Imperial College London

Predicting high-dimensional dynamical systems with irregular time steps presents significant challenges for current data-driven algorithms. These irregularities arise from missing data, sparse observations, or adaptive computational techniques, reducing prediction accuracy. To address these limitations, we propose a novel method: a Physics-Spatiotemporal Masked Autoencoder. This method integrates convolutional autoencoders for spatial feature extraction with masked autoencoders optimised for irregular time series, leveraging attention mechanisms to reconstruct the entire physical sequence in a single prediction pass. The model avoids the need for data imputation while preserving the physical integrity of the system. Here, ’physics’ refers to high-dimensional fields generated by underlying dynamical systems, rather than the enforcement of explicit physical constraints or PDE residuals. We evaluate this approach on multiple simulated datasets and real-world ocean temperature data. The results demonstrate that our method achieves significant improvements in prediction accuracy, robustness to nonlinearities, and computational efficiency over traditional convolutional and recurrent network methods. The model shows potential for capturing complex spatiotemporal patterns without requiring domain-specific knowledge, with applications in climate modelling, fluid dynamics, ocean forecasting, environmental monitoring, and scientific computing.

Stress echocardiography (SE) detects myocardial ischemia while coronary computed tomography angiography (CCTA) detects structural abnormalities associated with coronary artery disease (CAD). A pilot retrospective study of 270 consecutive patients who underwent both SE and CCTA to explore which test is more accurate in patients with CAD. Health records and risk factors, SE, and CCTA outcomes were reviewed. SE outcomes were positive (n = 36) and negative (n = 234); CCTA outcomes were based on CAD severity (absent, mild, moderate, and severe). 219 (81.1%) patients had a positive CCTA, indicating the presence of CAD with mild, moderate, or severe stenosis. Despite a positive CCTA (n = 191), 70.7% had a negative SE. This discrepancy was most prevalent in the moderate CCTA category, where 110 patients had a negative SE but positive CCTA. Additionally, 9 patients demonstrated a positive SE despite no significant CCTA findings, indicating that functional abnormalities could have remained undetected by anatomical imaging. It is suggested that significant CAD detected by CCTA does not necessarily suggest myocardial ischemia, while the presence of myocardial ischemia is not necessarily associated with significant CAD, suggesting other causes (e.g., microvascular angina). A positive CCTA but negative SE findings may suggest the presence of coronary collaterals.

Setting up a psychedelic study can be a long, arduous, and Kafkaesque process. Researchers are faced with a host of challenges in this rapidly evolving field, necessitating a range of considerations that remain largely unstandardised. Many of the complexities inherent to psychedelic research also challenge existing assumptions around, for example, approaches to psychiatric prescribing, the conceptual framing of the placebo effect, and definitions of selfhood. This review paper aims to formalise these unique considerations by addressing the sociocultural, political, legal, pharmacological, safety, study design and experiential facets inherent to a psychedelic study. We bring together several of the major psychedelic research teams across the United Kingdom, identify continuing areas of debate, and provide a practical, comprehensive, experience-based guide, with recommendations for policymakers and future researchers intending to set up a psychedelic research study or clinical trial.

Structural health monitoring often involves temperature measurement. However, traditional sensors cannot measure subsurface temperature non-invasively, making them unsuitable for monitoring temperature-driven damage mechanisms such as high-cycle thermal fatigue. This limitation arises, in part, due to effective thermal low-pass filtering caused by material properties. A previous feasibility study demonstrated that subsurface temperature can be inferred non-invasively in mild steel subjected to uniform heating. This was achieved using the ultrasonic-based inverse thermal modelling (ITM) method, which assumes the temperature of a component can be described by a 1D system. This study investigated the behaviour of ITM under non-uniform heating applied to the 'inaccessible' surface of a stainless steel sample through experiments and simulations. The experimental results show that ITM over-predicts temperature by as much as 120% when the heated region is small compared with the 10mm ultrasonic beam size. In simulation, the overestimation was reduced as the size of the heating source increased, effectively making the temperature distribution more uniform across the volume through which the ultrasonic wave travels. Despite the overestimation under non-uniform heating, ITM overcomes the thermal low-pass filtering, allowing the detection of thermal transients compared with a thermocouple mounted on the 'accessible' surface of a component.

• Mutational scanning indicates more processing sites in K2 than initially recognized. • Insights into precursor processing help to predict the mature toxin structure. • Single mutations can enhance killer toxin function and change target specificity. • Yeast killer toxins are part of structural families despite low sequence identity. Yeast killer toxins (YKTs) are antimicrobial proteins secreted by yeast with potential applications ranging from food preservation to therapeutic agents in human health. However, the practical use of many YKTs is limited by specific pH requirements, low temperature stability, low production yields, and narrow target specificity. While protein engineering could potentially overcome these challenges, progress is hindered by a lack of detailed knowledge about sequence-function relationships and structural data for these often multi-step processed proteins. In this study, we focused on the YKT K2, encoded by the M2 satellite dsRNA in Saccharomyces cerevisiae . Using alanine scanning mutagenesis of the full open reading frame and structure predictions combined with molecular dynamics simulations, we generated a comprehensive sequence-structure–function map, refined the model for the proteolytic processing of the K2 precursor, and predicted the mature toxin structure. Our findings also demonstrate that K2 can be engineered toward enhanced toxicity and altered target specificity through single-site mutations. Furthermore, we identified structural homology between K2 and other killer toxins, including the SMK toxin from the yeast Millerozyma farinosa . Our cost-effective workflow provides a platform to broadly map YKT sequence-structure–function relationships.

Social anxiety disorder (SAD) frequently co-occurs with major depressive disorder (MDD). We examined the prevalence and clinical correlates of SAD across three heterogeneous MDD cohorts. Secondary analyses were conducted in adult (CO-MED: n=482; GSRD: n=1398) and late-life (IRL-GREY: n=438) patients. SAD was assessed dimensionally (PDSQ) and categorically (MINI; SCID-1). Cohort-specific instruments were used to assess depressive severity (QIDS; MADRS), suicidality (CHRT; MADRS; SIS) and hypomanic symptoms (ASRM; YMRS). In late-life depression, neurocognitive tests were administered. Multivariate models were adjusted for depression and anxiety levels. SAD prevalence varied markedly by samples and diagnostic definitions (CO-MED: 46.7% with PDSQ≥6; 17.0% with PDSQ>12; GSRD: 3.0%; IRL-GREY: 8.7%). Across cohorts, SAD was associated with earlier MDD onset (CO-MED: p<0.001 d=-0.36; GSRD: p=0.014 d=-0.39; IRL-GREY: p=0.002 d=-0.54) and greater anxiety comorbidity (CO-MED: GAD: p<0.001 d=1.24; panic: p<0.001 d=1.29; GSRD: GAD: p<0.001; panic: p<0.001; IRL-GREY: agoraphobia: p<0.001). After controlling for depression and anxiety levels, SAD was linked to higher suicide risk in both adult cohorts (CO-MED: OR 1.04 (1.01-1.08); GSRD: OR 1.29 (1.04-1.64)) and to lower 6-week remission in CO-MED cohort (OR 0.51 (0.29-0.91)). In MDD, SAD has consistent associations with earlier illness onset and anxiety burden, modest associations with suicidality and antidepressant outcomes, and reduced impact in late life.

Chronic hepatitis B (CHB) affects an estimated 254 million people globally. Historically, WHO and professional society guidelines have recommended treatment for individuals at high risk of disease progression, including those with hepatitis B virus (HBV) DNA concentrations >20 000 IU/mL. Whether individuals at lower risk, including those with HBV DNA <20 000 IU/mL and normal alanine aminotransferase concentrations, benefit from treatment remains uncertain. To inform 2024 WHO guidelines and the potential expansion of treatment threshold recommendations, we conducted two linked systematic reviews and meta-analyses; this analysis examines the incidence of clinical outcomes in untreated adults with non-cirrhotic CHB stratified by baseline HBV DNA and ALT concentrations. In this systematic review and meta-analysis, we searched PubMed, Embase, Web of Science, and the Cochrane Library for longitudinal prospective and retrospective cohort studies, randomised controlled trials, and non-randomised studies of interventions, published in any language, from Jan 1, 2000, to Feb 6, 2023. We also reviewed the reference lists of included studies and systematic reviews and consulted networks of experts to identify other data sources. Included studies followed up adults (≥18 years) with non-cirrhotic CHB who had no history of antiviral therapy at enrolment, had baseline HBV DNA quantification and ALT measurement, had a follow-up period of at least 48 weeks, and assessed one or more of six key clinical outcomes (hepatocellular carcinoma, cirrhosis, liver-related mortality, all-cause mortality, liver decompensation, and indication for liver transplantation) or six additional outcomes (advanced liver fibrosis, progression of liver fibrosis, becoming eligible for antiviral treatment, hepatitis flare, and HBsAg and HBeAg seroclearance). Outcomes were required to be stratified by HBV DNA concentrations (<200 IU/mL, <2000 IU/mL, 2000-19 999 IU/mL, 20 000-199 999 IU/mL, and ≥200 000 IU/mL) or ALT concentrations (less than the upper limit of normal [ULN], 1·0-1·9 × ULN, and ≥2·0 × ULN). We excluded studies that focused solely on individuals who were pregnant; were co-infected with HIV, hepatitis C virus, or hepatitis D virus; or had another underlying condition other than CHB. We extracted aggregate data and used random-effects meta-analysis to pool incidence rates per 100 person-years. This study was registered with PROSPERO (CRD42023431652). Of 13 231 studies screened, 71 met the inclusion criteria. A concentration-response relationship was observed between single baseline HBV DNA measurements and hepatocellular carcinoma incidence rates per 100 person-years: 0·131 (95% CI 0·097 to 0·177, I2=0%) for HBV DNA concentrations <200 IU/mL, 0·176 (0·117-0·266, I2=88%) for <2000 IU/mL, 0·311 (0·245-0·393, I2=13%) for 2000-19 999 IU/mL, 0·862 (0·725-1·026, I2=0%) for 20 000-199 999 IU/mL, and 0·941 (0·668-1·324, I2=58%) for ≥200 000 IU/mL (p<0·0001 for between-group difference). Similar concentration-response patterns were observed for development of cirrhosis (0·298 [95% CI 0·214-0·415], two studies, for <200 IU/mL; 0·300 [0·146-0·616], I2=88%, for <2000 IU/mL; 0·712 [0·624-0·814], I2=46%, for 2000-19 999 IU/mL; 1·436 [0·991-2·082], I2=76%, for 20 000-199 999 IU/mL; and 2·193 [1·690-2·846], I2=82%, for ≥200 000 IU/mL) and liver-related mortality (0·086 [95% CI 0·054-0·136], one study; 0·083 [0·015-0·451], I2=0%; 0·303 [0·228-0·402], I2=0%; 0·766 [0·591-0·993], one study; and 1·118 [0·951-1·314], two studies), but no concentration-response relationship was observed between a single baseline HBV DNA assessment and all-cause mortality or liver decompensation. Compared with ALT below the ULN, incidence rates of hepatocellular carcinoma, cirrhosis, and liver-related mortality were higher in individuals with baseline ALT concentrations 1·0-1·9 × ULN or ≥2·0 × ULN. A similar pattern was not found for all-cause mortality, and other outcomes were not meta-analysed due to small numbers. Within the HBV DNA strata of <2000 IU/mL, 2000-19 999 IU/mL, and ≥200 000 IU/mL, hepatocellular carcinoma incidence rates were 2-3 times higher in individuals with ALT 1·0-1·9 × ULN than in those with ALT concentrations less than the ULN. Individuals with persistently low HBV DNA concentrations (<2000 IU/mL) or persistently normal ALT concentrations across multiple assessments had slightly lower hepatocellular carcinoma incidence rates compared with those assessed by a single measurement. Our risk of bias assessment found 15 (21%) of 71 studies to be rated as good quality, 40 (56%) as fair quality, and 16 (23%) as poor quality. These findings, alongside the linked systematic review and meta-analysis of antiviral treatment efficacy, supported the 2024 WHO guidelines expansion of treatment criteria to include individuals with HBV DNA concentrations >2000 IU/mL and ALT concentrations above the ULN. For those with HBV DNA concentrations <2000 IU/mL and persistently normal ALT concentrations, treatment can be deferred. WHO.