Excellence in Research: Developing a Knowledge Graph Driven Integrative Framework for Explainable Protein Function Prediction via Generative Deep Learning

蛋白质是生命的基石，具有多种功能，包括但不限于酶催化反应，抗体参与人体防御机制，形成结构和运输重要的化学物质。蛋白质之间的相互作用描述了疾病的分子机制，并为疾病的预防、诊断和治疗提供了潜在的重要见解。因此，蛋白质的功能表征对于帮助理解生命、疾病和开发危及生命的疾病的新疗法至关重要。尽管最近取得了一些进展，但由于性能低、缺乏可解释的结果以及不可重复的研究传播，预测蛋白质功能仍然是一个悬而未决的问题，这突出了利用最近关于蛋白质的生物医学数据激增改进方法的必要性。本研究的目的是设计、实现和评估一个蛋白质功能预测管道，使用一种新的由异构知识图驱动的生成式深度学习方法来解决多组学数据集成、可解释功能预测和可重复性的挑战。本研究将通过三个相互关联的任务进行：1)研究一种新的基于知识图谱的生成式深度学习模型；2)通过大语言模型整合多组学特征；3)可复制软件的开发。该项目的成功完成将导致一个强大的、更准确的、可重复的和可解释的蛋白质功能预测管道。该项目将创造新的教育和推广机会，以极大地加强Meharry医学院利用现代人工智能技术在计算生物学方面的培训和研究活动。Meharry主要招收非裔美国学生。Meharry大学90%以上的数据科学专业学生是非裔美国人，其中大多数是女性。该项目将提高梅哈里大学女性和少数族裔学生在人工智能/机器学习、定量基因组学和数据科学研究方面的STEM教育意识、影响和机会。可复制的开源软件将极大地促进更广泛的科学界致力于改进蛋白质功能预测。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Proteins are the building blocks of life performing multitudes of functions that include but not limited to catalyzing reactions as enzymes, participating in the body’s defense mechanism as antibodies, forming structures and transporting important chemicals. The interactions among proteins describe the molecular mechanism of diseases, and convey potentially important insights about the disease prevention, diagnosis, and treatments. Therefore, functional characterization of proteins is crucial to helping understand life, diseases, and developing novel treatments for life threatening illness. Despite recent advancements, predicting protein function remains an open problem due to low performance, lack of explainable outcomes, and irreproducible research dissemination highlighting the need for improved methodologies leveraging the recent proliferation of biomedical data about proteins. The objective of this research is to design, implement, and evaluate a protein function prediction pipeline using a novel generative deep learning approach powered by heterogeneous knowledge graph to address the challenge of multi-omics data integration, explainable function prediction, and reproducibility. The research will be carried out through three interrelated tasks: 1) investigation of a novel generative deep learning model on knowledge graph; 2) integration of multi-omics features through large language model; and, 3) development of reproducible software. Successful completion of this project will lead to a robust, more accurate, reproducible and explainable protein function prediction pipeline. The project will create new education and outreach opportunities to greatly strengthen the training and research activities in computational biology leveraging modern AI technologies at Meharry Medical College, a leading HBCU. Meharry dominantly enrolls African American students. More than 90% data science students at Meharry are African Americans and majority are women. This project will increase STEM education awareness, impact, and opportunity to the women and minority students at Meharry to excel in AI/ML, quantitative genomics and data science research. The reproducible open-source software will greatly facilitate broader scientific community working to improve protein function prediction.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.