SIFTER: A Systems Biology Platform for Protein Function Prediction

蛋白质是参与细胞内几乎所有过程的关键生物分子。如代谢、细胞信号、免疫反应等，蛋白质功能的知识对于获得对细胞活动的基本了解至关重要。由于核苷酸测序技术的最新进展，可用基因组序列的数量大约每12个月翻一番，这一惊人的速度远远超过了摩尔定律。破译蛋白质功能所需的实验技术却没有这么快的进展。事实上，尽管在综合Uniprot数据库中大约有1000万个蛋白质序列，但只有0.2%有实验验证的功能注释。这种序列与函数之间的差距正在迅速扩大，而计算方法的发展对于有效利用这些大量的序列数据至关重要。在这项工作中，我们开发了SIFTER，这是一个大规模的系统生物学平台，可以从高通量数据中准确预测蛋白质功能。基于一个有前途的基于系统基因组学的原型，我们将交互网络纳入我们的模型以提高性能。交互数据本质上耦合了这样的网络中成千上万的蛋白质，我们使用变分推理和并行实现来解决这个具有挑战性的计算问题。我们还探索了基于低秩矩阵分解的函数预测技术，并在此过程中引入了新的基于采样的方法来加快计算速度。此外，我们开发了算法来量化SIFTER预测中的不确定性，以帮助指导未来的实验工作。这些新算法是对经典自举抽样的大规模扩展，一般适用于任何涉及海量数据的问题。最后，我们与实验生物学家合作评估SIFTER，使我们能够确定相关的用例，并在生物医学界产生广泛影响的有效方法。

Proteins are key biomolecules involved in virtually all processes within cells,e.g., metabolism, cell signaling, immune response, etc., and knowledge ofprotein function is vital to obtain a basic understanding of cellular activity.Due to recent advances in nucleotide sequencing technology, the number ofavailable genomic sequences is doubling in size roughly every 12 months, anincredibly fast pace vastly exceeding Moore's law. Experimental technologiesrequired to decipher protein function have not progressed nearly as fast. Infact, although there are roughly 10 million protein sequences in thecomprehensive Uniprot database, only 0.2% have experimentally validatedfunction annotations. This sequence-function gap is rapidly expanding, and thedevelopment of computational methods is of crucial importance to effectivelyutilize this deluge of sequence data.In this work, we develop SIFTER, a large-scale, systems biology platform toaccurately predict protein function from high-throughput data. Building upon apromising phylogenomic-based prototype, we incorporate interaction networksinto our model to improve performance. Interaction data intrinsically couplesthe thousands to millions of proteins within such networks, and we usevariational inference and parallelized implementations to address thischallenging computational problem. We also explore techniques for functionprediction based on low-rank matrix factorization, and along the way, introducenovel sampling-based approaches to speed up computation. Additionally, wedevelop algorithms to quantify uncertainty in SIFTER's predictions tohelp guide future experimental work. These novel algorithms are large-scaleextensions to classical bootstrap sampling and are generally applicable to anyproblem involving massive data. Finally, we evaluate SIFTER incollaboration with experimental biologists, allowing us to pinpoint relevantuse cases and resulting in an effective method with widespread impact withinthe biomedical community.