3
$\begingroup$

In my lab we are trying to extract spatial features from protein structures. The software we develop makes use of CUDA for all heavy number-lifting, thus we are limited by the GPU's memory (12GB). Using standard voxel-based 3D-representations proved to be too memory hungry, hence we are trying to find a way to reduce the dimensionality to 2D, while preserving as much spatial information as possible. Our absolute goal is to preserve spatial colocalisation of different chains and secondary structures, to understand how important that is during folding. Standard techniques, such as PCA and PCoA on euclidian distances between amino acids, seem to preserve too little information on chain folds. We believe there should be a better way of doing that, but, having little experience in structural biology, we struggle to find any relevant work regarding this issue. Can you recommend any methods or relevant readings?

$\endgroup$
  • $\begingroup$ I'm not entirely sure I understand the question. What more spacial information do you need beyond .pdb files which hold the relative atomic co-ordinates? After this, dss can be ran to evaluate chain angles and overlay a secondary structure. $\endgroup$ – James Oct 27 '16 at 9:33
  • $\begingroup$ Maybe you could use a protein SSE graph? ptgl.uni-frankfurt.de/about.php $\endgroup$ – gilleain Nov 10 '16 at 11:13
  • $\begingroup$ @James the question is how to project 3D structures from .pdb into a lower-dimensional space without losing the information on relative position of chains. $\endgroup$ – Eli Korvigo Nov 10 '16 at 13:41
  • $\begingroup$ @gilleain this is actually quite interesting, though it lacks residue-level resolution. $\endgroup$ – Eli Korvigo Nov 10 '16 at 13:52
  • $\begingroup$ I suppose you already solved this, but did you consider something like non-linear dimensionality reduction with a regularization energy term that penalized the sort of chain fold deformations youre trying to avoid? E.g. using diffusion mappings with a kernel that combined say geodesic distances with an extra term for handling the chain folds. $\endgroup$ – user3658307 May 9 '17 at 22:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.