OrgNet: orientation-gnostic protein stability assessment using convolutional neural networks.

Area of Science:

• Computational biology
• Biotechnology
• Structural biology

Background:

• Accurate prediction of single-point mutation effects on protein stability is vital for understanding disease mechanisms and advancing protein engineering.
• Deep learning (DL) models show promise for predicting protein thermostability, potentially outperforming traditional methods.
• Existing structure-based DL models, like convolutional neural networks (CNNs), suffer from orientation bias, impacting prediction consistency.

Purpose of the Study:

• To introduce OrgNet, a novel orientation-agnostic DL model for predicting protein thermostability changes upon point mutation.
• To address and eliminate the orientation bias inherent in structure-based DL models for protein stability prediction.

Main Methods:

• OrgNet utilizes 3D CNNs to encode protein structures as voxel grids, capturing detailed atomic features.
• The model incorporates spatial transforms to standardize protein orientations, mitigating orientation bias.
• OrgNet is evaluated on established benchmarks like Ssym and S669.

Main Results:

• OrgNet achieves state-of-the-art performance in predicting protein thermostability changes.
• The model demonstrates superior accuracy and robust performance compared to existing prediction methods.
• Elimination of orientation bias leads to more consistent and reliable predictions.

Conclusions:

• OrgNet offers a significant advancement in predicting the impact of mutations on protein stability.
• The orientation-agnostic approach overcomes a key limitation of previous structure-based DL models.
• OrgNet holds potential for applications in disease research and protein engineering.