SAINT
SAINT predicts protein secondary structure at 8-class (Q8) resolution using a self-attention–augmented Inception-Inside-Inception (Deep3I) neural network to model short- and long-range amino acid residue interactions.
Key Features:
- Self-Attention Augmentation: Incorporates a self-attention mechanism (originating from natural language processing) to focus on relevant sequence positions and capture long-range dependencies.
- Deep Inception-Inside-Inception (Deep3I) Architecture: Uses the Deep3I network to extract multi-scale local patterns and short-range interactions.
- Short- and Long-Range Interaction Modeling: Combines self-attention with Deep3I to model both short-range and long-range residue interactions important for secondary structure.
- Q8 Resolution: Produces eight-class secondary structure (Q8) predictions for detailed structural annotation.
- Interpretability: Provides a more interpretable framework relative to traditional deep neural network methods.
- Benchmark Performance: Demonstrated best-known Q8 accuracy on benchmark datasets TEST2016, TEST2018, CASP12, and CASP13.
Scientific Applications:
- Protein Secondary Structure Annotation: Generates Q8-level annotations to support structural characterization of proteins.
- Inference of Protein Interactions and Function: Q8 predictions provide insights that aid interpretation of protein interactions and functional sites.
- Complement to Experimental Structural Biology: Serves as a computational alternative or complement to experimental methods such as X-ray crystallography and nuclear magnetic resonance (NMR) spectroscopy when those are costly or time-intensive.
- Benchmarking and Method Comparison: Used for performance comparison on standard datasets including TEST2016, TEST2018, CASP12, and CASP13.
Methodology:
SAINT integrates a self-attention mechanism with the Deep Inception-Inside-Inception (Deep3I) network to model short- and long-range residue interactions for Q8 secondary-structure prediction and was evaluated on TEST2016, TEST2018, CASP12, and CASP13.
Topics
Details
- Programming Languages:
- Python
- Added:
- 1/9/2020
- Last Updated:
- 1/16/2021
Operations
Publications
Uddin MR, Mahbub S, Rahman MS, Bayzid MS. SAINT: Self-Attention Augmented Inception-Inside-Inception Network Improves Protein Secondary Structure Prediction. Unknown Journal. 2019. doi:10.1101/786921.
DOI: 10.1101/786921