CNN extract the useful input festures from the raw data on its own. Then why we need to choose input features by ourself like solvent accessibility, secodary structures, evolutionary information etc. Why don't we simple give it a raw data of protein sequences to learn from and give predictions.
Mukhtar Ahmad Sofi The free energy landscape of protein is very complex. Just knowing the sequence doesn't allow to predict the proper conformation of protein. The folding of the protein takes in the range of microseconds to minute. In order to predict the correct conformation normal MD simulation time is not enough. In this respect, the enhanced sampling methods are used such as replica exchange MD (REMD) or metadynamics. The temperature increases from cold to hot in REMD. This assist to escape from the metastable state and find the stable conformation of the protein.
That is no longer true, provided the right network architecture, it can learn from sequence data alone (without manually extracting features). Article End-to-End Differentiable Learning of Protein Structure
A sequence of letters alone means nothing both to a natural and an artificial illiterate.Ii acquires meaning only after learning a lot about the object it encodes and implements this knowledge in a persons brain or its artificial companions in
silico including the methods to be used and the extra factors (that in the case of a protein will determine what will happen with it in a prescribed environment(temperature,solvent,ions,pH,ligands,other proteins,etc,etc).I do not think that anybody ( human or artificial) can predict the correct 3D structure(s) of
a protein given the sequence.we must be more modest and patient and be happy with the things we can presently describe ,thanks to decades of work of myriad
dedicated scientists and the computing machines thereby created.
- Badges
- Science topic