Pathogenicity classification of missense mutations based on deep generative model.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Missense mutations in human proteins are linked to various diseases.
• Accurate identification and pathogenicity classification of these mutations are crucial for understanding disease mechanisms and protein function.

Purpose of the Study:

• To propose MLAE (Method based on LSTM-Ladder AutoEncoder), a novel deep learning model for identifying disease-associated missense mutations and classifying their pathogenicity.
• To leverage the Variational AutoEncoder (VAE) framework with LSTM networks and a Ladder structure to improve information retention and model learning.

Main Methods:

• Developed MLAE, a deep learning model integrating LSTM networks and a Ladder structure within the VAE framework.
• Applied MLAE to classify all 27,572 possible missense variants for three input proteins.

Main Results:

• Achieved an average classification AUC of 0.941, demonstrating MLAE's effectiveness in predicting pathogenicity.
• Obtained an average Hamming loss of 0.196 for multi-label classification, indicating proficiency in classifying complex variants.

Conclusions:

• MLAE effectively captures amino acid sequence information for accurate pathogenicity prediction of missense mutations.
• The model offers a valuable analytical tool for studying genetic disease bases and disease prevention strategies.