Search

Funder: Phenix Industrial Consortium, Experimental structure determination can be accelerated with artificial intelligence (AI)-based structure-prediction methods such as AlphaFold. Here, an automatic procedure requiring only sequence information and crystallographic data is presented that uses AlphaFold predictions to produce an electron-density map and a structural model. Iterating through cycles of structure prediction is a key element of this procedure: a predicted model rebuilt in one cycle is used as a template for prediction in the next cycle. This procedure was applied to X-ray data for 215 structures released by the Protein Data Bank in a recent six-month period. In 87% of cases our procedure yielded a model with at least 50% of Cα atoms matching those in the deposited models within 2 Å. Predictions from the iterative template-guided prediction procedure were more accurate than those obtained without templates. It is concluded that AlphaFold predictions obtained based on sequence information alone are usually accurate enough to solve the crystallographic phase problem with molecular replacement, and a general strategy for macromolecular structure determination that includes AI-based prediction both as a starting point and as a method of model optimization is suggested.

SummaryCryo-EM observation of biological samples enables visualization of sample heterogeneity, in the form of discrete states that are separatable, or continuous heterogeneity as a result of local protein motion before flash freezing. Variability analysis of this continuous heterogeneity describes the variance between a particle stack and a volume, and results in a map series describing the various steps undertaken by the sample in the particle stack. While this observation is absolutely stunning, it is very hard to pinpoint structural details to elements of the maps. In order to bridge the gap between observation and explanation, we designed a tool that refines an ensemble of structures into all the maps from variability analysis. Using this bundle of structures, it is easy to spot variable parts of the structure, as well as the parts that are not moving. Comparison with molecular dynamics simulations highlight the fact that the movements follow the same directions, albeit with different amplitudes. Ligand can also be investigated using this method. Variability refinement is available in thePhenixsoftware suite, accessible under the program namephenix.varref.

Quantum refinement (Q|R) of crystallographic or cryo-EM derived structures of biomolecules within the Q|R project aims at using ab initio computations instead of library-based chemical restraints. An atomic model refinement requires the calculation of the gradient of the objective function. While it is not a computational bottleneck in classic refinement it is a roadblock if the objective function requires ab initio calculations. A solution to this problem adopted in Q|R is to divide the molecular system into manageable parts and do computations for these parts rather than using the whole macromolecule. This work focuses on the validation and optimization of the automatic divide-and-conquer procedure developed within the Q|R project. Also, we propose an atomic gradient error score that can be easily examined with common molecular visualization programs. While the tool is designed to work within the Q|R setting the error score can be adapted to similar fragmentation methods. The gradient testing tool presented here allows a prioridetermination of the computationally efficient strategy given available resources for the potentially time-expensive refinement process. The procedure is illustrated using a peptide and small protein models considering different quantum mechanical (QM) methodologies from Hartree-Fock, including basis set and dispersion corrections, to the modern semi-empirical method from the GFN-xTB family. The results obtained provide some general recommendations for the reliable and effective quantum refinement of larger peptides and proteins.

AI-based methods such as AlphaFold have revolutionized structural biology, often making it possible to predict protein structures with high accuracy. The accuracies of these predictions vary, however, and they do not include ligands, covalent modifications or other environmental factors. Here we focus on very-high-confidence parts of AlphaFold predictions, evaluating how well they can be expected to describe the structure of a protein in a particular environment. We compare predictions with experimental crystallographic maps of the same proteins for 102 crystal structures. In many cases, those parts of AlphaFold predictions that were predicted with very high confidence matched experimental maps remarkably closely. In other cases, these predictions differed from experimental maps on a global scale through distortion and domain orientation, and on a local scale in backbone and side-chain conformation. Overall, Cαatoms in very-high-confidence parts of AlphaFold predictions differed from corresponding crystal structures by a median of 0.6 Å, and about 10% of these differed by more than 2 Å, each about twice the values found for pairs of crystal structures containing the same components but determined in different space groups. We suggest considering AlphaFold predictions as exceptionally useful hypotheses. We further suggest that it is important to consider the confidence in prediction when interpreting AlphaFold predictions and to carry out experimental structure determination to verify structural details, particularly those that involve interactions not included in the prediction.

Bulk solvent is a major component of bio-macromolecular crystals and therefore contributes significantly to diffraction intensities. Accurate modeling of the bulk-solvent region has been recognized as important for many crystallographic calculations, from computing of R-factors and density maps to model building and refinement. Owing to its simplicity and computational and modeling power, the flat (mask-based) bulk-solvent model introduced by Jiang & Brunger (1994) is used by most modern crystallographic software packages to account for disordered solvent. In this manuscript we describe further developments of the mask-based model that improves the fit between the model and the data and aids in map interpretation. The new algorithm, here referred to as mosaic bulk-solvent model, considers solvent variation across the unit cell. The mosaic model is implemented in the computational crystallography toolbox and can be used in Phenix in most contexts where accounting for bulk-solvent is required. It has been optimized and validated using a sufficiently large subset of the Protein Data Bank entries that have crystallographic data available.SynopsisA mosaic bulk-solvent method models disordered solvent more accurately than current flat bulk solvent model. This improves the fit between the model and the data, improves map quality and allows for the solution of problems previously inaccessible.