Deep learning approaches for efficient interpretation of MS peptide spectra.

点击翻译按钮获取中文摘要

The project is to investigate the application of deep neural networks to identify signals in raw MS spectra that can be reliably associated with specific cell phenotypes and/or disease states. This can potentially identify new complex biomarkers with clinical value and also help to improve the efficiency of analysing MS-based proteomics data by avoiding algorithmic preprocessing of the raw MS data, which is time consuming and introduces sources of error. The project will benefit from the availability of existing very large collections of highly annotated MS proteomics data in the Lamond Laboratory, the ability to generate new targeted data sets using cell models and clinical tissues and access to existing large-scale computational infrastructure.The objectives are to compare different deep learning structures to design neural networks best suited to identifying indicative features within large sets of complex MS-based proteomics data at different levels of granularity (e.g. MS1 spectra, MS2 spectra, peptide IDs, proteins). To apply the resulting optimal neural network to interpret MS data derived from analysis of healthy and diseased cells and thereby identify complex biomarkers diagnostic for specific cell phenotypes. Project provides training in Quantitative & Interdisciplinary Skills.