Soft Ngram Representation and Modeling for Protein Remote Homology Detection

Area of Science:

• Bioinformatics
• Computational Biology
• Protein Science

Background:

• Remote homology detection is crucial for understanding protein function, especially when sequence similarity is low.
• Existing methods often use sequence profiles but may not fully capture evolutionary information.

Purpose of the Study:

• To propose a novel profile-based sequence representation called soft Ngram.
• To enhance remote homology detection by more deeply exploiting evolutionary information in sequence profiles.

Main Methods:

• Developed the soft Ngram representation, extending traditional Ngrams by incorporating evolutionary frequencies from sequence profiles.
• Illustrated two modeling approaches to derive feature vectors for classification.
• Utilized support vector machines (SVM) for classification.

Main Results:

• The soft Ngram approach significantly outperforms traditional Ngram-based methods.
• Demonstrated very promising results compared to a wider range of existing techniques.
• Validated through thorough evaluation on three benchmark datasets.

Conclusions:

• Soft Ngrams offer a powerful new method for remote homology detection.
• This approach effectively models and utilizes crucial evolutionary information from protein sequence profiles.
• The method shows significant potential for advancing bioinformatics and protein function prediction.