I have many pdb files (amino sequences are alike).
And I would like to use these data for machine learning.
In my problem, amino sequences have same length, but the number of atoms is not different.
I would like to deal with these data not losing structural information and organize them for ML.
I found the article below.
Conference Paper Encoding protein structure with functions on graphs
If you know other methods for this problem, please teach me.
If possible, let me know programs (Python and R are best for me).
Hello Ryosaku,
Unfortunately, if you need constant-sized vectors it's going to be tricky to do all-atom things unless you're working on an identical polypeptide and looking at some perturbations specifically (ligand binding, temperature, etc.). Most packages I have encountered have converged on using the C-alphas of a polypeptide for representing the entire structure, or some other course-grained bead method.
If this is a possible representation for what you're trying to do, I recommend you look into the Bio3D R package. I use it for a lot of MD trajectory analysis and structural analyses, as well as some graph theory type functionalities that are included in the package.
If you provide more information on what you're trying to classify (and maybe what classifier if you know already) specifically with all of this protein information, that may help finding a more specific answer.
Good luck!
Here are some other sources in RG that can be helpful for you:
https://www.researchgate.net/search.Search.html?type=publication&query=How%20to%20change%20protein%20structure%20to%20constant-size%20vector?
Best regards,
Debra
You can use the double-centroid reduced representation (DCRR) of proteins that I developed some years ago. It worked well for me in predicting protein-ligand binding sites (in the protein) and protein-protein interaction sites in binary complexes (on both proteins). It is published already.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs