome aims to study the main aspects of protein interactions in a living system. The interactome is rather dynamic as the interactions and ultimately functions are manifested in a temporal and spatial manner. To understand the complex cellular mechanisms involved in a biological system, it is necessary to study the nature and specificity of these interactions and the dynamics involved in it at the molecular level, for which prediction of protein-protein interactions (PPIs) has played a significant role. **Although prediction of PPIs has been studied from many different perspectives in solving different problems, the main aspects that are studied include: sites of interfaces (where), arrangement of proteins in a complex (how - aka docking), type of protein complex (what), molecular interaction event (if), and temporal and spatial trends (dynamics). This proposal focuses mostly on prediction of types, interaction events and temporal aspects of PPIs, more specifically, on devising machine learning approaches for prediction and analysis of PPIs from high-throughput data, understanding the dynamic aspects of these interactions and their relationships with genomic and transcriptional features. The main goals of this research are to: (1) develop new machine learning approaches for prediction of high-throughput PPIs, which include structural and sequence-based information; (2) analyze and elucidate the main properties of the dynamics associated with high-throughput PPIs; (3) integrate transcriptomics data from next generation sequencing techniques with interactomics in applications for detecting biomarkers and understanding the transcriptogenomics mechanisms involved in prostate and breast cancer. **Predicting interactions and unraveling the important properties and dynamic aspects of PPIs will help understand cellular and molecular mechanisms, improving disease diagnosis and treatment, and drug development. This will provide valuable information for researchers in biology, biochemistry and life sciences. In addition, integrative approaches for transcriptomics are novel in applications such as prostate cancer, which is one of the applicant's collaborative projects. This will bring a much better insight than the current approaches that associate biomarkers (chimeric events or splicing) with genes. However, isoforms or forms of proteins that interact with other proteins via domains or short motifs are part of interaction networks which are also dynamic. The advantage of integrating transcriptomics and interactomics to produce more robust biomarkers will benefit disease screening, diagnosis, treatment and follow-up. One of the primary goals in cancer studies is detection at early stages, and in many cases, cancer patients are over-treated. Prostate and breast cancer are among the most common types of cancer in Canada, and have been among the main causes of cancer death over the past ten years. Using machine learning techniques for prediction and discovery of reliable biomarkers can lead to improvements in cancer diagnosis, especially in early detection. By means of multi-disciplinary collaborations, the findings in interactomics and integration with transcriptoimcs will benefit the development of novel therapeutic strategies and lower recurrence in cancer.