This page provides access to the neighbor-dependent Ramachandran distributions described in Ting et al., PLOS Comp. Biol. (April, 2010).
This work was funded by NIH grant P20 GM76222 from the National Institutes of Health under the National
Institute for General Medical Sciences, as part of the Protein Structure Initiative
(PSI2).
License for the neighbor-dependent Ramachandran distributions for non-profit users: Click here
Description:
In
this work, Ramachandran probability distributions are presented for
residues in protein loops from a high-resolution data set with
filtering based on calculated electron densities. Distributions for
all 20 amino acids (with cis and trans proline treated separately)
have been determined, as well as 420 left-neighbor and 420
right-neighbor dependent distributions. The neighbor-independent and
neighbor-dependent probability densities have been accurately
estimated using Bayesian nonparametric statistical analysis based on
the Dirichlet process. In particular, we used hierarchical Dirichlet
process priors, which allow sharing of information between densities
for a particular residue type and different neighbor residue
types. The resulting distributions have been tested in a loop modeling
benchmark, and are shown to improve protein
loop conformation prediction significantly.
In the figure above, XXX.yyy indicates the Ramachandran distribution of residue type XXX with right neighbor yyy. In this file (35 MByte pdf, 43 pages), we show the Ramachandran distributions for the TCBIG set (Turns+Coil+Bridge+PiHelix+310Helix). yyy.XXX indicates the Ramachandran distribution of residue XXX with left neighbor yyy. The first page shows the neighbor-independent distributions.
The paper has been published in PLOS Computational Biology. A reprint is available.
Please cite the paper: Daniel Ting, Guoli Wang,
Maxim Shapovalov, Rajib Mitra, Michael I. Jordan, Roland L. Dunbrack,
Jr. Neighbor-dependent Ramachandran probability distributions of amino acids developed from a hierarchical Dirichlet process model. PLOS Comp. Biol. (April 2010).
Availability:
The NDRD is free to researchers in non-profit institutions. Obtaining the NDRD is fast and easy.
The non-profit/academic license form is available here. Just click and then fill out the form and click "I agree". You will get a page with your submitted data for you to check. Then make sure you hit "Send request" to complete the license request. Note: if you submit a blank request or nonsense information, you will not get a response from us.
Individuals in for-profit institutions should contact Roland Dunbrack to obtain information on a commercial license for the NDRD.
Usage:
Each file contains probabilities of phi,psi for specific residue
types given the residue type of a neighbor to the left or to the
right. "ALL" means all neighbor residues of the residue in question
were kept in the calculation. The format is this:
So for instance, here are some lines for LEU-right-PRO Ramachandran distribution
Res Dir. Neigh phi psi Probability log(prob) Cumulative_sum
LEU right PRO -175 -130 5.237406e-08 16.76485 2.082167e-04
LEU right PRO -175 -125 4.726758e-08 16.86744 2.082640e-04
LEU right PRO -175 -120 4.449988e-08 16.92778 2.083085e-04
LEU right PRO -175 -115 4.332896e-08 16.95444 2.083518e-04
LEU right PRO -175 -110 4.208621e-08 16.98355 2.083939e-04
LEU right PRO -175 -105 3.898562e-08 17.06007 2.084329e-04
LEU right PRO -175 -100 3.459067e-08 17.17968 2.084675e-04
Res = the residue type for the Ramachandran Distribution
Dir = the direction of the neighbor ("ALL" is ALL residue types at once)
phi and psi = the floors of 5x5 regions
Probability = the probability in the 5x5 regions with floor at that phi,psi (e.g. the point at -175,-130 covers the range (-175,-130) to (-170,-125)
log(prob) = followed by the log probaility.
Cumulative_sum = the cumulative sum and can be used for drawing random values from the probability distributions. The sum is 1.0 for each neighbor map.
CPR is cis proline as a central amino acid
Note: the probabilities cover the regions above and to the right of the phi,psi point; e.g., at phi,psi = {60,0}, the probability is for the region {60,0} -> {65,5}.
To calculate probabilies for triplets, use:
To calculate probabilies for triplets (center,left,right), use:
log p*(phi,psi | C,L,R) = log p(phi,psi |C,L) + log p(phi,psi |C,R) - log p(phi,psi |C,R=ALL)
Once log p*(phi,psi | C,L,R) is calculated, calculate p*(phi,psi |C,L,R) = exp(log(p*(phi,psi | C,L,R)))
Then sum them up for each Ramachandran map, and normalize the probabilities by dividing by the sum:
p(phi,psi, | C,L,R) = p*(phi,psi | C,L,R) / sum
There are four distribution files:
NDRD_TCBIG.txt = data from Turn, Coil, Bridge, PiHelix, and 310 Helix
NDRD_TCB.txt = data from Turn, Coil and Bridge
NDRD_Conly.txt = Coil only
NDRD_Tonly.txt = Turn only
Contact us
Roland Dunbrack
