Differentiating viral and bacterial infections: A machine learning model based on routine blood test values.

The growing threat of antibiotic resistance necessitates accurate differentiation between bacterial and viral infections for proper antibiotic administration. In this study, a Virus vs. Bacteria machine learning model was developed to distinguish between these infection types using 16 routine blood test results, C-reactive protein concentration (CRP), biological sex, and age. With a dataset of 44,120 cases from a single medical center, the model achieved an accuracy of 82.2 %, a sensitivity of 79.7 %, a specificity of 84.5 %, a Brier score of 0.129, and an area under the ROC curve (AUC) of 0.905, outperforming a CRP-based decision rule. Notably, the machine learning model enhanced accuracy within the CRP range of 10-40 mg/L, a range where CRP alone is less informative. These results highlight the advantage of integrating multiple blood parameters in diagnostics. The "Virus vs. Bacteria" model paves the way for advanced diagnostic tools, leveraging machine learning to optimize infection management.

COVID-19 diagnosis by routine blood tests using machine learning.

Physicians taking care of patients with COVID-19 have described different changes in routine blood parameters. However, these changes hinder them from performing COVID-19 diagnoses. We constructed a machine learning model for COVID-19 diagnosis that was based and cross-validated on the routine blood tests of 5333 patients with various bacterial and viral infections, and 160 COVID-19-positive patients. We selected the operational ROC point at a sensitivity of 81.9% and a specificity of 97.9%. The cross-validated AUC was 0.97. The five most useful routine blood parameters for COVID-19 diagnosis according to the feature importance scoring of the XGBoost algorithm were: MCHC, eosinophil count, albumin, INR, and prothrombin activity percentage. t-SNE visualization showed that the blood parameters of the patients with a severe COVID-19 course are more like the parameters of a bacterial than a viral infection. The reported diagnostic accuracy is at least comparable and probably complementary to RT-PCR and chest CT studies. Patients with fever, cough, myalgia, and other symptoms can now have initial routine blood tests assessed by our diagnostic tool. All patients with a positive COVID-19 prediction would then undergo standard RT-PCR studies to confirm the diagnosis. We believe that our results represent a significant contribution to improvements in COVID-19 diagnosis.

Application of lectin immobilized on polyHIPE monoliths for bioprocess monitoring of glycosylated proteins.

In-process monitoring of glycosylated protein concentration becomes very important with the introduction of perfusion bioprocesses. Affinity chromatography based on lectins allows selective monitoring when carbohydrates are accessible on the protein surface. In this work, we immobilized lectin on polyHIPE type of monoliths and implemented it for bioprocess monitoring. A spacer was introduced to lectin, which increased binding kinetics toward Fc-fusion protein, demonstrated by bio-layer interferometry. Furthermore, complete desorption using 0.25 M galactose was shown. Affinity column exhibited linearity in the range between 0.5 and 8 mg/ml and flow-unaffected binding for the flow-rates between 0.5 and 8 ml/min. Long-term stability over at least four months period was demonstrated. No unspecific binding of culture media components, including host cell proteins and DNA, was detected. Results obtained by affinity column matched concentration values obtained by a reference method.

Diagnosing brain tumours by routine blood tests using machine learning.

Routine blood test results are assumed to contain much more information than is usually recognised even by the most experienced clinicians. Using routine blood tests from 15,176 neurological patients we built a machine learning predictive model for the diagnosis of brain tumours. We validated the model by retrospective analysis of 68 consecutive brain tumour and 215 control patients presenting to the neurological emergency service. Only patients with head imaging and routine blood test data were included in the validation sample. The sensitivity and specificity of the adapted tumour model in the validation group were 96% and 74%, respectively. Our data demonstrate the feasibility of brain tumour diagnosis from routine blood tests using machine learning. The reported diagnostic accuracy is comparable and possibly complementary to that of imaging studies. The presented machine learning approach opens a completely new avenue in the diagnosis of these grave neurological diseases and demonstrates the utility of valuable information obtained from routine blood tests.

Proper evaluation of chemical cross-linking-based spatial restraints improves the precision of modeling homo-oligomeric protein complexes.

BACKGROUND: The function of oligomeric proteins is inherently linked to their quaternary structure. In the absence of high-resolution data, low-resolution information in the form of spatial restraints can significantly contribute to the precision and accuracy of structural models obtained using computational approaches. To obtain such restraints, chemical cross-linking coupled with mass spectrometry (XL-MS) is commonly used. However, the use of XL-MS in the modeling of protein complexes comprised of identical subunits (homo-oligomers) is often hindered by the inherent ambiguity of intra- and inter-subunit connection assignment. RESULTS: We present a comprehensive evaluation of (1) different methods for inter-residue distance calculations, and (2) different approaches for the scoring of spatial restraints. Our results show that using Solvent Accessible Surface distances (SASDs) instead of Euclidean distances (EUCs) greatly reduces the assignation ambiguity and delivers better modeling precision. Furthermore, ambiguous connections should be considered as inter-subunit only when the intra-subunit alternative exceeds the distance threshold. Modeling performance can also be improved if symmetry, characteristic for most homo-oligomers, is explicitly defined in the scoring function. CONCLUSIONS: Our findings provide guidelines for proper evaluation of chemical cross-linking-based spatial restraints in modeling homo-oligomeric protein complexes, which could facilitate structural characterization of this important group of proteins.

An application of machine learning to haematological diagnosis.

Quick and accurate medical diagnoses are crucial for the successful treatment of diseases. Using machine learning algorithms and based on laboratory blood test results, we have built two models to predict a haematologic disease. One predictive model used all the available blood test parameters and the other used only a reduced set that is usually measured upon patient admittance. Both models produced good results, obtaining prediction accuracies of 0.88 and 0.86 when considering the list of five most likely diseases and 0.59 and 0.57 when considering only the most likely disease. The models did not differ significantly, which indicates that a reduced set of parameters can represent a relevant "fingerprint" of a disease. This knowledge expands the model's utility for use by general practitioners and indicates that blood test results contain more information than physicians generally recognize. A clinical test showed that the accuracy of our predictive models was on par with that of haematology specialists. Our study is the first to show that a machine learning predictive model based on blood tests alone can be successfully applied to predict haematologic diseases. This result and could open up unprecedented possibilities for medical diagnosis.

Analysis of the N-terminal region of human MLKL, as well as two distinct MLKL isoforms, reveals new insights into necroptotic cell death.

The pseudokinase mixed lineage kinase domain-like (MLKL) is an essential effector of necroptotic cell death. Two distinct human MLKL isoforms have previously been reported, but their capacities to trigger cell death have not been compared directly. Herein, we examine these two MLKL isoforms, and further probe the features of the human MLKL N-terminal domain that are required for cell death. Expression in HEK293T cells of the N-terminal 201 amino acids (aa) of human MLKL is sufficient to cause cell death, whereas expression of the first 154 aa is not. Given that aa 1-125 are able to initiate necroptosis, our findings indicate that the helix that follows this region restrains necroptotic activity, which is again restored in longer constructs. Furthermore, MLKL isoform 2 (MLKL2), which lacks much of the regulatory pseudokinase domain, is a much more potent inducer of cell death than MLKL isoform 1 (MLKL1) in ectopic expression studies in HEK293T cells. Modelling predicts that a C-terminal helix constrains the activity of MLKL1, but not MLKL2. Although both isoforms are expressed by human monocyte-derived macrophages at the mRNA level, MLKL2 is expressed at much lower levels. We propose that it may have a regulatory role in controlling macrophage survival, either in the steady state or in response to specific stimuli.

Phosphorylation of C-terminal tyrosine residue 526 in FUS impairs its nuclear import.

Aberrant cytoplasmic aggregation of FUS, which is caused by mutations primarily in the C-terminal nuclear localisation signal, is associated with 3% of cases of familial amyotrophic lateral sclerosis (ALS). FUS aggregates are also pathognomonic for 10% of all frontotemporal lobar degeneration (FTLD) cases; however, these cases are not associated with mutations in the gene encoding FUS. This suggests that there are differences in the mechanisms that drive inclusion formation of FUS in ALS and FTLD. Here, we show that the C-terminal tyrosine residue at position 526 of FUS is crucial for normal nuclear import. This tyrosine is subjected to phosphorylation, which reduces interaction with transportin 1 and might consequentially affect the transport of FUS into the nucleus. Furthermore, we show that this phosphorylation can occur through the activity of the Src family of kinases. Our study implicates phosphorylation as an additional mechanism by which nuclear transport of FUS might be regulated and potentially perturbed in ALS and FTLD.

Crystal structure and its bearing towards an understanding of key biological functions of EpCAM.

EpCAM (epithelial cell adhesion molecule), a stem and carcinoma cell marker, is a cell surface protein involved in homotypic cell-cell adhesion via intercellular oligomerization and proliferative signalling via proteolytic cleavage. Despite its use as a diagnostic marker and being a drug target, structural details of this conserved vertebrate-exclusive protein remain unknown. Here we present the crystal structure of a heart-shaped dimer of the extracellular part of human EpCAM. The structure represents a cis-dimer that would form at cell surfaces and may provide the necessary structural foundation for the proposed EpCAM intercellular trans-tetramerization mediated by a membrane-distal region. By combining biochemical, biological and structural data on EpCAM, we show how proteolytic processing at various sites could influence structural integrity, oligomeric state and associated functionality of the molecule. We also describe the epitopes of this therapeutically important protein and explain the antigenicity of its regions.

N-terminally truncated forms of human cathepsin F accumulate in aggresome-like inclusions.

The contribution of individual cysteine cathepsins as positive mediators of programmed cell death is dependent on several factors, such as the type of stimuli, intensity and duration of the stimulus, and cell type involved. Of the eleven human cysteine cathepsins, cathepsin F is the only cathepsin that exhibits an extended N-terminal proregion, which contains a cystatin-like domain. We predicted that the wild-type human cathepsin F contains three natively disordered regions within the enzyme's propeptide and various amino acid stretches with high fibrillation propensity. Wild-type human cathepsin F and its N-terminally truncated forms, Ala(20)-Asp(484) (Δ(19)CatF), Pro(126)-Asp(484) (Δ(125)CatF), and Met(147)-Asp(484) (Δ(146)CatF) were cloned into the pcDNA3 vector and overexpressed in HEK 293T cells. Wild-type human cathepsin F displayed a clear vesicular labeling and colocalized with the LAMP2 protein, a lysosomal marker. However, all three N-terminally truncated forms of human cathepsin F were recovered as insoluble proteins, suggesting that the deletion of at least the signal peptides (Δ(19)CatF), results in protein aggregation. Noteworthy, they concentrated large perinuclear-juxtanuclear aggregates that accumulated within aggresome-like inclusions. These inclusions showed p62-positive immunoreactivity and were colocalized with the autophagy marker LC3B, but not with the LAMP2 protein. In addition, an approximately 2-3 fold increase in DEVDase activity was not sufficient to induce apoptotic cell death. These results suggested the clearance of the N-terminally truncated forms of human cathepsin F via the autophagy pathway, underlying its protective and prosurvival mechanisms.

Modulation of contact order effects in the two-state folding of stefins A and B.

It is well established that contact order and folding rates are correlated for small proteins. The folding rates of stefins A and B differ by nearly two orders of magnitude despite sharing an identical native fold and hence contact order. We break down the determinants of this behavior and demonstrate that the modulation of contact order effects can be accounted for by the combined contributions of a framework-like mechanism, characterized by intrinsic helix stabilities, together with nonnative helical backbone conformation and nonnative hydrophobic interactions within the folding transition state. These contributions result in the formation of nonnative interactions in the transition state as evidenced by the opposing effects on folding rate and stability of these proteins.

Mechanisms of amyloid fibril formation--focus on domain-swapping.

Conformational diseases constitute a group of heterologous disorders in which a constituent host protein undergoes changes in conformation, leading to aggregation and deposition. To understand the molecular mechanisms of the process of amyloid fibril formation, numerous in vitro and in vivo studies, including model and pathologically relevant proteins, have been performed. Understanding the molecular details of these processes is of major importance to understand neurodegenerative diseases and could contribute to more effective therapies. Many models have been proposed to describe the mechanism by which proteins undergo ordered aggregation into amyloid fibrils. We classify these as: (a) templating and nucleation; (b) linear, colloid-like assembly of spherical oligomers; and (c) domain-swapping. In this review, we stress the role of domain-swapping and discuss the role of proline switches.

Structural and functional characterization of three DsbA paralogues from Salmonella enterica serovar typhimurium.

In prototypic Escherichia coli K-12 the introduction of disulfide bonds into folding proteins is mediated by the Dsb family of enzymes, primarily through the actions of the highly oxidizing protein EcDsbA. Homologues of the Dsb catalysts are found in most bacteria. Interestingly, pathogens have developed distinct Dsb machineries that play a pivotal role in the biogenesis of virulence factors, hence contributing to their pathogenicity. Salmonella enterica serovar (sv.) Typhimurium encodes an extended number of sulfhydryl oxidases, namely SeDsbA, SeDsbL, and SeSrgA. Here we report a comprehensive analysis of the sv. Typhimurium thiol oxidative system through the structural and functional characterization of the three Salmonella DsbA paralogues. The three proteins share low sequence identity, which results in several unique three-dimensional characteristics, principally in areas involved in substrate binding and disulfide catalysis. Furthermore, the Salmonella DsbA-like proteins also have different redox properties. Whereas functional characterization revealed some degree of redundancy, the properties of SeDsbA, SeDsbL, and SeSrgA and their expression pattern in sv. Typhimurium indicate a diverse role for these enzymes in virulence.

Averaged kick maps: less noise, more signal... and probably less bias.

Use of reliable density maps is crucial for rapid and successful crystal structure determination. Here, the averaged kick (AK) map approach is investigated, its application is generalized and it is compared with other map-calculation methods. AK maps are the sum of a series of kick maps, where each kick map is calculated from atomic coordinates modified by random shifts. As such, they are a numerical analogue of maximum-likelihood maps. AK maps can be unweighted or maximum-likelihood (sigma(A)) weighted. Analysis shows that they are comparable and correspond better to the final model than sigma(A) and simulated-annealing maps. The AK maps were challenged by a difficult structure-validation case, in which they were able to clarify the problematic region in the density without the need for model rebuilding. The conclusion is that AK maps can be useful throughout the entire progress of crystal structure determination, offering the possibility of improved map interpretation.

Crystallography and protein-protein interactions: biological interfaces and crystal contacts.

Crystallography is commonly used for studying the structures of protein-protein complexes. However, a crystal structure does not define a unique protein-protein interface, and distinguishing a 'biological interface' from 'crystal contacts' is often not straightforward. A number of computational approaches exist for distinguishing them, but their error rate is high, emphasizing the need to obtain further data on the biological interface using complementary structural and functional approaches. In addition to reviewing the computational and experimental approaches for addressing this problem, we highlight two relevant examples. The first example from our laboratory involves the structure of acyl-CoA thioesterase 7, where each domain of this two-domain protein was crystallized separately, but both yielded a non-functional assembly. The structure of the full-length protein was uncovered using a combination of complementary approaches including chemical cross-linking, analytical ultracentrifugation and mutagenesis. The second example involves the platelet glycoprotein Ibalpha-thrombin complex. Two groups reported the crystal structures of this complex, but all the interacting interfaces differed between the two structures. Our computational analysis did not fully resolve the reasons for the discrepancies, but provided interesting insights into the system. This review highlights the need to complement crystallographic studies with complementary experimental and computational approaches.

Kap95p binding induces the switch loops of RanGDP to adopt the GTP-bound conformation: implications for nuclear import complex assembly dynamics.

The asymmetric distribution of the nucleotide-bound state of Ran across the nuclear envelope is crucial for determining the directionality of nuclear transport. In the nucleus, Ran is primarily in the guanosine 5'-triphosphate (GTP)-bound state, whereas in the cytoplasm, Ran is primarily guanosine 5'-diphosphate (GDP)-bound. Conformational changes within the Ran switch I and switch II loops are thought to modulate its affinity for importin-beta. Here, we show that RanGDP and importin-beta form a stable complex with a micromolar dissociation constant. This complex can be dissociated by importin-beta binding partners such as importin-alpha. Surprisingly, the crystal structure of the Kap95p-RanGDP complex shows that Kap95p induces the switch I and II regions of RanGDP to adopt a conformation that resembles that of the GTP-bound form. The structure of the complex provides insights into the structural basis for the gradation of affinities regulating nuclear protein transport.

A general target selection method for crystallographic proteomics.

Increasing the success in obtaining structures and maximizing the value of the structures determined are the two major goals of target selection in structural proteomics. This chapter presents an efficient and flexible target selection procedure supplemented with a Web-based resource that is suitable for small- to large-scale structural genomics projects that use crystallography as the major means of structure determination. Based on three criteria, biological significance, structural novelty, and "crystallizability," the approach first removes (filters) targets that do not meet minimal criteria and then ranks the remaining targets based on their "crystallizability" estimates. This novel procedure was designed to maximize selection efficiency, and its prevailing criteria categories make it suitable for a broad range of structural proteomics projects.

A medium or high throughput protein refolding assay.

Expression of insoluble protein in E. coli is a major bottleneck of high throughput structural biology projects. Refolding proteins into native conformations from inclusion bodies could significantly increase the number of protein targets that can be taken on to structural studies. This chapter presents a simple assay for screening insoluble protein targets and identifying those that are most amenable to refolding. The assay is based on the observation that when proteins are refolded while bound to metal affinity resin, misfolded proteins are generally not eluted by imidazole. This difference is exploited here to distinguish between folded and misfolded proteins. Two implementations of the assay are described. The assay fits well into a standard high throughput structural biology pipeline, because it begins with the inclusion body preparations that are a byproduct of small-scale, automated expression and purification trials and does not require additional facilities. Two formats of the assay are described, a manual assay that is useful for screening small numbers of targets, and an automated implementation that is useful for large numbers of targets.

Overview of the pipeline for structural and functional characterization of macrophage proteins at the university of queensland.

This chapter describes the methodology adopted in a project aimed at structural and functional characterization of proteins that potentially play an important role in mammalian macrophages. The methodology that underpins this project is applicable to both small research groups and larger structural genomics consortia. Gene products with putative roles in macrophage function are identified using gene expression information obtained via DNA microarray technology. Specific targets for structural and functional characterization are then selected based on a set of criteria aimed at maximizing insight into function. The target proteins are cloned using a modification of Gateway cloning technology, expressed with hexa-histidine tags in E. coli, and purified to homogeneity using a combination of affinity and size exclusion chromatography. Purified proteins are finally subjected to crystallization trials and/or NMR-based screening to identify candidates for structure determination. Where crystallography and NMR approaches are unsuccessful, chemical cross-linking is employed to obtain structural information. This resulting structural information is used to guide cell biology experiments to further investigate the cellular and molecular function of the targets in macrophage biology. Jointly, the data sheds light on the molecular and cellular functions of macrophage proteins.