An interpretable deep learning model for classifying adaptor protein complexes from sequence information

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Adaptor proteins (APs) are crucial for intracellular membrane trafficking.
• Defects in APs are linked to various human disorders.
• Existing methods for AP identification are time-consuming and lack comprehensive strategies for subtype complex discrimination.

Purpose of the Study:

• To develop a novel computational method for discriminating AP complexes within the AP family.
• To create a benchmark dataset for AP complex classification.
• To enhance the understanding of AP functions and their role in diseases.

Main Methods:

• Utilized an interpretable deep neural network architecture.
• Employed sequence-based encoding features for AP complex recognition.
• Developed a benchmark dataset from UniProt and GeneOntology databases.
• Compared performance against various machine learning algorithms and feature extraction strategies.
• Applied t-SNE, UMAP, and SHAP for model interpretation and feature visualization.

Main Results:

• Achieved promising performance in distinguishing AP complexes.
• Demonstrated the robustness of the proposed deep learning method.
• Provided insights into the distribution of AP complexes on optimal features through visualization techniques.
• Established a new benchmark dataset for AP complex analysis.

Conclusions:

• The proposed interpretable deep neural network offers an effective strategy for AP complex distinction.
• This method can assist researchers in protein sequence analysis and understanding AP-related disorders.
• The developed dataset and model are valuable resources for the scientific community.