DeepMitoDB

DeepMitoDB provides experimental and predicted sub-mitochondrial localizations and functional annotations for mitochondrial proteins across humans, mice, flies, yeast, and Arabidopsis thaliana.

Key Features:

• Species coverage: Annotations cover mitochondrial proteomes from five species: humans, mice, flies, yeast, and Arabidopsis thaliana.
• Localization types: Includes both experimental and predicted subcellular localizations at sub-mitochondrial resolution.
• Sub-mitochondrial compartments: Assigns proteins to four distinct sub-mitochondrial compartments.
• Prediction algorithm: Uses the DeepMito method based on a 1-Dimensional Convolutional Neural Network (1D-CNN) to predict sub-mitochondrial localization.
• Proteome scale: Integrates annotations for a total of 4,307 mitochondrial proteins across the five species.
• Functional annotation sources: Incorporates functional annotations derived from experimental data and from state-of-the-art prediction tools.
• Novel identifications: Includes proteins lacking prior experimental localization, filling annotation gaps in mitochondrial proteomes.

Scientific Applications:

• Functional annotation: Support for assigning biological roles to mitochondrial proteins by combining localization and functional evidence.
• Comparative proteomics: Enables cross-species comparison of mitochondrial protein localization and composition.
• Annotation gap filling: Identification and characterization of proteins without prior experimental sub-mitochondrial localization data.
• Compartment-specific studies: Facilitates investigation of compartment-specific mitochondrial organization and function.

Methodology:

Annotations were generated by applying the DeepMito method, which leverages a 1-Dimensional Convolutional Neural Network (1D-CNN), to mitochondrial proteomes from five species to assign proteins to four sub-mitochondrial compartments and by integrating functional annotations from experimental data and prediction tools.