Accurate Protein Structure Prediction Research Articles

Improving protein structure prediction with model-based search

De novo protein structure prediction can be formulated as search in a high-dimensional space. One of the most frequently used computational tools to solve such search problems is the Monte Carlo method. We present a novel search technique, called model-based search. This method samples the high-dimensional search space to build an approximate model of the underlying function. This model is incrementally refined in areas of interest, whereas areas that are not of interest are excluded from further exploration. Model-based search derives its efficiency from the fact that the information obtained during the exploration of the search space is used to guide further exploration. In contrast, Monte Carlo-based techniques lack memory and exploration is performed based on random walks, ignoring the information obtained in previous steps. Model-based search is applied to protein structure prediction, where search is employed to find the global minimum of the protein's energy landscape. We show that model-based search uses computational resources more efficiently to find lower-energy conformations of proteins than one of the leading protein structure prediction methods, which relies on a tailored Monte Carlo method to perform a search. The performance improvements become more pronounced as the dimensionality of the search problem increases. We argue that model-based search will enable more accurate protein structure prediction than was previously possible. Furthermore, we believe that similar performance improvements can be expected in other problems that are currently solved using Monte Carlo-based search methods. An implementation of model-based search can be obtained by contacting the authors.

Enhanced sampling near the native conformation using statistical potentials for local side‐chain and backbone interactions

In the preceding article in this issue of Proteins, an empirical energy function consisting of 4 statistical potentials that quantify local side-chain-backbone and side-chain-side-chain interactions has been demonstrated to successfully identify the native conformations of short sequence fragments and the native structure within large sets of high-quality decoys. Because this energy function consists entirely of interactions between residues separated by fewer than 5 positions, it can be used at the earliest stage of ab initio structure prediction to enhance the efficiency of conformational search. In this article, protein fragments are generated de novo by recombining very short segments of protein structures (2, 4, or 6 residues), either selected at random or optimized with respect this local energy function. When local energy is optimized in selected fragments, more efficient sampling of conformational space near the native conformation is consistently observed for 450 randomly selected single turn fragments, with turn lengths varying from 3 to 12 residues and all 4 combinations of flanking secondary structure. These results further demonstrate the energetic significance of local interactions in protein conformations. When used in combination with longer range energy functions, application of these potentials should lead to more accurate prediction of protein structure.

FRankenstein becomes a cyborg: the automatic recombination and realignment of fold recognition models in CASP6.

In the course of CASP6, we generated models for all targets using a new version of the "FRankenstein's monster approach." Previously (in CASP5) we were able to build many very accurate full-atom models by selection and recombination of well-folded fragments obtained from crude fold recognition (FR) results, followed by optimization of the sequence-structure fit and assessment of alternative alignments on the structural level. This procedure was however very arduous, as most of the steps required extensive visual and manual input from the human modeler. Now, we have automated the most tedious steps, such as superposition of alternative models, extraction of best-scoring fragments, and construction of a hybrid "monster" structure, as well as generation of alternative alignments in the regions that remain poorly scored in the refined hybrid model. We have also included the ROSETTA method to construct those parts of the target for which no reasonable structures were generated by FR methods (such as long insertions and terminal extensions). The analysis of successes and failures of the current version of the FRankenstein approach in modeling of CASP6 targets reveals that the considerably streamlined and automated method performs almost as well as the initial, mostly manual version, which suggests that it may be a useful tool for accurate protein structure prediction even in the hands of nonexperts.

IMIPRAMINE AND "DRINAMYL" IN DEPRESSIVE ILLNESS: A COMPARATIVE TRIAL.

<h3>Abstract</h3> Accurate prediction of protein structures is critical for understanding the biological function of proteins. Nevertheless, most structure optimization methods are built upon pre-defined statistical energy functions, which may be sub-optimal in formulating the conformation space. In this paper, we propose an end-to-end approach for protein structure optimization, powered by SE(3)-equivariant energy-based models. The conformation space is characterized by a SE(3)-equivariant graph neural network, with substantial modifications to embed the protein-specific domain knowledge. Furthermore, we introduce continuously-annealed Langevin dynamics as a novel sampling algorithm, and demonstrate that such process converges to native protein structures with theoretical guarantees. Extensive experiments indicate that SE(3)-Fold achieves comparable structure optimization accuracy, compared against state-of-the-art baselines, with over 1-2 orders of magnitude speed-up.