DeepPI: Alignment-Free Analysis of Flexible Length Proteins Based on Deep Learning and Image Generator

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• High-throughput sequencing generates vast protein data, necessitating efficient computational analysis.
• Experimental methods for protein function determination are costly and time-consuming.
• Existing deep learning methods for protein analysis are limited by fixed-length sequences and adjacent amino acid information.

Purpose of the Study:

• To develop a deep learning-based tool, DeepPI, for analyzing large-scale protein databases.
• To overcome the limitations of fixed-length sequence analysis in current computational methods.
• To improve the accuracy of protein function inference through flexible-length sequence analysis.

Main Methods:

• DeepPI utilizes Global Average Pooling for flexible-length protein sequence analysis, minimizing information loss.
• An image generator converts 1D sequences into 2D structures to extract common features across diverse shapes.
• Automated filtering techniques ensure representative data selection and coverage of large protein databases.

Main Results:

• DeepPI was successfully applied to large databases like Pfam-A.
• Comparative experiments demonstrated the impact of different image generators on feature extraction.
• Filtering techniques proved effective and scalable for large datasets.
• DeepPI achieved superior protein family classification accuracy compared to existing methods.

Conclusions:

• DeepPI offers an effective deep learning solution for analyzing flexible-length protein sequences.
• The tool enhances protein function inference accuracy, particularly in large-scale database applications.
• DeepPI addresses key limitations of previous computational approaches in bioinformatics.