DeepEP

DeepEP predicts essential proteins by combining graph embeddings from protein-protein interaction (PPI) networks and multi-scale convolutional neural networks on gene expression profiles to identify proteins required for cellular survival.

Key Features:

• Node2Vec embedding: Applies the node2vec algorithm to PPI networks to automatically learn dense vectors that capture topological and semantic properties of proteins.
• Multi-scale CNNs on gene expression: Treats gene expression profiles as image-like input and uses multi-scale convolutional neural networks to extract hierarchical patterns and features.
• Sampling for class imbalance: Implements a per-epoch sampling strategy that balances training by selecting equal numbers of majority and minority class samples to mitigate imbalance bias.

Scientific Applications:

• Essential protein prediction: Integrates PPI-derived embeddings and gene expression features to predict essential proteins implicated in cellular viability.
• Benchmarking against alternative methods: Serves as a deep learning alternative to traditional centrality measures and shallow machine learning approaches for identifying essential proteins.

Methodology:

Node2vec is applied to PPI networks to generate embeddings; multi-scale CNNs are used on gene expression profiles treated as images to extract features; and a sampling method balances classes during training by selecting equal numbers of majority and minority samples each epoch.