In-Pero

In-Pero predicts sub-peroxisomal localization of proteins, distinguishing matrix and membrane compartments to support studies of peroxisomal protein targeting and related disorders.

Key Features:

• Deep Learning Embeddings: Generates multi-dimensional deep-learning embeddings of protein amino acid sequences that capture sequence features relevant to localization.
• High Classification Accuracy: Achieves classification accuracy exceeding 0.9 in distinguishing peroxisomal matrix and membrane proteins.
• Robust Validation Methodology: Employs double cross-validation and is trained and tested on a curated dataset of 160 peroxisomal proteins with experimentally validated sub-peroxisomal localizations.
• Adaptability: Methodology can be adapted to mitochondrial localization prediction, and the In-Mito variant shows superior performance for mitochondrial matrix and inner-membrane proteins compared to existing tools.

Scientific Applications:

• Peroxisomal disorder research: Supports investigation of molecular mechanisms and mislocalization in peroxisomal disorders and the identification of potential therapeutic targets.
• Mitochondrial research: Enables prediction of mitochondrial matrix and inner-membrane localization to inform studies of mitochondrial function and dysfunction.

Methodology:

Generates deep-learning embeddings of protein amino acid sequences, integrates them with standard machine learning approaches, and evaluates performance via double cross-validation on a curated dataset of 160 experimentally validated peroxisomal proteins.