Bayesian machine learning improves single-wavelength anomalous diffraction phasing

Machine learning in crystallography and structural science

An overview of the virtual collection on machine learning (ML) in crystallography and structural science, as represented in Acta Crystallographica Sections A, B and D, IUCrJ and Journal of Synchrotron Radiation, is presented. Some terms and concepts related to artificial intelligence and machine learning are briefly introduced and described, and a short history of ML in structural science as it appeared in these IUCr journals is given to whet the appetite for the rest of the collection.

Correcting systematic errors in diffraction data with modern scaling algorithms

X-ray diffraction enables the routine determination of the atomic structure of materials. Key to its success are data-processing algorithms that allow experimenters to determine the electron density of a sample from its diffraction pattern. Scaling, the estimation and correction of systematic errors in diffraction intensities, is an essential step in this process. These errors arise from sample heterogeneity, radiation damage, instrument limitations and other aspects of the experiment. New X-ray sources and sample-delivery methods, along with new experiments focused on changes in structure as a function of perturbations, have led to new demands on scaling algorithms. Classically, scaling algorithms use least-squares optimization to fit a model of common error sources to the observed diffraction intensities to force these intensities onto the same empirical scale. Recently, an alternative approach has been demonstrated which uses a Bayesian optimization method, variational inference, to simultaneously infer merged data along with corrections, or scale factors, for the systematic errors. Owing to its flexibility, this approach proves to be advantageous in certain scenarios. This perspective briefly reviews the history of scaling algorithms and contrasts them with variational inference. Finally, appropriate use cases are identified for the first such algorithm, Careless, guidance is offered on its use and some speculations are made about future variational scaling methods.

M, Katona G. Survivin prevents the polycomb repressor complex 2 from methylating histone 3 lysine 27

This study investigates the role of survivin in epigenetic control of gene transcription through interaction with the polycomb repressive complex 2 (PRC2). PRC2 is responsible for silencing gene expression by trimethylating lysine 27 on histone 3. We observed differential expression of PRC2 subunits in CD4+ T cells with varying levels of survivin expression, and ChIP-seq results indicated that survivin colocalizes with PRC2 along DNA. Inhibition of survivin resulted in a significant increase in H3K27 trimethylation, implying that survivin prevents PRC2 from functioning. Peptide microarray showed that survivin interacts with peptides from PRC2 subunits, and machine learning revealed that amino acid composition contains relevant information for predicting survivin interaction. NMR and BLI experiments supported the interaction of survivin with PRC2 subunit EZH2. Finally, protein-protein docking revealed that the survivin-EZH2 interaction interface overlaps with catalytic residues of EZH2, potentially inhibiting its H3K27 methylation activity. These findings suggest that survivin inhibits PRC2 function.

Emerging Time-Resolved X-Ray Diffraction Approaches for Protein Dynamics

Proteins guide the flows of information, energy, and matter that make life possible by accelerating transport and chemical reactions, by allosterically modulating these reactions, and by forming dynamic supramolecular assemblies. In these roles, conformational change underlies functional transitions. Time-resolved X-ray diffraction methods characterize these transitions either by directly triggering sequences of functionally important motions or, more broadly, by capturing the motions of which proteins are capable. To date, most successful have been experiments in which conformational change is triggered in light-dependent proteins. In this review, I emphasize emerging techniques that probe the dynamic basis of function in proteins lacking natively light-dependent transitions and speculate about extensions and further possibilities. In addition, I review how the weaker and more distributed signals in these data push the limits of the capabilities of analytical methods. Taken together, these new methods are beginning to establish a powerful paradigm for the study of the physics of protein function.

reciprocalspaceship: a Python library for crystallographic data analysis

Crystallography uses the diffraction of X-rays, electrons or neutrons by crystals to provide invaluable data on the atomic structure of matter, from single atoms to ribosomes. Much of crystallography's success is due to the software packages developed to enable automated processing of diffraction data. However, the analysis of unconventional diffraction experiments can still pose significant challenges - many existing programs are closed source, sparsely documented, or challenging to integrate with modern libraries for scientific computing and machine learning. Described here is reciprocalspaceship, a Python library for exploring reciprocal space. It provides a tabular representation for reflection data from diffraction experiments that extends the widely used pandas library with built-in methods for handling space groups, unit cells and symmetry-based operations. As is illustrated, this library facilitates new modes of exploratory data analysis while supporting the prototyping, development and release of new methods.

Estimating the probability of coincidental similarity between atomic displacement parameters with machine learning

High-resolution diffraction studies of macromolecules incorporate the tensor form of the anisotropic displacement parameter (ADP) of atoms from their mean position. The comparison of these parameters requires a statistical framework that can handle the experimental and modeling errors linked to structure determination. Here, a Bayesian machine learning model is introduced that approximates ADPs with the random Wishart distribution. This model allows for the comparison of random samples from a distribution that is trained on experimental structures. The comparison revealed that the experimental similarity between atoms is larger than predicted by the random model for a substantial fraction of the comparisons. Different metrics between ADPs were evaluated and categorized based on how useful they are at detecting non-accidental similarity and whether they can be replaced by other metrics. The most complementary comparisons were provided by Euclidean, Riemann and Wasserstein metrics. The analysis of ADP similarity and the positional distance of atoms in bovine trypsin revealed a set of atoms with striking ADP similarity over a long physical distance, and generally the physical distance between atoms and their ADP similarity do not correlate strongly. A substantial fraction of long- and short-range ADP similarities does not form by coincidence and are reproducibly observed in different crystal structures of the same protein.

Likelihood-based estimation of substructure content from single-wavelength anomalous diffraction (SAD) intensity data

SAD phasing can be challenging when the signal-to-noise ratio is low. In such cases, having an accurate estimate of the substructure content can determine whether or not the substructure of anomalous scatterer positions can successfully be determined. Here, a likelihood-based target function is proposed to accurately estimate the strength of the anomalous scattering contribution directly from the measured intensities, determining a complex correlation parameter relating the Bijvoet mates as a function of resolution. This gives a novel measure of the intrinsic anomalous signal. The SAD likelihood target function also accounts for correlated errors in the measurement of intensities from Bijvoet mates, which can arise from the effects of radiation damage. When the anomalous signal is assumed to come primarily from a substructure comprising one anomalous scatterer with a known value of f'' and when the protein composition of the crystal is estimated correctly, the refined complex correlation parameters can be interpreted in terms of the atomic content of the primary anomalous scatterer before the substructure is known. The maximum-likelihood estimation of substructure content was tested on a curated database of 357 SAD cases with useful anomalous signal. The prior estimates of substructure content are highly correlated to the content determined by phasing calculations, with a correlation coefficient (on a log-log basis) of 0.72.

An Integrative Analysis Reveals the Potential Mechanism between Herbal Medicine Yinchen and Immunoregulation in Hepatocellular Carcinoma

Aims. Abundant evidences in traditional Chinese medicine (TCM) supported the therapeutic value of herbal medicine Yinchen in hepatocellular carcinoma (HCC), but the underlying mechanism remains to be investigated. Main Methods. The intersection of immune gene set, module genes, HCC-associated genes, and target genes of Yinchen was employed for further analyses. The module genes were identified by weighted gene coexpression network analysis, and the other three gene sets were obtained from public databases. Subsequently, we further explored the clinical value and immunoregulation of the hub gene of intersection. The relevant pathways related to hub gene expression were investigated by gene set enrichment analysis. Finally, the interaction of active compounds and target genes was validated by molecular docking. Key Findings. Thirteen active compounds and 90 target genes of Yinchen were included. After constructing the network among Yinchen, target genes, and HCC, BIRC5 was identified as the hub gene. Significant difference was found between the high-expressed group and the low-expressed group in survival and stage. Different immune subtypes also presented significant difference in BIRC5 expression. Moreover, NK cell and T cell (CD4+ effector memory and CD4+ memory resting) were negatively correlated with BIRC5 expression, while CTLA4 and LAG3 were positively correlated. The results of molecular docking further validated a good binding activity of quercetin-BIRC5 interaction. Significance. In summary, our research identified for the first time a novel underlying association among herbal medicine Yinchen, BIRC5, immunotherapy, and HCC. We speculated that Yinchen may target the immune checkpoints (CTLA4 and LAG3) and activate the immune cells by suppressing BIRC5.

Advances in long-wavelength native phasing at X-ray free-electron lasers

Long-wavelength pulses from the Swiss X-ray free-electron laser (XFEL) have been used for de novo protein structure determination by native single-wavelength anomalous diffraction (native-SAD) phasing of serial femtosecond crystallography (SFX) data. In this work, sensitive anomalous data-quality indicators and model proteins were used to quantify improvements in native-SAD at XFELs such as utilization of longer wavelengths, careful experimental geometry optimization, and better post-refinement and partiality correction. Compared with studies using shorter wavelengths at other XFELs and older software versions, up to one order of magnitude reduction in the required number of indexed images for native-SAD was achieved, hence lowering sample consumption and beam-time requirements significantly. Improved data quality and higher anomalous signal facilitate so-far underutilized de novo structure determination of challenging proteins at XFELs. Improvements presented in this work can be used in other types of SFX experiments that require accurate measurements of weak signals, for example time-resolved studies.