DeepCellEss: cell line-specific essential protein prediction with attention-based interpretable deep learning.

Area of Science:

• Computational biology
• Genomics
• Proteomics

Background:

• Protein essentiality is a conditional trait influenced by cellular environments.
• Current computational methods often fail to account for cell-specific factors.
• Limited model interpretability hinders the analysis of essential protein predictions.

Purpose of the Study:

• To develop DeepCellEss, a sequence-based interpretable deep learning framework for cell line-specific essential protein predictions.
• To address the limitations of existing methods by incorporating cell-specific characteristics and enhancing model interpretability.

Main Methods:

• Utilized a convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM) to learn from protein sequences.
• Incorporated a multi-head self-attention mechanism for residue-level interpretability.
• Trained the model on a large-scale benchmark dataset comprising 323 cell lines.

Main Results:

• DeepCellEss demonstrated effective prediction performance across different cell lines.
• The framework outperformed existing sequence-based and network-based centrality measures.
• Case studies highlighted the importance of cell-specific predictions and DeepCellEss's superiority.

Conclusions:

• DeepCellEss provides accurate and interpretable predictions of essential proteins.
• The framework underscores the necessity of cell line-specific analysis in predicting protein essentiality.
• DeepCellEss serves as a valuable tool for essential protein prediction in diverse cellular contexts.