S-PSorter

S-PSorter predicts protein subcellular localization from immunohistochemistry images by integrating cell-structure relationships to improve classification accuracy.

Key Features:

• Cell structure-driven approach: Incorporates inherent structural relationships among cellular compartments into classifier construction rather than treating compartments as independent classes.
• Error Correcting Output Coding (ECOC): Employs an ECOC framework with a codeword matrix that encodes structural relationships among cellular components.
• Multi-Kernel Support Vector Machine: Trains multiple classifiers using a multi-kernel SVM approach, each classifier corresponding to a column of the ECOC codeword matrix.
• Classifier ensemble via majority voting: Combines outputs of the multiple classifiers through majority voting to produce final predictions.

Scientific Applications:

• Protein subcellular localization: Predicts locations of proteins from immunohistochemistry images to support interpretation of protein function within cellular contexts.
• Human proteome studies: Applies to analysis of the human proteome and characterization of protein functions and interactions in cells.
• Benchmarking on Human Protein Atlas: Evaluated on 1,636 images from the Human Protein Atlas, reporting an overall accuracy of 89.0% versus existing methods.

Methodology:

Construct an ECOC codeword matrix to encode structural relationships among cellular components; train multiple classifiers using multi-kernel SVM corresponding to each codeword column; combine classifier outputs by majority voting to generate final predictions.