De novo profile generation based on sequence context specificity with the long short-term memory network.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Amino acid sequence profiles are vital for bioinformatics tasks like similarity searches and evolutionary analyses.
• The increasing volume of biological sequence data necessitates scalable profile generation methods.
• Deep learning, particularly Long Short-Term Memory (LSTM) networks, shows promise for analyzing sequential data.

Purpose of the Study:

• To develop a novel, assumption-free amino acid sequence profile generator using LSTM.
• To evaluate the performance of the proposed generator against existing methods.
• To investigate the capability of LSTM in capturing long-range interactions within biological sequences.

Main Methods:

• Employed a Long Short-Term Memory (LSTM) network architecture.
• Developed a new de novo profile generator named SPBuild.
• Benchmarked SPBuild against CSBuild and RPS-BLAST using profile-sequence similarity search performance.
• Analyzed the impact of LSTM's memory capacity on detecting amino acid interactions.

Main Results:

• The novel LSTM-based profile generator, SPBuild, demonstrated superior performance compared to existing methods.
• SPBuild achieved better profile-sequence similarity search results with linear computational costs.
• LSTM's memory effectively captured long-range amino acid interactions, beneficial for predicting structures like beta-strands.
• The method proved effective for biological sequence analysis, particularly for regions with complex interactions.

Conclusions:

• Sequence context is crucial for biological sequence analysis, and LSTM is a feasible deep learning approach.
• The SPBuild generator leverages LSTM's memory to achieve high performance in profile prediction, especially for beta-strands.
• Findings highlight the potential of machine learning, particularly LSTM, for advancing biological sequence prediction methods.