Automated Retrieval of Heterogeneous Proteomic Data for Machine Learning.

Area of Science:

• Proteomics
• Bioinformatics
• Machine Learning
• Deep Learning
• Computational Biology

Background:

• Proteomics instrumentation and bioinformatics tools have advanced significantly.
• Deep learning applications in proteomics are emerging.
• Existing proteomics data is dispersed across various repositories and formats, hindering advanced analysis.

Purpose of the Study:

• To develop a workflow for consolidating publicly available proteomics data.
• To create a unified, linked dataset for machine learning applications.
• To facilitate the use of deep learning for heart disease prediction and modeling.

Main Methods:

• Mapping publicly available proteomics repositories (e.g., ProteomeXchange) and publications.
• Extracting tandem mass spectrometry (MS/MS) data and patient history.
• Developing a data scraping and crawling strategy to build a comprehensive database.

Main Results:

• A proposed workflow enables the creation of a large, linked dataset of heart-related proteomics data.
• The consolidated dataset can be efficiently applied to machine learning and deep learning algorithms.
• Overcomes challenges associated with data dispersion on the internet.

Conclusions:

• The developed workflow facilitates the application of advanced bioinformatics tools to proteomics data.
• This approach holds potential for future predictions and modeling of heart diseases.
• Emphasizes the need for caution regarding ethical, legal, and data quality issues in data harvesting.