A Bayesian mixture modelling approach for spatial proteomics.

Area of Science:

• Proteomics
• Cell Biology
• Computational Biology

Background:

• Understanding protein function requires analyzing their spatial distribution within cells.
• Proteins can exist in multiple locations, dynamically change locations, or reside in unknown compartments, complicating single-location assignments.
• Current mass spectrometry (MS)-based spatial proteomics uses machine learning but fails to quantify the uncertainty in sub-cellular location assignments.

Purpose of the Study:

• To develop a novel statistical framework for probabilistic sub-cellular protein localization.
• To introduce uncertainty quantification into spatial proteomics analysis.
• To provide a more informative and flexible approach compared to existing methods.

Main Methods:

• Developed a Bayesian generative classifier using Gaussian mixture models.
• Employed Expectation-Maximization (EM) and Markov-chain Monte-Carlo (MCMC) for Bayesian computation.
• Reformulated the statistical analysis framework for MS-based spatial proteomics data.

Main Results:

• The proposed Bayesian model assigns proteins probabilistically to sub-cellular locations, providing a probability distribution.
• The methodology enables proteome-wide uncertainty quantification, adding a new layer to spatial proteomics analysis.
• The Bayesian approach performs competitively with state-of-the-art machine learning methods while providing additional information and resolving ambiguities where other methods fail.

Conclusions:

• This work presents the first Bayesian model for MS-based spatial proteomics data.
• The flexible framework allows analysis of diverse biological systems and opens new modeling avenues.
• Probabilistic assignment and uncertainty quantification offer deeper biological insights into protein localization and function.