BITES: balanced individual treatment effect for survival data.

MOTIVATION: Estimating the effects of interventions on patient outcome is one of the key aspects of personalized medicine. Their inference is often challenged by the fact that the training data comprises only the outcome for the administered treatment, and not for alternative treatments (the so-called counterfactual outcomes). Several methods were suggested for this scenario based on observational data, i.e. data where the intervention was not applied randomly, for both continuous and binary outcome variables. However, patient outcome is often recorded in terms of time-to-event data, comprising right-censored event times if an event does not occur within the observation period. Albeit their enormous importance, time-to-event data are rarely used for treatment optimization. We suggest an approach named BITES (Balanced Individual Treatment Effect for Survival data), which combines a treatment-specific semi-parametric Cox loss with a treatment-balanced deep neural network; i.e. we regularize differences between treated and non-treated patients using Integral Probability Metrics (IPM). RESULTS: We show in simulation studies that this approach outperforms the state of the art. Furthermore, we demonstrate in an application to a cohort of breast cancer patients that hormone treatment can be optimized based on six routine parameters. We successfully validated this finding in an independent cohort. AVAILABILITY AND IMPLEMENTATION: We provide BITES as an easy-to-use python implementation including scheduled hyper-parameter optimization (https://github.com/sschrod/BITES). The data underlying this article are available in the CRAN repository at https://rdrr.io/cran/survival/man/gbsg.html and https://rdrr.io/cran/survival/man/rotterdam.html. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

UVA-induced metabolic changes in non-malignant skin cells and the potential role of pyruvate as antioxidant.

The exposure to UVA (320-400 nm) irradiation is a major threat to human skin concerning photoaging and carcinogenesis. It has been shown that UVA irradiation can induce reactive oxygen species (ROS) and DNA mutations, such as 8-hydroxydeoxyguanosine. Furthermore, UVA induces the expression of photoaging-associated matrix metalloproteases (MMPs), especially of matrix metalloprotease 1 (MMP 1) and matrix metalloprotease 3 (MMP 3). In addition to this, it was recently shown that UVA-induced ROS also increase glucose metabolism of melanoma cells, however, the influence of UVA on the glucose metabolism of non-malignant cells of the human skin has, so far, not been investigated in detail. Here, we investigated the UVA-induced changes in glucose metabolism and the functional relevance of these changes in primary fibroblasts-normal non-malignant cells of the skin. These cells showed an UVA-induced enhanced glucose consumption and lactate production and changes in pyruvate production. As it has been proposed that pyruvate could have antioxidant properties we tested the functional relevance of pyruvate as protective agent against UVA-induced ROS. Our initial experiments support earlier publications, demonstrating that pyruvate treated with H2O2 is non-enzymatically transformed to acetate. Furthermore, we show that this decarboxylation of pyruvate to acetate also occurs upon UVA irradiation. In addition to this, we could show that in fibroblasts pyruvate has antioxidant properties as enhanced levels of pyruvate protect cells from UVA-induced ROS and partially from a DNA mutation by the modified base 8-hydroxydeoxyguanosine. Furthermore, we describe for the first time, that the interaction of UVA with pyruvate is relevant for the regulation of photoaging-associated MMP 1 and MMP 3 expression.

Molecular signatures that can be transferred across different omics platforms.

MOTIVATION: Molecular signatures for treatment recommendations are well researched. Still it is challenging to apply them to data generated by different protocols or technical platforms. RESULTS: We analyzed paired data for the same tumors (Burkitt lymphoma, diffuse large B-cell lymphoma) and features that had been generated by different experimental protocols and analytical platforms including the nanoString nCounter and Affymetrix Gene Chip transcriptomics as well as the SWATH and SRM proteomics platforms. A statistical model that assumes independent sample and feature effects accounted for 69-94% of technical variability. We analyzed how variability is propagated through linear signatures possibly affecting predictions and treatment recommendations. Linear signatures with feature weights adding to zero were substantially more robust than unbalanced signatures. They yielded consistent predictions across data from different platforms, both for transcriptomics and proteomics data. Similarly stable were their predictions across data from fresh frozen and matching formalin-fixed paraffin-embedded human tumor tissue. AVAILABILITY AND IMPLEMENTATION: The R-package 'zeroSum' can be downloaded at https://github.com/rehbergT/zeroSum . Complete data and R codes necessary to reproduce all our results can be received from the authors upon request. CONTACT: rainer.spang@ur.de.

Short communication: Use of genomic and metabolic information as well as milk performance records for prediction of subclinical ketosis risk via artificial neural networks.

Subclinical ketosis is one of the most prevalent metabolic disorders in high-producing dairy cows during early lactation. This renders its early detection and prevention important for both economical and animal-welfare reasons. Construction of reliable predictive models is challenging, because traits like ketosis are commonly affected by multiple factors. In this context, machine learning methods offer great advantages because of their universal learning ability and flexibility in integrating various sorts of data. Here, an artificial-neural-network approach was applied to investigate the utility of metabolic, genetic, and milk performance data for the prediction of milk levels of ß-hydroxybutyrate within and across consecutive weeks postpartum. Data were collected from 218 dairy cows during their first 5wk in milk. All animals were genotyped with a 50,000 SNP panel, and weekly information on the concentrations of the milk metabolites glycerophosphocholine and phosphocholine as well as milk composition data (milk yield, fat and protein percentage) was available. The concentration of ß-hydroxybutyric acid in milk was used as target variable in all prediction models. Average correlations between observed and predicted target values up to 0.643 could be obtained, if milk metabolite and routine milk recording data were combined for prediction at the same day within weeks. Predictive performance of metabolic as well as milk performance-based models was higher than that of models based on genetic information.

Nuclear magnetic resonance and mass spectrometry-based milk metabolomics in dairy cows during early and late lactation.

Milk production in dairy cows has dramatically increased over the past few decades. The selection for higher milk yield affects the partitioning of available nutrients, with more energy being allocated to milk synthesis and less to physiological processes essential to fertility and fitness. In this study, the abundance of numerous milk metabolites in early and late lactation was systematically investigated, with an emphasis on metabolites related to energy metabolism. The aim of the study was the identification and correlation of milk constituents to the metabolic status of the cows. To investigate the influence of lactation stage on physiological and metabolic variables, 2 breeds of different productivity were selected for investigation by high-resolution nuclear magnetic resonance spectroscopy and gas chromatography-mass spectrometry. We could reliably quantify 44 different milk metabolites. The results show that biomarkers such as acetone and beta-hydroxybutyrate are clearly correlated to the metabolic status of the individual cows during early lactation. Based on these data, the selection of cows that cope well with the metabolic stress of early lactation should become an option.

On the use of simulated annealing to automatically assign decorrelated components in second-order blind source separation.

In this paper, an automatic assignment tool, called BSS-AutoAssign, for artifact-related decorrelated components within a second-order blind source separation (BSS) is presented. The latter is based on the recently proposed algorithm dAMUSE, which provides an elegant solution to both the BSS and the denoising problem simultaneously. BSS-AutoAssign uses a local principal component analysis (PCA)to approximate the artifact signal and defines a suitable cost function which is optimized using simulated annealing. The algorithms dAMUSE plus BSS-AutoAssign are illustrated by applying them to the separation of water artifacts from two-dimensional nuclear overhauser enhancement (2-D NOESY) spectroscopy signals of proteins dissolved in water.

Solution structure of the Ras binding domain of the protein kinase Byr2 from Schizosaccharomyces pombe.

BACKGROUND: After activation, small GTPases such as Ras transfer the incoming signal to effectors by specifically interacting with the binding domain of these proteins. Structural details of the binding domain of different effectors determine which pathway is predominantly activated. Byr2 from fission yeast is a functional homolog of Raf, which is the direct downstream target of Ras in mammalians that initiates a protein kinase cascade. The amino acid sequence of Byr2's Ras binding domain is only weakly related to that of Raf, and Byr2's three-dimensional structure is unknown. RESULTS: We have solved the 3D structure of the Ras binding domain of Byr2 (Byr2RBD) from Schizosaccharomyces pombe in solution. The structure consists of three alpha helices and a mixed five-stranded beta pleated sheet arranged in the topology betabetaalphabetabetaalphabetaalpha with the first seven canonic secondary structure elements forming a ubiquitin superfold. 15N-(1)H-TROSY-HSQC spectroscopy of the complex of Byr2RBD with Ras*Mg(2+)*GppNHp reveals that the first and second beta strands and the first alpha helix of Byr2 are mainly involved in the protein-protein interaction as observed in other Ras binding domains. Although the putative interaction site of H-Ras from human and Ras1 from S. pombe are identical in sequence, binding to Byr2 leads to small but significant differences in the NMR spectra, indicating a slightly different binding mode. CONCLUSIONS: The ubiquitin superfold appears to be the general structural motif for Ras binding domains even in cases with vanishing sequence identity. However, details of the 3D structure and the interacting interface are different, thereby determining the specifity of the recognition of Ras and Ras-related proteins.

Overcoming the problems associated with poor spectra quality of the protein kinase Byr2 using residual dipolar couplings.

For the Ras-binding domain of the protein kinase Byr2, only a limited number of NOE contacts could be initially assigned unambiguously, as the quality of the NOESY spectra was too poor. However, the use of residual (1)H-(15)N dipolar couplings in the beginning of the structure determination process allows to overcome this problem. We used a three-step recipe for this procedure. A previously unknown structure could be calculated reasonably well with only a limited number of unambiguously assigned NOE contacts.

Computer assisted assignment of 13C or 15N edited 3D-NOESY-HSQC spectra using back calculated and experimental spectra.

A new tool, for the simulation of 15N or 13C edited 3D-NOESY-HSQC spectra using the complete relaxation matrix approach, has been developed and integrated in the program AURELIA. This tool should be particularly useful for the fast and reliable computer assisted assignment of 3D-NOESY-HSQC spectra by comparing back-calculated and experimental spectra in an iterative process. Folded spectra are sometimes used to enhance the digital resolution in the indirect dimensions of multidimensional spectra. However, these spectra are usually difficult to analyze. To simplify this assignment process we have incorporated the simulation and automated annotation of folded peaks into the program. It is hereby possible to simulate multiple folding in all three dimensions of 3D 15N- or 13C-NOESY-HSQC spectra. By comparing experimental 3D-NOESY-HSQC spectra with spectra back calculated from a single trial structure or a set of trial structures, a user can easily check if the final structures explain all experimental NOEs. The new feature has been successfully tested with the histidine-containing phosphocarrier protein HPr from Staphylococcus carnosus.

ApoA-I-binding protein (AI-BP) and its homologues hYjeF_N2 and hYjeF_N3 comprise the YjeF_N domain protein family in humans with a role in spermiogenesis and oogenesis.

The screening for additional human YjeF_N domain containing proteins beside the apolipoprotein A-I interacting protein (AI-BP), identified two other genes designated hYjeF_N2-15q23 (formerly human homologue of yeast edc3) and hYjeF_N3-19p13.11 comprising the human YjeF_N family. AI-BP is ubiquitously expressed, with a predominance of these tissues where the homologues were found to be restricted including brain, mammary gland, testes and ovaries. Immunohistochemistry of human testes and ovaries showed an expression of hYjeF_N3-19p13.11 only in Leydig cells and theca cells, respectively, indicating a role in steroid hormone metabolism. Interestingly, the protein was also strongly expressed in Leydig cell tumors and in thecofibromas. The identification of hYjeF_N2-15q23 in theca cells and granulosa cells in ovaries, in human spermatids of meiotic division part II and the apical membrane of Sertoli cells in testes suggest similar functions in oogenesis and sperm maturation which is strengthened by the identification of the spermatogenesis regulator HMGA1 as a conserved transcription factor. However, in contrast to AI-BP, both homologous proteins are unable to bind apoA-I. These results relate the human YjeF_N domain containing protein family to cholesterol processing and steroid hormone metabolism in spermiogenesis and oogenesis, and AI-BP may link this function to the HDL pathway.

NMR in the SPINE Structural Proteomics project.

This paper describes the developments, role and contributions of the NMR spectroscopy groups in the Structural Proteomics In Europe (SPINE) consortium. Focusing on the development of high-throughput (HTP) pipelines for NMR structure determinations of proteins, all aspects from sample preparation, data acquisition, data processing, data analysis to structure determination have been improved with respect to sensitivity, automation, speed, robustness and validation. Specific highlights are protonless (13)C-direct detection methods and inferential structure determinations (ISD). In addition to technological improvements, these methods have been applied to deliver over 60 NMR structures of proteins, among which are five that failed to crystallize. The inclusion of NMR spectroscopy in structural proteomics pipelines improves the success rate for protein structure determinations.

Assessment by 1H NMR spectroscopy of the structural behaviour of human parathyroid-hormone-related protein(1-34) and its close relationship with the N-terminal fragments of human parathyroid hormone in solution.

Human parathyroid-hormone-related protein (hPTHrP) is a hormone that is over-expressed by a large number of tumors and is produced by a variety of normal cells. Its main biological functions are shown by the N-terminal fragment (1-34) and are similar to those of parathyroid hormone with which it shares a common G-protein-coupled receptor. Hence to gain insight into the structure-function relationship of these hormones we have investigated the solution structure of hPTHrP(1-34) in pure water alone and have monitored the effect of adding TFE. CD spectra in pure water showed that it only possesses a small content of alpha-helical secondary structure, which from the NMR data, consists of a short unstable helix between Gln-16 and Leu-24 with the rest of the peptide in an essentially unstructured state. On adding 50% TFE (v/v) there was a considerable increase in stable secondary structure, without any evidence of stable tertiary structure. The subsequent structure calculations showed the presence of two well defined helices, from Ser-3 to Gly-12 and from Asp-17 to Thr-33, connected by a flexible linker. The similarity in behaviour of hPTHrP(1-34) and the N-terminal fragments of PTH under various solution conditions is shown from the 1H NMR data presented here and an extensive review of the literature data.

The structure of human parathyroid hormone from a study of fragments in solution using 1H NMR spectroscopy and its biological implications.

In order to gain insight into the structure of human parathyroid hormone (hPTH), four fragments [hPTH(1-34), hPTH(18-48), hPTH(28-48), and hPTH(53-84)], which cover all regions of the intact hormone, have been investigated by CD and NMR spectroscopy in combination with distance geometry, and restrained molecular dynamics and energy minimization calculations, under a variety of solution conditions. Significantly, all fragments showed little propensity to form stable structures in aqueous solution alone, and it was only on the addition of trifluoroethanol (TFE) that defined structural features were observed. In an extension of earlier work [Klaus et al. (1991) Biochemistry 30, 6936-6942], hPTH(1-34) in 70% trifluoroethanol (TFE) showed two helices that were longer than in 10% TFE, but essentially showed the same characteristics. Although overlap in the 1H NMR spectra prevented the determination of quantitative NOE data for residues 26-30, the combination of the alpha-proton chemical shift data and quantitative NOE data indicated the helices extend from residues 3 to 13 and 15 to 29. No evidence was found for interaction of the two helical regions. The nature and extent of this second helix in the intact hormone were better defined from the data for hPTH(18-48). Under limiting solution conditions, where the fragment assumed its maximum helical content, a well-defined helix was observed between residues 21 and 38 with a possible discontinuity between Leu-28 and Gln-29. There was little evidence of any form of secondary structure between Gly-38 and the terminus of this fragment, Ser-48. In keeping with this result, the shorter fragment, hPTH(28-48), showed little evidence of stable secondary structure on addition of TFE. From the alpha-proton chemical shifts residues 23-27 appeared to sustain helical structure more readily than the rest of molecule under all solution regimes in both hPTH(1-34) and hPTH(18-48). In contrast to the other two longer fragments hPTH(53-84) showed little propensity for helical secondary structure even at the highest TFE concentrations. However, there was evidence that the molecule did adopt a defined three-dimensional structure. Various long-range NOE's were observed in 10% TFE that allowed the calculation of an open tertiary structure consisting of an initial series of turns surrounded by a loop structure of several loose turns.(ABSTRACT TRUNCATED AT 400 WORDS)

The ice-binding site of sea raven antifreeze protein is distinct from the carbohydrate-binding site of the homologous C-type lectin.

Antifreeze proteins lower the freezing point of their solution by binding to ice and inhibiting its growth. One of several structurally different antifreeze proteins in fishes (type II) is homologous to the carbohydrate-recognition domain of Ca2+-dependent lectins and adopts the same three-dimensional fold. Type II antifreeze proteins from herring and smelt require Ca2+ for binding to ice, whereas this same antifreeze protein in sea raven binds to ice in the absence of Ca2+ and has only two of the five Ca2+-liganding amino acids that are present in the lectin. To locate the ice-binding site, site-directed mutants of the 15 kDa, globular, disulfide-bonded sea raven antifreeze protein were produced by secretion from Pichia pastoris. Pairs of amino acid replacements, insertions, and a peptide loop swap were made in the region equivalent to the sugar-binding site of the lectin that encompasses loops 3 and 4 and beta-sheets 7 and 8. Even the most extensive mutation caused only a 25% decrease in antifreeze activity and demonstrated that the residues corresponding to the Ca2+-binding site are only peripherally involved in ice binding. When adjacent surface residues were mutated, the replacement of one residue, Ser120 by His, caused a 35% decrease in activity by itself and an 80% loss in conjunction with the peptide loop swap mutation. This pivotal sea raven antifreeze protein amino acid does not coincide with the herring ice-binding epicenter, but is located within the region corresponding to the proposed CaCO3-binding surface of a third homologue, the pancreatic stone protein. Intron and exon structure of the sea raven AFP gene also suggests that it might be more closely related to the stone protein gene than to the lectin gene. These results support the notion that this family of proteins has evolved more than one binding surface from the same protein scaffold.

RFAC, a program for automated NMR R-factor estimation.

A computer program (RFAC) has been developed, which allows the automated estimation of residual indices (R-factors) for protein NMR structures and gives a reliable measure for the quality of the structures. The R-factor calculation is based on the comparison of experimental and simulated 1H NOESY NMR spectra. The approach comprises an automatic peak picking and a Bayesian analysis of the data, followed by an automated structure based assignment of the NOESY spectra and the calculation of the R-factor. The major difference to previously published R-factor definitions is that we take the non-assigned experimental peaks into account as well. The number and the intensities of the non-assigned signals are an important measure for the quality of an NMR structure. It turns out that for different problems optimally adapted R-factors should be used which are defined in the paper. The program allows to compute a global R-factor, different R-factors for the intra residual NOEs, the inter residual NOEs, sequential NOEs, medium range NOEs and long range NOEs. Furthermore, R-factors can be calculated for various user defined parts of the molecule or it is possible to obtain a residue-by-residue R-factor. Another possibility is to sort the R-factors according to their corresponding distances. The summary of all these different R-factors should allow the user to judge the structure in detail. The new program has been successfully tested on two medium sized proteins, the cold shock protein (TmCsp) from Termotoga maritima and the histidine containing protein (HPr) from Staphylococcus carnosus. A comparison with a previously published R-factor definition shows that our approach is more sensitive to errors in the calculated structure.

NMR characterization of side chain flexibility and backbone structure in the type I antifreeze protein at near freezing temperatures.

The flexibility of the polar side chains in the alpha-helical Type I antifreeze protein (AFP) near the solution freezing temperature was investigated by two-dimensional nuclear magnetic resonance spectroscopy. These experiments were conducted to define the rotameric conformations of the proposed ice-binding groups, threonines and asparagines, in order to probe the molecular mechanism for ice binding. On the basis of the 3J alpha beta 2 NMR coupling constant values of 7.1, 8.5, 8.5, and 6.8 Hz for residues T2, T13, T24, and T35, respectively, it can be calculated that the regularly spaced ice-binding threonines sample many possible rotameric states prior to ice binding. The lack of a dominant side chain rotamer is further corroborated by nuclear Overhauser distance measurements for T13 and T24. N16 and N27, both with 3J alpha beta 2 and 3J alpha beta 3 coupling constants of 8.4 and 4.5 Hz, respectively, show a slight preference for the side chain conformation with a chi 1 of -60 degrees. These data suggest that prior to ice binding the threonine and asparagine side chains are free to rotate and that a unique preformed ice-binding structure in solution is not apparent. These observations do not support the rigid side chain model proposed recently by an X-ray study [Sicheri, F., & Yang, D. S. C. (1995) Nature 375, 427-431].

Structure of recombinant human parathyroid hormone in solution using multidimensional NMR spectroscopy.

The solution structure of human parathyroid hormone, in the form of recombinant prolyl-hPTH(1-84), has been investigated by multidimensional NMR spectroscopy under conditions (aqueous trifluoroethanol) which favour the structured-state of the protein. Spin systems were identified from 3D 1H DQF (double-quantum filtered)-COSY and TOCSY spectra and sequence-specific assignments were from 2D 1H phase-sensitive NOESY spectra. Signal overlap was resolved in a 3D-NOESY-TOCSY spectrum and assignments were confirmed with 2D NOESY-15N-HMQC (heteronuclear multiple-quantum coherence) spectra taken of a sample universally labeled with 15N. A satisfactory set of final structures was calculated from the quantitative NOE data using restrained molecular dynamics and energy minimization calculations. The N-terminus is dominated by three, well defined helices between Ser-3 to Asn-10, Ser-17 to Lys-27 and Asp-30 to Leu-37, while the most significant structural features in the C-terminus are a short, less-well defined helix between Asn-57 to Ser-62 and a series of loose turns. These two terminal units are joined by an unstructured mid-region. The molecule shows a tendency towards tertiary structure, defined by a number of long-range NOEs. A detailed RMS deviation analysis allowed the final refined structures to be classified into a limited ensemble of stable conformations that reflect the inherent flexibility of the hormone in solution.

Solution NMR structure of the cold-shock protein from the hyperthermophilic bacterium Thermotoga maritima.

Cold-shock proteins (Csps) are a subgroup of the cold-induced proteins preferentially expressed in bacteria and other organisms on reduction of the growth temperature below the physiological temperature. They are related to the cold-shock domain found in eukaryotes and are some of the most conserved proteins known. Their exact function is still not known, but translational regulation, possibly via RNA chaperoning, has been discussed. Here we present the structure of a hyperthermophilic member of the Csp family. The NMR solution structure of TmCsp from Thermotoga maritima, the hyperthermophilic member of this class of proteins, was solved on the basis of 1015 conformational constraints. It contains five beta strands combined in two antiparallel beta sheets making up a beta barrel structure, in which beta strands 1-4 are arranged in a Greek-key topology. The side chain of R2, which is exclusively found in thermophilic members of the Csp family, probably participates in a peripheral ion cluster involving residues D20, R2, E47 and K63, suggesting that the thermostability of TmCsp is based on the peripheral ion cluster around the side chain of R2.

CAMRA: chemical shift based computer aided protein NMR assignments.

A suite of programs called CAMRA (Computer Aided Magnetic Resonance Assignment) has been developed for computer assisted residue-specific assignments of proteins. CAMRA consists of three units: ORB, CAPTURE and PROCESS. ORB predicts NMR chemical shifts for unassigned proteins using a chemical shift database of previously assigned homologous proteins supplemented by a statistically derived chemical shift database in which the shifts are categorized according to their residue, atom and secondary structure type. CAPTURE generates a list of valid peaks from NMR spectra by filtering out noise peaks and other artifacts and then separating the derived peak list into distinct spin systems. PROCESS combines the chemical shift predictions from ORB with the spin systems identified by CAPTURE to obtain residue specific assignments. PROCESS ranks the top choices for an assignment along with scores and confidence values. In contrast to other auto-assignment programs, CAMRA does not use any connectivity information but instead is based solely on matching predicted shifts with observed spin systems. As such, CAMRA represents a new and unique approach for the assignment of protein NMR spectra. CAMRA will be particularly useful in conjunction with other assignment methods and under special circumstances, such as the assignment of flexible regions in proteins where sufficient NOE information is generally not available. CAMRA was tested on two medium-sized proteins belonging to the chemokine family. It was found to be effective in predicting the assignment providing a database of previously assigned proteins with at least 30% sequence identity is available. CAMRA is versatile and can be used to include and evaluate heteronuclear and three-dimensional experiments.