A lightweight classification of adaptor proteins using transformer networks.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Adaptor proteins are crucial for intercellular signal transduction; their dysfunction leads to disease.
• Understanding protein structure is key to addressing associated conditions.
• Bioinformatics and computational biology research is vital for studying adaptor proteins.

Purpose of the Study:

• To introduce a novel, compact, and superior machine learning model for protein classification.
• To advance the field of adaptor protein recognition using innovative algorithms.

Main Methods:

• A novel transformer-based deep learning model incorporating convolutional blocks and a fully connected layer was developed.
• Protein sequences were inputted from a database, and Position-Specific Scoring Matrix (PSSM) features were extracted.
• The extracted features were processed through the proposed deep learning architecture.

Main Results:

• The model achieved state-of-the-art performance, demonstrated by high area under the receiver operating characteristic curve and Matthew's Correlation Coefficient.
• The model is highly efficient and compact, utilizing only 20 hidden nodes, approximately 1% of previous methods.
• Superior results and computational efficiency were achieved compared to existing approaches.

Conclusions:

• This is the first transformer model specifically designed for adaptor protein recognition.
• The model, utilizing PSSM profiles with convolutional blocks, transformer, and fully connected layers, outperforms all existing methods for adaptor protein classification.