SUPERMAGO: Protein Function Prediction Based on Transformer Embeddings.

Area of Science:

• Computational Biology
• Bioinformatics
• Protein Science

Background:

• Experimental determination of protein amino acid sequences is advancing rapidly.
• Analyzing protein function from sequences is crucial but challenging due to cost and time.
• Computational methods using amino acid sequences are vital for protein function categorization.

Purpose of the Study:

• To introduce SUPERMAGO, a novel method for predicting protein functions from amino acid sequences.
• To present SUPERMAGO+, an enhanced ensemble method, and SUPERMAGO+Web, a web server version.
• To evaluate the performance of these methods against state-of-the-art approaches.

Main Methods:

• Utilizing Transformer architectures pre-trained on protein data for feature extraction.
• Employing multilayer perceptrons for classification and stacking neural networks for prediction aggregation.
• Developing SUPERMAGO+ as a neural network-based ensemble of SUPERMAGO and DIAMOND with a novel weighting mechanism.

Main Results:

• SUPERMAGO and SUPERMAGO+ demonstrated superior performance compared to existing methods.
• The stacking neural network significantly enhanced prediction accuracy.
• SUPERMAGO+ introduced a novel weighting mechanism for improved functional prediction.

Conclusions:

• SUPERMAGO and SUPERMAGO+ are highly effective computational tools for predicting protein functions solely from amino acid sequences.
• These methods represent a significant advancement in the field, offering improved accuracy and efficiency.
• SUPERMAGO+Web provides a resource-efficient solution for broader accessibility.