DeepSP: A Deep Learning Framework for Spatial Proteomics.

Area of Science:

• Proteomics
• Cell Biology
• Bioinformatics

Background:

• Protein subcellular localization (PSL) is crucial for understanding protein function.
• Mass spectrometry (MS)-based spatial proteomics offers high-throughput PSL prediction.
• Existing machine learning predictors have limitations in PSL annotation accuracy.

Purpose of the Study:

• To develop a novel deep learning framework, DeepSP, for improved PSL prediction.
• To enhance the accuracy and robustness of PSL prediction using MS-based spatial proteomics data.

Main Methods:

• DeepSP utilizes a novel feature map derived from a difference matrix of protein occupancy profiles.
• Incorporates a convolutional block attention module to refine predictions.
• Evaluated on independent test sets and for predicting unknown PSLs.

Main Results:

• DeepSP demonstrated significant improvements in accuracy and robustness compared to existing methods.
• The framework effectively captures detailed changes in protein distribution across subcellular fractions.
• Achieved superior performance in predicting both known and unknown PSLs.

Conclusions:

• DeepSP provides an efficient and robust framework for PSL prediction in spatial proteomics.
• Facilitates deeper insights into protein functions and biological process regulation.
• Expected to advance the field of spatial proteomics research.