Improving prediction of secondary structure, local backbone angles and solvent accessible surface area of proteins by iterative deep learning
Problem
Predicting protein structural features (secondary structure, backbone angles, solvent accessibility) directly from sequence is a challenging problem.
Why it matters
Protein structure prediction from sequence is a long-standing challenging problem in biology.
What the AI did
A deep learning neural network used three iterations of predicted secondary structure, backbone angles and solvent accessibility to progressively improve all predictions.
Results
82% secondary-structure accuracy; 0.76 ASA correlation; MAEs 19°/30° (φ/ψ) and 8°/32° (θ/τ) on independent test set.
Impact
Peer-reviewed accuracy gains across several protein prediction sub-tasks on independent data.
Record
| Venue | Scientific Reports |
|---|---|
| Published | 2015 |
| Work type | article |
| Authors | Rhys Heffernan, Kuldip K. Paliwal, James Lyons, Abdollah Dehzangi, Alok K. Sharma, Jihua Wang, Abdul Rahman Sattar, Yuedong Yang et al. |
| Citations | 351 (OpenAlex, at import time) |
| References | 49 |
| Model or method | Deep learning neural network (three iterations) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/srep11476 |
| Open access | gold |
| Keywords | protein secondary structure, deep learning, iterative prediction, solvent accessibility, backbone angles |
Importance rating
real but narrower advance
Protein structure prediction from sequence is a long-standing biology problem; this work improves several sub-tasks but is an incremental methodological advance with no stated scale.