Protein abundance inference via expectation-maximization in fluorosequencing

Area of Science:

• Proteomics
• Computational Biology
• Biotechnology

Background:

• Fluorosequencing generates millions of peptide reads, but lacks a robust method for quantitative protein abundance estimation.
• Accurate protein quantification is crucial for understanding biological processes and disease mechanisms.

Purpose of the Study:

• To develop a probabilistic framework for estimating relative protein abundances from fluorosequencing data.
• To evaluate the scalability and accuracy of the proposed method for proteome-wide quantification.

Main Methods:

• Adapted expectation-maximization (EM) algorithm to the fluorosequencing measurement process.
• Utilized posterior peptide probabilities from existing classifiers to estimate protein abundances.
• Evaluated performance using five-protein simulations and full human-proteome simulations.

Main Results:

• The EM-based method significantly reduced mean absolute error in protein abundance estimation compared to a uniform-abundance guess.
• Demonstrated scalability, processing ten million reads from a human proteome simulation in under four hours on a GPU.
• Showed that accuracy gains are modest under current error rates but increase markedly with reduced chemistry error rates.

Conclusions:

• EM-based inference provides a scalable, model-driven solution for protein-level quantification in fluorosequencing.
• The framework can serve as a refinement step for other inference methods.
• Improvements in fluorosequencing chemistry would directly translate to more accurate quantitative proteomics.