Deep learning for protein secondary structure prediction: Pre and post-AlphaFold
Problem
Predicting protein secondary structure from amino acid sequence is a long-standing computational biology challenge.
Why it matters
Protein structure prediction underpins drug discovery and understanding of disease mechanisms.
What the AI did
Reviews how deep neural networks (CNNs, RNNs, graph and language models) became the primary method for secondary structure prediction, lifting three-state accuracy above 80%.
Results
Three-state secondary structure prediction accuracy above 80% (cited from previous studies)
Impact
The review documents deep learning as the dominant, effective approach in this field.
Record
| Venue | Computational and Structural Biotechnology Journal |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Dewi Pramudi Ismi, Reza Pulungan, Afiahayati |
| Citations | 50 (OpenAlex, at import time) |
| References | 203 |
| Model or method | not stated in source (reviews CNN, RNN, inception and graph networks) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1016/j.csbj.2022.11.012 |
| Open access | gold |
| Keywords | deep learning, protein secondary structure, AlphaFold, neural networks, review |
Importance rating
significant — thousands to millions affected
Protein secondary structure prediction is a long-standing core computational biology task; deep learning lifted accuracy above 80% per abstract, but no application scale stated.