DeepLoc

DeepLoc predicts eukaryotic protein subcellular localization from amino acid sequences using a deep neural network to inform proteomics analyses.

Key Features:

• Sequence-Based Prediction: Predicts localization solely from the amino acid sequence, enabling analysis of proteins lacking annotated homologues and assessment of sequence variants.
• Deep Neural Network Architecture: Employs a recurrent neural network (RNN) that processes entire protein sequences to capture sequential dependencies and contextual information.
• Attention Mechanism: Integrates an attention mechanism to identify specific sequence regions critical for determining subcellular localization, enhancing interpretability.
• Performance: Achieves an overall accuracy of 78% across ten subcellular categories and up to 92% for distinguishing membrane-bound versus soluble proteins.

Scientific Applications:

• Proteomics Research: Provides localization predictions that aid in elucidating protein function and interaction networks within the cell.
• Functional Annotation of Novel Proteins: Supplies localization-based annotations for newly discovered proteins without experimental data to guide experimental validation and functional studies.
• Impact Assessment of Sequence Variants: Predicts how mutations or sequence variants might alter protein localization, informing hypotheses about effects on cellular function and disease mechanisms.

Methodology:

Model employs a recurrent neural network with an integrated attention mechanism and was trained and validated on a UniProt-derived dataset of proteins with experimentally annotated localizations, with performance evaluated against that dataset.