Neither experimental, nor computational methods are currently able to bridge the gap between known protein sequences and known protein structures. While experimental methods are too labor-intensive and time-consuming, computational methods seem promising, but have problems on their own.
Comparative modeling can usually provide good models for an unsolved structure because there is a limited number of folds, and most of them are probably already known. The best results are attained by utilizing sequence similarity to identify templates. If this is not possible, structural information can be leveraged by fold recognition methods, albeit at a lower success rate. If no template can be found, ab initio modeling can be used. It searches the approximated energy function of a protein for the global minimum, which is deemed to resemble the native state. This approach suffers most of all from the vastness of the search space that grows exponentially with each residue.
The goal was to develop an algorithm for retrieving templates that is independent of sequence similarity. In order to do this, we combine ab initio with comparative modeling by looking for similarities in ab initio decoys to identify good templates. We assume that decoys of a target will show similarities to decoys of structurally related proteins for one of two reasons. For one, they may simply be similar because decoys are similar to their respective native structures. These similarities may overlap in the decoys of two different related proteins and thus enable the distinction from decoys of unrelated proteins. Another possible cause is that the energy landscapes of proteins with related folds may have similar minima. This concept is supported by experimental data for a small region of the energy landscape. It has been shown that transition state ensembles of some structurally similar proteins also share similar structures. Despite the restriction of experimental data on this specific region, this concept might possess more generality.
Description of Work
The developed method works by generating ab initio decoys for all targets and a set of pre-filtered templates and comparing them by various metrics. In order to detect structural similarities between the decoys, we evaluate the similarity of predicted structural motif classes, predicted contacts, and structures between the sampled decoys. We used the retrieved structural similarities to create rankings of the potential templates, which we then compared to a state of the art template retrieval tool.
We tested the method on a set of 14 targets. Each target has 8 − 44 templates
that were retrieved by HHsearch. Looking at the top 5% templates ranked according
to the best variants of our method, the best template was found 41% of the time.
The top 5% of the ranking from HHsearch contained the best possible template only
21% of the time. We have discovered that the presented method mainly succeeds
because of similarities of the decoys to the native structures. Nevertheless, we could
also produce evidence for similar energy landscapes, although this notion did not
have a major influence on the results.