4. A Density Estimation Example
In the top panel of Figure 16 of our Triogram paper, we present three data sets that are natural candidates for Triogram density estimation. The points in these plots represent a collection of amino acids obtained form 100 protein structures taken from the Brookhaven Protein Data Bank (Hobohm, Scharf, Schneider and Sander 1992). In order to characterize the local environment of each amino acid within a given protein structure, three pieces of information were recorded: the local context of the protein at the given amino acid (whether the protein is twisting around a helix, for example), the fraction of the amino acid side-chain area that is buried in the protein structure, and the fraction of the side-chain area that is covered by polar atoms. Since the unburied portion of the amino acid is exposed to a polar solvent, the final two quantities are restricted to the upper triangle of the unit square. The plots in the top row of Figure 16 correspond to data collected from the amino acid Lysine found in a helix, a coil and a sheet.Bivariate density estimates computed for each amino acid and each local protein structure are the basis for an approach to solving the so-called inverse folding problem (Bowie, Luthy and Eisenberg 1991; Zhang and Eisenberg 1994). Evaluating the structure of a given protein is extremely difficult. Fortunately, determining the sequence of amino acids that comprise the protein is relatively simple. It would seem reasonable, therefore, to attempt to infer the protein's structure from its amino acid sequence. Unfortunately, many rather different sequences produce very similar structures, so the objective of the inverse folding problem is to determine which amino acid sequences might result in a given known structure. This can be accomplished by studying the propensity for certain amino acids to occur in certain local environments in a large collection of known protein structures. The procedure described by Zhang and Eisenberg involves a log-odds calculations, the main ingredient of which is a set of bivariate density estimates for the type of data given in the top row of Figure 16 of the Triogram paper.
We first applied the Triogram procedure separately to each dataset corresponding to the three different local environments. The fitted log-densities corresponding to Lysine found in a Helix structure is given in Figure 2 below. At each step in the addition process, the set of candidate vertices consisted of the points with barycentric coordinates associated with K=5 relative to each of the triangles in the current triangulation. We did not enforce shape restrictions on the updated triangulation when choosing between the candidates, but did insist that each triangle must contain at least 25 points. After the deletion phase we selected a final model using BIC. In each case, the best fits were encountered during the stepwise deletion.
Figure 2. The fitted log-densities associated with the Lysine data found in a helix context. As in the previous dynamic figure, you can rotate the perspective plot by clicking in the panel and dragging the mouse. You can scroll through the model selection process by positioning your mouse in the perspective plot window (not the BIC window) and hitting:
Note that the model corresponding to the minimum BIC value is the one selected by the Triogram procedure.
- the 'n' key or the spacebar for the next step,
- the 'p' or 'b' keys for the previous step,
- the '0' (zero) key to return to the initial model, or
- the 'd' key to toggle between our fits and the underlying data .
You are currently reading Section 4, Triogram density estimation.
- Introduction and acknowledgements
- A short description of Triogram modeling
- A regression example with simulated data
- A bivariate density estimation example
- A regression surface with a ridge
- References
- A PostScript version of Triogram Models
- Send us comments !
You can also bail out and head back to my home page .