CIBR: Full-Spectrum Prediction of Peptide Tandem Mass Spectra using Deep Neural Networks

过去十年见证了质谱（MS）技术的快速发展。特别是液相色谱耦合串联质谱（LC-MS/MS）已成为微生物学、环境科学、植物生物学、农业和生物医学等生命科学许多分支中表征复杂蛋白质样品的流行分析工具。该项目旨在利用公开可用的蛋白质组学数据来预测肽的串联质量（MS/MS）光谱。成功预测多肽MS/MS谱具有重要的理论意义（有助于更好地理解质谱仪中多肽断裂的机制），并将显著提高多肽鉴定，这对复杂蛋白质样品的分析至关重要。该项目的pi积极参与并领导一些学校和部门的外展活动和活动，包括每年的女童子军夏令营。他们计划每年从HBCU各研究所招收学生参加该项目的暑期研究。该项目的pi提出了一个序列到序列（seq2seq）深度学习模型，用于直接从其序列预测肽的完整MS/MS谱，而无需假设片段规则。他们还将利用多任务学习（MTL）方法来预测含有翻译后修饰（PTMs）的肽的MS/MS谱，以及通过使用不同的离子激活方法（如电子转移解离（ETDs））获得的MS/MS谱。深度学习模型将在开源软件工具中实现并发布，供研究社区使用。该研究项目的最新情况将在项目网站上公布：http://www.predfull.com.This该奖项反映了美国国家科学基金会的法定使命，并通过基金会的智力价值和更广泛的影响审查标准进行评估，认为值得支持。

The last decade has witnessed rapid advances in mass spectrometry (MS) technology. In particular, the liquid chromatograph coupled tandem mass spectrometry (LC-MS/MS) has become a popular analytical tool for characterizing complex protein samples in many branches of life sciences, including microbiology, environmental science, plant biology, agriculture and biomedicine. This project aims to exploit publicly available proteomic data for predicting tandem mass (MS/MS) spectra of peptides. Successful prediction of peptide MS/MS spectra is of great theoretical interests (for better understanding mechanisms of peptide fragmentation in mass spectrometers), and will significantly improve the peptide identification, which is critical for the analyses of complex protein samples. The PIs of this project are actively involved and lead some the school and departments outreach activities and events, including the annual summer camp for girl scouts. They plan to recruit students from HBCU institutes to participate summer research each year in this project.The PIs of the project propose to a sequence-to-sequence (seq2seq) deep learning model for predicting the full MS/MS spectra of peptides directly from their sequences without any assumption of fragmentation rules. They will also exploit the multitask learning (MTL) approach for predicting the MS/MS spectra from peptides containing post-translation modification (PTMs), and the MS/MS spectra acquired by using different ion activation methods, such as Electron Transfer Dissociation (ETDs). The deep learning models will be implemented and released in open source software tools to be used by the research community. The update of the research project will be made available through the project website: http://www.predfull.com.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.