mLASSO-Hum: A LASSO-based interpretable human-protein subcellular localization predictor.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Accurate human protein subcellular localization is crucial for understanding disease mechanisms.
• Gene Ontology (GO) based methods outperform sequence-based methods but often lack interpretability and suffer from overfitting.
• Existing predictors struggle with high-dimensional feature vectors derived from GO terms.

Purpose of the Study:

• To develop an interpretable, large-scale predictor for human protein subcellular localization.
• To address the interpretability and overfitting issues in current GO-based methods.
• To identify key Gene Ontology (GO) terms critical for subcellular localization prediction.

Main Methods:

• Proposed mLASSO-Hum, an interpretable multi-label predictor utilizing one-vs-rest LASSO-based classifiers.
• Identified 87 significant GO terms out of over 8000 for predicting subcellular localization.
• Incorporated GO hierarchical information using depth distance to leverage remaining GO terms.

Main Results:

• mLASSO-Hum provides sparse and interpretable solutions for human protein subcellular localization.
• Identified 87 essential GO terms that explain protein localization and its underlying reasons.
• Demonstrated significantly improved performance compared to state-of-the-art predictors.
• Revealed the importance of GO terms from all three categories (not just cellular component) in classification.

Conclusions:

• mLASSO-Hum offers a powerful and interpretable approach for predicting human protein subcellular localization.
• The identified key GO terms provide insights into the 'why' behind protein localization.
• The method effectively utilizes GO hierarchical information, improving predictive accuracy.