Incorporating organelle correlations into semi-supervised learning for protein subcellular localization prediction

Area of Science:

• Computational Biology
• Bioinformatics
• Cell Biology

Background:

• Automated prediction of protein subcellular localization is crucial but hindered by limited high-quality annotated images.
• The small sample size learning problem and accurate annotation of multi-location proteins remain significant challenges.

Purpose of the Study:

• To develop a novel protein subcellular localization prediction pipeline.
• To address the limitations of small sample size learning and multi-location protein annotation.

Main Methods:

• Integration of five semi-supervised algorithms to leverage lower-quality data.
• Development of a multi-label classification approach incorporating organelle correlations using Bayesian networks.
• Utilizing Bayesian network topology to guide classifier training order for label dependence.

Main Results:

• The proposed pipeline demonstrated efficiency in predicting protein subcellular localization.
• Experimental results validated the effectiveness on both immunohistochemistry and immunofluorescence images.
• The approach successfully handled small sample size learning and multi-location protein annotation.

Conclusions:

• The developed pipeline offers an effective solution for protein subcellular localization prediction.
• The integration of semi-supervised learning and organelle correlation modeling improves prediction accuracy.
• This work advances automated protein localization prediction, particularly for challenging multi-location proteins.