Generalising uncertainty improves accuracy and safety of deep learning analytics applied to oncology.

Uncertainty estimation is crucial for understanding the reliability of deep learning (DL) predictions, and critical for deploying DL in the clinic. Differences between training and production datasets can lead to incorrect predictions with underestimated uncertainty. To investigate this pitfall, we benchmarked one pointwise and three approximate Bayesian DL models for predicting cancer of unknown primary, using three RNA-seq datasets with 10,968 samples across 57 cancer types. Our results highlight that simple and scalable Bayesian DL significantly improves the generalisation of uncertainty estimation. Moreover, we designed a prototypical metric-the area between development and production curve (ADP), which evaluates the accuracy loss when deploying models from development to production. Using ADP, we demonstrate that Bayesian DL improves accuracy under data distributional shifts when utilising 'uncertainty thresholding'. In summary, Bayesian DL is a promising approach for generalising uncertainty, improving performance, transparency, and safety of DL models for deployment in the real world.

SARS-CoV-2 ORF6 disrupts innate immune signalling by inhibiting cellular mRNA export.

SARS-CoV-2 is a betacoronavirus and the etiological agent of COVID-19, a devastating infectious disease. Due to its far-reaching effect on human health, there is an urgent and growing need to understand the viral molecular biology of SARS-CoV-2 and its interaction with the host cell. SARS-CoV-2 encodes 9 predicted accessory proteins, which are presumed to be dispensable for in vitro replication, most likely having a role in modulating the host cell environment to aid viral replication. Here we show that the ORF6 accessory protein interacts with cellular Rae1 to inhibit cellular protein production by blocking mRNA export. We utilised cell fractionation coupled with mRNAseq to explore which cellular mRNA species are affected by ORF6 expression and show that ORF6 can inhibit the export of many mRNA including those encoding antiviral factors such as IRF1 and RIG-I. We also show that export of these mRNA is blocked in the context of SARS-CoV-2 infection. Together, our studies identify a novel mechanism by which SARS-CoV-2 can manipulate the host cell environment to supress antiviral responses, providing further understanding to the replication strategies of a virus that has caused an unprecedented global health crisis.

A deep sleep stage in Drosophila with a functional role in waste clearance.

Sleep is a highly conserved state, suggesting that sleep's benefits outweigh the increased vulnerability it brings. Yet, little is known about how sleep fulfills its functions. Here, we used video tracking in tethered flies to identify a discrete deep sleep stage in Drosophila, termed proboscis extension sleep, that is defined by repeated stereotyped proboscis extensions and retractions. Proboscis extension sleep is accompanied by highly elevated arousal thresholds and decreased brain activity, indicative of a deep sleep state. Preventing proboscis extensions increases injury-related mortality and reduces waste clearance. Sleep deprivation reduces waste clearance and during subsequent rebound sleep, sleep, proboscis extensions, and waste clearance are increased. Together, these results provide evidence of a discrete deep sleep stage that is linked to a specific function and suggest that waste clearance is a core and ancient function of deep sleep.

Verifying explainability of a deep learning tissue classifier trained on RNA-seq data.

For complex machine learning (ML) algorithms to gain widespread acceptance in decision making, we must be able to identify the features driving the predictions. Explainability models allow transparency of ML algorithms, however their reliability within high-dimensional data is unclear. To test the reliability of the explainability model SHapley Additive exPlanations (SHAP), we developed a convolutional neural network to predict tissue classification from Genotype-Tissue Expression (GTEx) RNA-seq data representing 16,651 samples from 47 tissues. Our classifier achieved an average F1 score of 96.1% on held-out GTEx samples. Using SHAP values, we identified the 2423 most discriminatory genes, of which 98.6% were also identified by differential expression analysis across all tissues. The SHAP genes reflected expected biological processes involved in tissue differentiation and function. Moreover, SHAP genes clustered tissue types with superior performance when compared to all genes, genes detected by differential expression analysis, or random genes. We demonstrate the utility and reliability of SHAP to explain a deep learning model and highlight the strengths of applying ML to transcriptome data.

Duplication and divergence of the retrovirus restriction gene Fv1 in Mus caroli allows protection from multiple retroviruses.

Viruses and their hosts are locked in an evolutionary race where resistance to infection is acquired by the hosts while viruses develop strategies to circumvent these host defenses. Forming one arm of the host defense armory are cell autonomous restriction factors like Fv1. Originally described as protecting laboratory mice from infection by murine leukemia virus (MLV), Fv1s from some wild mice have also been found to restrict non-MLV retroviruses, suggesting an important role in the protection against viruses in nature. We surveyed the Fv1 genes of wild mice trapped in Thailand and characterized their restriction activities against a panel of retroviruses. An extra copy of the Fv1 gene, named Fv7, was found on chromosome 6 of three closely related Asian species of mice: Mus caroli, M. cervicolor, and M. cookii. The presence of flanking repeats suggested it arose by LINE-mediated retroduplication within their most recent common ancestor. A high degree of natural variation was observed in both Fv1 and Fv7 and, on top of positive selection at certain residues, insertions and deletions were present that changed the length of the reading frames. These genes exhibited a range of restriction phenotypes, with activities directed against gamma-, spuma-, and lentiviruses. It seems likely, at least in the case of M. caroli, that the observed gene duplication may expand the breadth of restriction beyond the capacity of Fv1 alone and that one or more such viruses have recently driven or continue to drive the evolution of the Fv1 and Fv7 genes.

A helical LC3-interacting region mediates the interaction between the retroviral restriction factor Trim5α and mammalian autophagy-related ATG8 proteins.

The retroviral restriction factor tripartite motif-containing 5α (Trim5α) acts during the early postentry stages of the retroviral life cycle to block infection by a broad range of retroviruses, disrupting reverse transcription and integration. The mechanism of this restriction is poorly understood, but it has recently been suggested to involve recruitment of components of the autophagy machinery, including members of the mammalian autophagy-related 8 (ATG8) family involved in targeting proteins to the autophagosome. To better understand the molecular details of this interaction, here we utilized analytical ultracentrifugation to characterize the binding of six ATG8 isoforms and determined the crystal structure of the Trim5α Bbox coiled-coil region in complex with one member of the mammalian ATG8 proteins, autophagy-related protein LC3 B (LC3B). We found that Trim5α binds all mammalian ATG8s and that, unlike the typical LC3-interacting region (LIR) that binds to mammalian ATG8s through a ß-strand motif comprising approximately six residues, LC3B binds to Trim5α via the α-helical coiled-coil region. The orientation of the structure demonstrated that LC3B could be accommodated within a Trim5α assembly that can bind the retroviral capsid. However, mutation of the binding interface does not affect retroviral restriction. Comparison of the typical linear ß-strand LIR with our atypical helical LIR reveals a conservation of the presentation of residues that are required for the interaction with LC3B. This observation expands the range of LC3B-binding proteins to include helical binding motifs and demonstrates a link between Trim5α and components of the autophagosome.

Evolutionary journey of the retroviral restriction gene Fv1.

Both exogenous and endogenous retroviruses have long been studied in mice, and some of the earliest mouse studies focused on the heritability of genetic factors influencing permissivity and resistance to infection. The prototypic retroviral restriction factor, Fv1, is now understood to exhibit a degree of control across multiple retroviral genera and is highly diverse within Mus To better understand the age and evolutionary history of Fv1, a comprehensive survey of the Muroidea was conducted, allowing the progenitor integration to be dated to ∼45 million years. Intact coding potential is visible beyond Mus, and sequence analysis reveals strong signatures of positive selection also within field mice, ApodemusFv1's survival for such a period implies a recurring and shifting retroviral burden imparting the necessary selective pressures-an influence likely also common to analogous factors. Regions of Fv1 adapt cooperatively, highlighting its preference for repeated structures and suggesting that this functionally constrained aspect of the retroviral capsid lattice presents a common target in the evolution of intrinsic immunity.

Acute control of the sleep switch in Drosophila reveals a role for gap junctions in regulating behavioral responsiveness.

Sleep is a dynamic process in most animals, involving distinct stages that probably perform multiple functions for the brain. Before sleep functions can be initiated, it is likely that behavioral responsiveness to the outside world needs to be reduced, even while the animal is still awake. Recent work in Drosophila has uncovered a sleep switch in the dorsal fan-shaped body (dFB) of the fly's central brain, but it is not known whether these sleep-promoting neurons also govern the acute need to ignore salient stimuli in the environment during sleep transitions. We found that optogenetic activation of the sleep switch suppressed behavioral responsiveness to mechanical stimuli, even in awake flies, indicating a broader role for these neurons in regulating arousal. The dFB-mediated suppression mechanism and its associated neural correlates requires innexin6 expression, suggesting that the acute need to reduce sensory perception when flies fall asleep is mediated in part by electrical synapses.

Oscillatory brain activity in spontaneous and induced sleep stages in flies.

Sleep is a dynamic process comprising multiple stages, each associated with distinct electrophysiological properties and potentially serving different functions. While these phenomena are well described in vertebrates, it is unclear if invertebrates have distinct sleep stages. We perform local field potential (LFP) recordings on flies spontaneously sleeping, and compare their brain activity to flies induced to sleep using either genetic activation of sleep-promoting circuitry or the GABAA agonist Gaboxadol. We find a transitional sleep stage associated with a 7-10 Hz oscillation in the central brain during spontaneous sleep. Oscillatory activity is also evident when we acutely activate sleep-promoting neurons in the dorsal fan-shaped body (dFB) of Drosophila. In contrast, sleep following Gaboxadol exposure is characterized by low-amplitude LFPs, during which dFB-induced effects are suppressed. Sleep in flies thus appears to involve at least two distinct stages: increased oscillatory activity, particularly during sleep induction, followed by desynchronized or decreased brain activity.

SAMHD1 enhances nucleoside-analogue efficacy against HIV-1 in myeloid cells.

SAMHD1 is an intracellular enzyme that specifically degrades deoxynucleoside triphosphates into component nucleoside and inorganic triphosphate. In myeloid-derived dendritic cells and macrophages as well as resting T-cells, SAMHD1 blocks HIV-1 infection through this dNTP triphosphohydrolase activity by reducing the cellular dNTP pool to a level that cannot support productive reverse transcription. We now show that, in addition to this direct effect on virus replication, manipulating cellular SAMHD1 activity can significantly enhance or decrease the anti-HIV-1 efficacy of nucleotide analogue reverse transcription inhibitors presumably as a result of modulating dNTP pools that compete for recruitment by viral polymerases. Further, a variety of other nucleotide-based analogues, not normally considered antiretrovirals, such as the anti-herpes drugs Aciclovir and Ganciclovir and the anti-cancer drug Clofarabine are now revealed as potent anti-HIV-1 agents, under conditions of low dNTPs. This in turn suggests novel uses for nucleotide analogues to inhibit HIV-1 in differentiated cells low in dNTPs.

Expression levels of Fv1: effects on retroviral restriction specificities.

BACKGROUND: The mouse protein Fv1 is a factor that can confer resistance to retroviral infection. The two major Fv1 alleles from laboratory mice, Fv1 (n) and Fv1 (b) , restrict infection by different murine leukaemia viruses (MLVs). Fv1(n) restricts B-tropic MLV, but not N-tropic MLV or NB-tropic MLV. In cells expressing Fv1(b) at natural levels, only N-MLV is restricted, however restriction of NB-MLV and partial restriction of B-MLV were observed when recombinant Fv1(b) was expressed from an MLV promoter in Fv1 null Mus dunni tail fibroblast cells. To investigate the relationship between expression level and restriction specificity we have developed new retroviral delivery vectors which allow inducible expression of Fv1, and yet allow sufficient production of fluorescent reporter proteins for analysis in our FACS-based restriction assay. RESULTS: We demonstrated that at concentrations close to the endogenous expression level, Fv1(b) specifically restricts only N-MLV, but restriction of NB-MLV, and to a lesser extent B-MLV, could be gained by increasing the protein level of Fv1(b). By contrast, we found that even when Fv1(n) is expressed at very high levels, no significant inhibition of N-MLV or NB-MLV could be observed. Study of Fv1 mutants using this assay led to the identification of determinants for N/B tropism at an expression level close to that of endogenous Fv1(n) and Fv1(b). We also compared the recently described restriction activities of wild mice Fv1 proteins directed against non-MLV retroviruses when expressed at different levels. Fv1 from M. spretus restricted N-MLV, B-MLV and equine infectious anaemia virus equally even at low concentrations, while Fv1 from M. macedonicus showed even stronger restriction against equine infectious anaemia virus than to N-MLV. Restriction of feline foamy virus by Fv1 of M. caroli occurred at levels equivalent to MLV restriction. CONCLUSIONS: Our data indicate that for some but not all Fv1 proteins, gain of restriction activities could be achieved by increasing the expression level of Fv1. However such a concentration dependent effect is not seen with most Fv1s and cannot explain the recently reported activities against non-MLVs. It will be interesting to examine whether overexpression of other capsid binding restriction factors such as TRIM5α or Mx2 result in novel restriction specificities.

Structural studies of postentry restriction factors reveal antiparallel dimers that enable avid binding to the HIV-1 capsid lattice.

Restriction factors (RFs) form important components of host defenses to retroviral infection. The Fv1, Trim5α, and TrimCyp RFs contain N-terminal dimerization and C-terminal specificity domains that target assembled retroviral capsid (CA) proteins enclosing the viral core. However, the molecular detail of the interaction between RFs and their CA targets is unknown. Therefore, we have determined the crystal structure of the B-box and coiled-coil (BCC) region from Trim5α and used small-angle X-ray scattering to examine the solution structure of Trim5α BCC, the dimerization domain of Fv1 (Fv1Ntd), and the hybrid restriction factor Fv1Cyp comprising Fv1NtD fused to the HIV-1 binding protein Cyclophilin A (CypA). These data reveal that coiled-coil regions of Fv1 and Trim5α form extended antiparallel dimers. In Fv1Cyp, two CypA moieties are located at opposing ends, creating a molecule with a dumbbell appearance. In Trim5α, the B-boxes are located at either end of the coiled-coil, held in place by interactions with a helical motif from the L2 region of the opposing monomer. A comparative analysis of Fv1Cyp and CypA binding to a preformed HIV-1 CA lattice reveals how RF dimerization enhances the affinity of interaction through avidity effects. We conclude that the antiparallel organization of the NtD regions of Fv1 and Trim5α dimers correctly positions C-terminal specificity and N-terminal effector domains and facilitates stable binding to adjacent CA hexamers in viral cores.

Evolution of the retroviral restriction gene Fv1: inhibition of non-MLV retroviruses.

Fv1 is the prototypic restriction factor that protects against infection by the murine leukemia virus (MLV). It was first identified in cells that were derived from laboratory mice and was found to be homologous to the gag gene of an endogenous retrovirus (ERV). To understand the evolution of the host restriction gene from its retroviral origins, Fv1s from wild mice were isolated and characterized. Most of these possess intact open reading frames but not all restricted N-, B-, NR-or NB-tropic MLVs, suggesting that other viruses could have played a role in the selection of the gene. The Fv1s from Mus spretus and Mus caroli were found to restrict equine infectious anemia virus (EIAV) and feline foamy virus (FFV) respectively, indicating that Fv1 could have a broader target range than previously thought, including activity against lentiviruses and spumaviruses. Analyses of the Fv1 sequences revealed a number of residues in the C-terminal region that had evolved under positive selection. Four of these selected residues were found to be involved in the novel restriction by mapping studies. These results strengthen the similarities between the two capsid binding restriction factors, Fv1 and TRIM5α, which support the hypothesis that Fv1 defended mice against waves of retroviral infection possibly including non-MLVs as well as MLVs.

Apparent effect of rabbit endogenous lentivirus type K acquisition on retrovirus restriction by lagomorph Trim5αs.

To test the hypothesis that rabbit endogenous lentivirus type K (RELIK) could play a role in shaping the evolution of TRIM5α, the susceptibility of viruses containing the RELIK capsid (CA) to TRIM5 restriction was evaluated. RELIK CA-containing viruses were susceptible to the TRIM5αs from Old World monkeys but were unaffected by most ape or New World monkey factors. TRIM5αs from various lagomorph species were also isolated and tested for anti-retroviral activity. The TRIM5αs from both cottontail rabbit and pika restrict a range of retroviruses, including HIV-1, HIV-2, FIV, EIAV and N-MLV. TRIM5αs from the European and cottontail rabbit, which have previously been found to contain RELIK, also restricted RELIK CA-containing viruses, whereas a weaker restriction was observed with chimeric TRIM5α containing the B30.2 domain from the pika, which lacks RELIK. Taken together, these results could suggest that the pika had not been exposed to exogenous RELIK and that endogenized RELIK might exert a selective pressure on lagomorph TRIM5α.

A unique spumavirus Gag N-terminal domain with functional properties of orthoretroviral matrix and capsid.

The Spumaretrovirinae, or foamyviruses (FVs) are complex retroviruses that infect many species of monkey and ape. Although FV infection is apparently benign, trans-species zoonosis is commonplace and has resulted in the isolation of the Prototypic Foamy Virus (PFV) from human sources and the potential for germ-line transmission. Despite little sequence homology, FV and orthoretroviral Gag proteins perform equivalent functions, including genome packaging, virion assembly, trafficking and membrane targeting. In addition, PFV Gag interacts with the FV Envelope (Env) protein to facilitate budding of infectious particles. Presently, there is a paucity of structural information with regards FVs and it is unclear how disparate FV and orthoretroviral Gag molecules share the same function. Therefore, in order to probe the functional overlap of FV and orthoretroviral Gag and learn more about FV egress and replication we have undertaken a structural, biophysical and virological study of PFV-Gag. We present the crystal structure of a dimeric amino terminal domain from PFV, Gag-NtD, both free and in complex with the leader peptide of PFV Env. The structure comprises a head domain together with a coiled coil that forms the dimer interface and despite the shared function it is entirely unrelated to either the capsid or matrix of Gag from other retroviruses. Furthermore, we present structural, biochemical and virological data that reveal the molecular details of the essential Gag-Env interaction and in addition we also examine the specificity of Trim5α restriction of PFV. These data provide the first information with regards to FV structural proteins and suggest a model for convergent evolution of gag genes where structurally unrelated molecules have become functionally equivalent.

A dynamic deep sleep stage in Drosophila.

How might one determine whether simple animals such as flies sleep in stages? Sleep in mammals is a dynamic process involving different stages of sleep intensity, and these are typically associated with measurable changes in brain activity (Blake and Gerard, 1937; Rechtschaffen and Kales, 1968; Webb and Agnew, 1971). Evidence for different sleep stages in invertebrates remains elusive, even though it has been well established that many invertebrate species require sleep (Campbell and Tobler, 1984; Hendricks et al., 2000; Shaw et al., 2000; Sauer et al., 2003). Here we used electrophysiology and arousal-testing paradigms to show that the fruit fly, Drosophila melanogaster, transitions between deeper and lighter sleep within extended bouts of inactivity, with deeper sleep intensities after ∼15 and ∼30 min of inactivity. As in mammals, the timing and intensity of these dynamic sleep processes in flies is homeostatically regulated and modulated by behavioral experience. Two molecules linked to synaptic plasticity regulate the intensity of the first deep sleep stage. Optogenetic upregulation of cyclic adenosine monophosphate during the day increases sleep intensity at night, whereas loss of function of a molecule involved in synaptic pruning, the fragile-X mental retardation protein, increases sleep intensity during the day. Our results show that sleep is not homogenous in insects, and suggest that waking behavior and the associated synaptic plasticity mechanisms determine the timing and intensity of deep sleep stages in Drosophila.

TRIM proteins and the innate immune response to viruses.

Many TRIM proteins are up-regulated by interferons, suggesting that they might be involved in the innate immune response against viruses. Indeed, some members of the family have been shown to be either regulators of the interferon pathways or to be directly involved in the restriction of viruses. While the mechanisms of actions are varied, the modular organization of these proteins seems to be important for their activities, many of which are linked to the ubiquitination/proteasomal degradation system. The different domains enable the TRIM proteins to interact with either viral components or signaling molecules in the interferon induction pathways.

HIV-1 restriction factor SAMHD1 is a deoxynucleoside triphosphate triphosphohydrolase.

SAMHD1, an analogue of the murine interferon (IFN)-γ-induced gene Mg11 (ref. 1), has recently been identified as a human immunodeficiency virus-1 (HIV-1) restriction factor that blocks early-stage virus replication in dendritic and other myeloid cells and is the target of the lentiviral protein Vpx, which can relieve HIV-1 restriction. SAMHD1 is also associated with Aicardi-Goutières syndrome (AGS), an inflammatory encephalopathy characterized by chronic cerebrospinal fluid lymphocytosis and elevated levels of the antiviral cytokine IFN-α. The pathology associated with AGS resembles congenital viral infection, such as transplacentally acquired HIV. Here we show that human SAMHD1 is a potent dGTP-stimulated triphosphohydrolase that converts deoxynucleoside triphosphates to the constituent deoxynucleoside and inorganic triphosphate. The crystal structure of the catalytic core of SAMHD1 reveals that the protein is dimeric and indicates a molecular basis for dGTP stimulation of catalytic activity against dNTPs. We propose that SAMHD1, which is highly expressed in dendritic cells, restricts HIV-1 replication by hydrolysing the majority of cellular dNTPs, thus inhibiting reverse transcription and viral complementary DNA (cDNA) synthesis.

A systematic review of different models of home and community care services for older persons.

BACKGROUND: Costs and consumer preference have led to a shift from the long-term institutional care of aged older people to home and community based care. The aim of this review is to evaluate the outcomes of case managed, integrated or consumer directed home and community care services for older persons, including those with dementia. METHODS: A systematic review was conducted of non-medical home and community care services for frail older persons. MEDLINE, PsycINFO, CINAHL, AgeLine, Scopus and PubMed were searched from 1994 to May 2009. Two researchers independently reviewed search results. RESULTS: Thirty five papers were included in this review. Evidence from randomized controlled trials showed that case management improves function and appropriate use of medications, increases use of community services and reduces nursing home admission. Evidence, mostly from non-randomized trials, showed that integrated care increases service use; randomized trials reported that integrated care does not improve clinical outcomes. The lowest quality evidence was for consumer directed care which appears to increase satisfaction with care and community service use but has little effect on clinical outcomes. Studies were heterogeneous in methodology and results were not consistent. CONCLUSIONS: The outcomes of each model of care differ and correspond to the model's focus. Combining key elements of all three models may maximize outcomes.

Novel escape mutants suggest an extensive TRIM5α binding site spanning the entire outer surface of the murine leukemia virus capsid protein.

After entry into target cells, retroviruses encounter the host restriction factors such as Fv1 and TRIM5α. While it is clear that these factors target retrovirus capsid proteins (CA), recognition remains poorly defined in the absence of structural information. To better understand the binding interaction between TRIM5α and CA, we selected a panel of novel N-tropic murine leukaemia virus (N-MLV) escape mutants by a serial passage of replication competent N-MLV in rhesus macaque TRIM5α (rhTRIM5α)-positive cells using a small percentage of unrestricted cells to allow multiple rounds of virus replication. The newly identified mutations, many of which involve changes in charge, are distributed over the outer 'top' surface of N-MLV CA, including the N-terminal ß-hairpin, and map up to 29 A(o) apart. Biological characterisation with a number of restriction factors revealed that only one of the new mutations affects restriction by human TRIM5α, indicating significant differences in the binding interaction between N-MLV and the two TRIM5αs, whereas three of the mutations result in dual sensitivity to Fv1(n) and Fv1(b). Structural studies of two mutants show that no major changes in the overall CA conformation are associated with escape from restriction. We conclude that interactions involving much, if not all, of the surface of CA are vital for TRIM5α binding.