PhosVarDeep

PhosVarDeep predicts phospho-variants from protein sequence information to assess how single nucleotide variants (SNVs) disrupt protein phosphorylation.

Key Features:

• Deep-Learning Architecture: A Siamese-like Convolutional Neural Network (CNN) with two identical subnetworks processes paired reference and variant protein sequences.
• Sequence Feature Extraction: Each subnetwork uses a pre-trained sequence feature encoding network to extract phosphorylation-related sequence features.
• Variant-Aware Phosphorylation Analysis: A CNN module captures variant-specific phosphorylation sequence features to detect subtle differences between reference and variant sequences.
• Integrated Prediction Module: Outputs from both subnetworks are integrated by a prediction module to produce phospho-variant predictions.

Scientific Applications:

• Variant effect prediction: Predicts the impact of SNVs on protein phosphorylation sites within protein sequences.
• Signaling and disease mechanism analysis: Provides information for interpreting how phosphorylation disruptions may alter cellular signaling pathways related to complex diseases.

Methodology:

PhosVarDeep employs a pre-trained sequence feature encoding network followed by a Siamese-like CNN architecture where two subnetworks analyze reference and variant sequences; a CNN module extracts variant-specific phosphorylation features and a prediction module integrates subnetwork outputs to classify phospho-variants.