Machine Learning on Large-Scale Proteomics Data Identifies Tissue and Cell-Type Specific Proteins.

Area of Science:

• Proteomics
• Machine Learning
• Bioinformatics

Background:

• Accurate identification of tissue and cell types is crucial for biological research and clinical applications.
• Existing methods for proteomic analysis can be limited in specificity and scalability.
• Publicly available datasets offer a valuable resource for developing advanced analytical tools.

Purpose of the Study:

• To develop and validate a machine learning model for classifying human tissues and cell types based on protein abundance data.
• To identify key protein markers that distinguish specific tissues and cell types.
• To assess the model's performance in accurately predicting sample origin.

Main Methods:

• Utilized 183 public human datasets from the PRIDE database.
• Employed ionbot for data searching and manually annotated tissue/cell type information.
• Trained a Random Forest model using protein abundances from physiological samples.
• Applied one-vs-all classification and feature importance analysis to identify discriminating proteins.

Main Results:

• The machine learning model achieved 98% accuracy in predicting tissues and 99% accuracy in predicting cell types based on protein abundance alone.
• F-scores clearly delineated tissue-specific proteins and expression patterns.
• Feature analysis revealed the model's ability to detect biologically relevant patterns, with minor confusion between physiologically similar tissues.

Conclusions:

• Protein abundance patterns are highly predictive of human tissue and cell types.
• The developed machine learning model demonstrates significant potential for applications in liquid biopsy analysis and studying organoid/cell line proteomes.
• This approach offers a robust method for identifying the origin of proteins in complex biological samples.