DeepPredict: a state-of-the-art web server for protein secondary structure and relative solvent accessibility...

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Accurate prediction of protein secondary structure (PSSP) and relative solvent accessibility (RSA) is crucial for understanding protein function and structure.
• Existing methods often rely on multiple sequence alignments (MSAs), which can be time-consuming and computationally intensive.
• The development of deep learning models has shown promise in improving prediction accuracy and efficiency.

Purpose of the Study:

• To introduce DeepPredict, a novel web server for PSSP and RSA prediction.
• To integrate two advanced deep learning models, Porter6 and PaleAle6, into a user-friendly platform.
• To leverage pre-trained protein language models (PLMs) to bypass the need for MSAs.

Main Methods:

• DeepPredict utilizes a deep learning framework incorporating pre-trained protein language models (PLMs), specifically ESM-2.
• It integrates Porter6 for PSSP prediction and PaleAle6 for RSA prediction.
• The server provides a web interface for easy access and prediction generation.

Main Results:

• DeepPredict demonstrates state-of-the-art performance in both PSSP and RSA prediction tasks.
• The use of ESM-2 eliminates the requirement for MSAs, leading to rapid predictions.
• The server outperforms existing prediction methods in accuracy and speed.

Conclusions:

• DeepPredict provides a powerful and accessible tool for predicting protein secondary structure and solvent accessibility.
• Its deep learning approach and reliance on PLMs offer significant advantages over traditional methods.
• The user-friendly interface makes it valuable for both computational and experimental researchers in structural biology.