DeepMito

DeepMito predicts protein sub-mitochondrial localization (matrix, outer membrane, inner membrane, intermembrane space) using a 1-dimensional convolutional neural network to provide fine-grained functional annotation of mitochondrial proteins.

Key Features:

• Deep Learning Architecture: Employs a 1-Dimensional Convolutional Neural Network (1D-CNN) for sequence-based classification of sub-mitochondrial compartments.
• High-Quality Training Data: Trained on a dataset of 424 mitochondrial proteins with experimentally validated sub-organelle localizations.
• Cross-Validation Testing: Performance was evaluated using cross-validation and reported to outperform other recent approaches for sub-mitochondrial localization prediction.
• DeepMitoDB: Integrates fine-grain localization information with functional annotations and predicted Gene Ontology (GO) terms.

Scientific Applications:

• Genomic-Scale Annotation: Enables genomic-scale predictions on highly curated datasets of human mitochondrial proteins.
• Multi-Species Analysis: Applied to mitochondrial proteomes from humans (Homo sapiens), mouse (Mus musculus), fly (Drosophila), yeast (Saccharomyces cerevisiae), and Arabidopsis thaliana.
• Functional Characterization: Supports functional characterization by providing precise sub-mitochondrial localizations and associated GO term annotations.

Methodology:

Data collection: assembling a high-quality dataset of 424 proteins with experimentally validated localizations. Model training: training 1D-CNNs on this dataset. Validation: employing cross-validation to assess prediction robustness and accuracy.