SPIFFED

SPIFFED predicts protein-protein interactions and protein complexes from co-fractionation coupled with mass spectrometry (CF-MS) data by inferring interactomes without handcrafted feature extraction.

Key Features:

• Feature-extraction-free approach: Employs a balanced end-to-end learning architecture that integrates raw CF-MS data representation with interactome prediction using convolutional neural networks (CNNs).
• Improved sensitivity for true PPIs: Demonstrates improved sensitivity for true protein-protein interactions, particularly when trained on balanced datasets, and can outperform existing methods under imbalanced training conditions.
• Ensemble model with voting schemes: Implements an ensemble model with voting schemes to integrate predicted PPIs from multiple CF-MS datasets.
• Integration with clustering software: Integrates with clustering software such as ClusterONE to infer high-confidence protein complexes from predicted interactions.

Scientific Applications:

• Protein complex inference: Infers high-confidence protein complexes from CF-MS interaction predictions.
• Global interactome reconstruction: Enables global interactome inference from CF-MS datasets to map protein interaction networks.
• Study of dynamic cellular interactions: Supports analysis of dynamic interactions within cells to aid understanding of complex biological systems.

Methodology:

Processes raw CF-MS data directly with convolutional neural networks (CNNs) in a balanced end-to-end learning architecture that bypasses handcrafted feature extraction; employs ensemble models with voting schemes to combine predictions across CF-MS datasets; is compatible with ClusterONE for clustering predicted interactions into protein complexes; and uses training on balanced datasets to improve sensitivity and mitigate false positives from non-interacting proteins co-eluting by chance.