Automated Retrieval of Heterogeneous Proteomic Data for Machine Learning

Abstract

Proteomics instrumentation and the corresponding bioinformatics tools have evolved at a rapid pace in the last 20 years, whereas the exploitation of deep learning techniques in proteomics is on the horizon. The ability to revisit proteomics raw data, in particular, could be a valuable resource for machine learning applications seeking new insight into protein expression and functions of previously acquired data from different instruments under various lab conditions. We map publicly available proteomics repositories (such as ProteomeXchange) and relevant publications to extract MS/MS data to form one large database that contains the patient history and mass spectrometric data acquired for the patient sample. The extracted mapped dataset should enable the research to overcome the issues attached to the dispersions of proteomics data on the internet, which makes it difficult to apply emerging new bioinformatics tools and deep learning algorithms. The workflow proposed in this study enables a linked large dataset of heart-related proteomics data, which could be easily and efficiently applied to machine learning and deep learning algorithms for futuristic predictions of heart diseases and modeling. Data scraping and crawling offer a powerful tool to harvest and prepare the training and test datasets; however, the authors advocate caution because of ethical and legal issues, as well as the need to ensure the quality and accuracy of the data that are being collected.