LightRoseTTA: High-Efficient and Accurate Protein Structure Prediction Using a Light-Weight Deep Graph Model.

Area of Science:

• Computational biology
• Structural bioinformatics
• Machine learning in biology

Background:

• Accurate protein structure prediction is crucial for biological research.
• Existing deep learning models like RoseTTAFold show promise but are resource-intensive.

Purpose of the Study:

• To develop a highly accurate and efficient protein structure prediction model.
• To reduce the computational cost and data requirements for protein structure prediction.

Main Methods:

• Implementation of LightRoseTTA, a light-weight deep graph network.
• Training and evaluation on benchmark datasets (CASP14, CAMEO) and MSA-insufficient datasets (Orphan, De novo, Orphan25).
• Comparison of training time, parameter count, and performance against RoseTTAFold.

Main Results:

• LightRoseTTA achieves accuracy competitive with RoseTTAFold.
• Significantly reduced training time (1 week vs. 30 days) and parameter count (1.4M vs. 130M).
• Superior performance on MSA-insufficient datasets and demonstrated transferability to antibody data.

Conclusions:

• LightRoseTTA offers a feasible and efficient alternative for protein structure prediction, especially in resource-limited settings.
• The light-weight approach democratizes advanced protein structure prediction capabilities.
• Open-source release aims to accelerate biological research.