DULoc: quantitatively unmixing protein subcellular location patterns in immunofluorescence images based on deep...

Area of Science:

• Cell Biology
• Proteomics
• Bioimaging

Background:

• Understanding protein subcellular localization is crucial for deciphering protein function and complex cellular processes.
• Multi-label proteins, residing in multiple locations, present challenges due to limitations in current descriptive annotations.
• Quantitative analysis of protein spatial distribution is essential for a deeper understanding of their functional mechanisms.

Purpose of the Study:

• To develop a quantitative method for estimating protein fractions across subcellular compartments using immunofluorescence images.
• To address the limitations of descriptive annotations for multi-label proteins.
• To enable a more comprehensive understanding of protein spatial distribution and function.

Main Methods:

• Developed a deep-learning-based pattern unmixing pipeline named DULoc.
• Utilized a deep convolutional neural network for feature representation.
• Integrated multiple nonlinear decomposing algorithms for pattern unmixing.

Main Results:

• DULoc achieved a correlation greater than 0.93 between estimated and true protein fractions on real and synthetic datasets.
• Large-scale application on the Human Protein Atlas showed consistent location orders for 70.52% of proteins compared to database annotations.
• The method provides quantitative insights into protein localization patterns.

Conclusions:

• DULoc offers a robust deep-learning approach for quantitative protein subcellular localization.
• The tool enhances the analysis of multi-label proteins by estimating fractional localization.
• This quantitative approach advances the understanding of protein function and cellular mechanisms.