Discovering protein structure and function through nonlinear system identification

从序列中预测蛋白质的结构和功能对生物医学研究具有重要意义。一个有效的解决方案有可能加速药物发现过程，并可能导致对癌症和其他疾病过程的更多了解。不幸的是，阐明蛋白质结构的实验方法非常昂贵，而且并不总是适用的。计算预测技术提供了一个有吸引力的替代方案；然而，当高度相似的先前已解决的蛋白质结构不可用或结构域更长时，直接从氨基酸序列数据准确预测3D三级蛋白质结构仍然是研究人员的难题。即使到那时，了解蛋白质的三级结构可能还不足以完全了解其功能：大多数真核蛋白质在其原始结构中没有功能，必须通过某种形式的翻译后修饰(PTM)来激活。我们最近开发了一种新的方法来预测蛋白质的二级结构，利用并行串级识别(PCI)，这是一种强大的非线性系统识别方法。结果的准确性不仅与目前最流行的基于神经网络和隐马尔可夫模型的预测方法相媲美，而且与这些方法一起使用时，PCI也被证明是非常有效的。虽然预测二级结构是一个有用的中间步骤，但蛋白质结构预测的最终目标是预测蛋白质活性构象(S)的完整三维结构，包括PTM信息，只需知道其氨基酸序列和环境。基于PCI作为蛋白质组学工具已取得的成功，本文提出的研究计划将在这些新的和具有挑战性的领域取得重大进展，并为蛋白质结构和功能的硅基药物预测做出真正的贡献。

Predicting protein structure and function from sequence is of fundamental importance to biomedical research. An effective solution has the potential to accelerate the drug discovery process and may lead to an increased understanding of cancer and other disease processes. Unfortunately, experimental methods for elucidating a protein's structure are very costly and are not always applicable. Computational prediction techniques provide an attractive alternative; however, the accurate prediction of 3D tertiary protein structure directly from amino acid sequence data continues to elude researchers when highly similar previously solved protein structures are not available or for longer domains. Even then, knowing a protein's tertiary structure may not be sufficient to fully understand its function: the majority of eukaryotic proteins are not functional in their original structure and must be activated through some form of post-translational modification (PTM). We have recently developed a novel approach to protein secondary structure prediction making use of Parallel Cascade Identification (PCI), a powerful method of nonlinear system identification. Resulting accuracy not only compared well with top contemporary prediction methods based on neural networks and hidden Markov models, but PCI was also shown to be highly effective when used in concert with such methods. While predicting secondary structure is a useful intermediate step, the ultimate goal of protein structure prediction is to predict the complete 3D structure of the active conformation(s) of a protein, including PTM information, given knowledge only of its amino acid sequence and its environment. Building on the established success of PCI as a proteomics tool, the research program proposed herein will make significant progress in these new and challenging areas, and make a real contribution towards in silico drug prediction of protein structure and function.