Variable Length Character N-Gram Embedding of Protein Sequences for Secondary Structure Prediction

Area of Science:

• Computational biology
• Structural bioinformatics
• Machine learning in proteomics

Background:

• Protein secondary structure prediction from amino acid sequences is vital for 3-D structure determination.
• While homologous sequence alignments improve predictions, they are not always available.
• Accurate prediction from single sequences is therefore necessary.

Purpose of the Study:

• To predict protein secondary structure from primary amino acid sequences.
• To leverage n-gram word embedding and deep recurrent neural networks for enhanced prediction.
• To capture both local and long-range residue dependencies.

Main Methods:

• Utilized variable-length character n-gram words to capture local residue context.
• Employed embedding vectors to represent n-gram words.
• Applied a bidirectional long short-term memory (Bi-LSTM) model to extract long-range sequence information.

Main Results:

• Evaluated the model on ss.txt, RS126, and CASP9 datasets.
• Achieved Q3 accuracies of 92.57% (ss.txt), 86.48% (RS126), and 89.66% (CASP9).

Conclusions:

• The proposed model demonstrates superior performance compared to existing state-of-the-art methods.
• The integration of n-gram embeddings and Bi-LSTM effectively predicts protein secondary structure.