iDeepSubMito

iDeepSubMito predicts submitochondrial localization of proteins using deep learning-based sequence representations and neural architectures to assign proteins to mitochondrial matrix, inner membrane, outer membrane, or intermembrane regions.

Key Features:

• Protein Sequence Representation: iDeepSubMito employs ProteinELMo to model probability distributions over protein sequences and represent sequences as continuous embedding vectors.
• Neural Network Architecture: The core model is a convolutional neural network that integrates a bidirectional Long Short-Term Memory (LSTM) network with a self-attention mechanism to capture contextual and semantic sequence features.
• Submitochondrial Compartments: The method targets four mitochondrial compartments: matrix, inner membrane, outer membrane, and intermembrane space.
• Cross-validation Evaluation: Performance was assessed by cross-validation on two datasets comprising 424 and 570 proteins, respectively.
• External Validation Datasets: Additional validation was conducted on M187, M983, and MitoCarta3.0 datasets.
• Performance and Interpretability: The method reported improved prediction performance over existing computational approaches and includes motif analysis and interpretability studies to provide biological insights.

Scientific Applications:

• Functional Genomics: Assigning proteins to submitochondrial compartments to support studies of mitochondrial protein function and organization.
• Disease Research: Investigating submitochondrial localization patterns relevant to mitochondrial dysfunctions implicated in neurodegenerative and metabolic diseases.
• Drug Discovery: Prioritizing potential drug targets by localizing candidate proteins to specific mitochondrial compartments.

Methodology:

ProteinELMo embeddings were used for sequence encoding; a CNN integrating bidirectional LSTM and self-attention performed predictions; model evaluation used cross-validation on datasets of 424 and 570 proteins and external validation on M187, M983, and MitoCarta3.0, alongside motif analysis and interpretability studies.