It is my understanding that machine learning approaches perform best for predicting secondary structures in proteins ( having prediction accuracy of up to 80% ). However, protein structure prediction with ML relies on finding homologous regions established from previously determined structures. So, it won't work for proteins for which no known homologues exist. My background is computer science and, not being from the field of biochemistry, I wonder whether non-ML methods like improved Chou-Fasman and GOR are still being worked on.
Hi Renata,
This paper details about it.
https://www.nature.com/articles/s41598-018-28084-8
A note: The paper mentioned above doesn't distinguish between homologous and non-homologous proteins. The reported average prediction accuracy is around 85%. So, we don't know how does the method perform for sequences without known homologues.
- Badges
- Science topic