Incorporating homologues into sequence embeddings for protein analysis

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Statistical and machine learning methods are vital in bioinformatics but often require data in Euclidean space.
• Discrete biological sequences, like protein sequences, cannot be directly analyzed by these methods.
• Existing methods for applying Euclidean-based techniques to sequences are limited.

Purpose of the Study:

• To introduce a novel, biologically motivated sequence embedding method called the homology kernel.
• To enable the direct application of powerful statistical and learning techniques to protein sequences.
• To demonstrate the utility of the homology kernel in various bioinformatics tasks.

Main Methods:

• Developed the homology kernel, an embedding that incorporates local alignment, sequence homology, and predicted secondary structure.
• Applied the homology kernel to protein family classification.
• Utilized the homology kernel for secondary structure prediction.
• Integrated the homology kernel into local sequence alignment to leverage information from homologous sequences.

Main Results:

• The homology kernel outperforms state-of-the-art methods in remote homology detection.
• The homology kernel achieves competitive performance in secondary structure prediction.
• The homology kernel effectively incorporates homologous sequence information into local sequence alignment.

Conclusions:

• The homology kernel is a versatile and powerful tool for analyzing protein sequences.
• This embedding method significantly advances the application of machine learning in bioinformatics.
• The homology kernel offers improved performance in protein classification, homology detection, and secondary structure prediction.