Search

An effective aggregation of node features into a graph-level representation via readout functions is an essential step in numerous learning tasks involving graph neural networks. Typically, readouts are simple and non-adaptive functions designed such that the resulting hypothesis space is permutation invariant. Prior work on deep sets indicates that such readouts might require complex node embeddings that can be difficult to learn via standard neighborhood aggregation schemes. Motivated by this, we investigate the potential of adaptive readouts given by neural networks that do not necessarily give rise to permutation invariant hypothesis spaces. We argue that in some problems such as binding affinity prediction where molecules are typically presented in a canonical form it might be possible to relax the constraints on permutation invariance of the hypothesis space and learn a more effective model of the affinity by employing an adaptive readout function. Our empirical results demonstrate the effectiveness of neural readouts on more than 40 datasets spanning different domains and graph characteristics. Moreover, we observe a consistent improvement over standard readouts (i.e., sum, max, and mean) relative to the number of neighborhood aggregation iterations and different convolutional operators., Comment: Published at NeurIPS 2022. 10 pages, 5 figures, 1 table

Hannah R Whittaker,1 Steven J Kiddle,2 Jennifer K Quint1 1National Heart and Lung Institute, Imperial College London, London, UK; 2MRC Biostatistics Unit, University of Cambridge, Cambridge, UKCorrespondence: Hannah R Whittaker Email h.whittaker@imperial.ac.ukBackground: Electronic healthcare records (EHR) are increasingly used in epidemiological studies but are often viewed as lacking quality compared to randomised control trials and prospective cohorts. Studies of patients with chronic obstructive pulmonary disease (COPD) often use the rate of forced expiratory volume in 1 second (FEV1) decline as an outcome; however, its definition and robustness in EHR have not been investigated. We aimed to investigate how the rate of FEV1 decline differs by the criteria used in an EHR database.Methods: Clinical Practice Research Datalink and Hospital Episode Statistics were used. Patient populations were defined using 8 sets of criteria around repeated FEV1 measurements. At a minimum, patients had a diagnosis of COPD, were ≥ 35 years old, were current or ex-smokers, and had data recorded from 2004. FEV1 measurements recorded during follow-up were identified. Thereafter, eight populations were defined based on criteria around: i) the exclusion of patients or individual measurements with potential measurement error; ii) minimum number of FEV1 measurements; iii) minimum time interval between measurements; iv) specific timing of measurements; v) minimum follow-up time; and vi) the use of linked data. For each population, the rate of FEV1 decline was estimated using mixed linear regression.Results: For 7/8 patient populations, rates of FEV1 decline (age and sex adjusted) were similar and ranged from − 18.7mL/year (95% CI − 19.2 to − 18.2) to − 16.5mL/year (95% CI − 17.3 to − 15.7). Rates of FEV1 decline in populations that excluded patients with potential measurement error ranged from − 79.4mL/year (95% CI − 80.7 to − 78.2) to − 46.8mL/year (95% CI − 47.6 to

Hannah R Whittaker,1 Jeanne M Pimenta,2 Deborah Jarvis,1 Steven J Kiddle,3 Jennifer K Quint1 1Respiratory Epidemiology, Occupational Medicine and Public Health, National Heart and Lung Institute, Imperial College London, London, UK; 2Epidemiology (Value Evidence and Outcomes), GlaxoSmithKline, R&D, Uxbridge, UK; 3MRC Biostatistics Unit, University of Cambridge, Cambridge, UKCorrespondence: Hannah R WhittakerImperial College, Emmanuel Kaye Building, National Heart and Lung Institute, London, UKEmail h.whittaker@imperial.ac.ukBackground: Estimates for lung function decline in chronic obstructive pulmonary disease (COPD) have differed by study setting and have not been described in a UK primary care population.Purpose: To describe rates of FEV1 and FVC decline in COPD and investigate characteristics associated with accelerated decline.Patients and Methods: Current/ex-smoking COPD patients (35 years+) who had at least 2 FEV1 or FVC measurements ≥ 6 months apart were included using Clinical Practice Research Datalink. Patients were followed up for a maximum of 13 years. Accelerated rate of lung function decline was defined as the fastest quartile of decline using mixed linear regression, and association with baseline characteristics was investigated using logistic regression.Results: A total of 72,683 and 50,649 COPD patients had at least 2 FEV1 or FVC measurements, respectively. Median rates of FEV1 and FVC changes or decline were − 18.1mL/year (IQR: − 31.6 to − 6.0) and − 22.7mL/year (IQR: − 39.9 to − 6.7), respectively. Older age, high socioeconomic status, being underweight, high mMRC dyspnoea and frequent AECOPD or severe AECOPD were associated with an accelerated rate of FEV1 and FVC decline. Current smoking, mild airflow obstruction and inhaled corticosteroid treatment were additionally associated with accelerated FEV1 decline whilst women, sputum production and severe airflow obstruction w

Alzheimer's disease (AD)-a complex disease showing multiple pathomechanistic alterations-is triggered by nonlinear dynamic interactions of genetic/epigenetic and environmental risk factors, which, ultimately, converge into a biologically heterogeneous disease. To tackle the burden of AD during early preclinical stages, accessible blood-based biomarkers are currently being developed. Specifically, next-generation clinical trials are expected to integrate positive and negative predictive blood-based biomarkers into study designs to evaluate, at the individual level, target druggability and potential drug resistance mechanisms. In this scenario, systems biology holds promise to accelerate validation and qualification for clinical trial contexts of use-including proof-of-mechanism, patient selection, assessment of treatment efficacy and safety rates, and prognostic evaluation. Albeit in their infancy, systems biology-based approaches are poised to identify relevant AD "signatures" through multifactorial and interindividual variability, allowing us to decipher disease pathophysiology and etiology. Hopefully, innovative biomarker-drug codevelopment strategies will be the road ahead towards effective disease-modifying drugs. .

Alzheimer's disease (AD) is highly heritable and recent studies have identified over 20 disease-associated genomic loci. Yet these only explain a small proportion of the genetic variance, indicating that undiscovered loci remain. Here, we performed a large genome-wide association study of clinically diagnosed AD and AD-by-proxy (71,880 cases, 383,378 controls). AD-by-proxy, based on parental diagnoses, showed strong genetic correlation with AD (rg = 0.81). Meta-analysis identified 29 risk loci, implicating 215 potential causative genes. Associated genes are strongly expressed in immune-related tissues and cell types (spleen, liver, and microglia). Gene-set analyses indicate biological mechanisms involved in lipid-related processes and degradation of amyloid precursor proteins. We show strong genetic correlations with multiple health-related outcomes, and Mendelian randomization results suggest a protective effect of cognitive ability on AD risk. These results are a step forward in identifying the genetic factors that contribute to AD risk and add novel insights into the neurobiology of AD.

Hannah R Whittaker,1 Hana Müllerova,2 Deborah Jarvis,1 Neil C Barnes,2 Paul W Jones,2 Chris H Compton,2 Steven J Kiddle,3 Jennifer K Quint11National Heart and Lung Institute, Imperial College London, London, UK; 2Respiratory Epidemiology, GlaxoSmithKline R&D, Uxbridge, UK; 3MRC Biostatistics Unit, University of Cambridge, Cambridge, UKBackground: Inhaled corticosteroid (ICS)-containing medications slow rate of decline of FEV1. Blood eosinophil (EOS) levels are associated with the degree of exacerbation reduction with ICS.Purpose: We investigated whether FEV1 decline differs between patients with and without ICS, stratified by blood EOS level.Patients and methods: The UK Clinical Practice Research Datalink (primary care records) and Hospital Episode Statistics (hospital records) were used to identify COPD patients aged 35 years or older, who were current or ex-smokers with ≥2 FEV1 measurements ≥6 months apart. Prevalent ICS use and the nearest EOS count to start of follow-up were identified. Patients were classified at baseline as higher stratum EOS (≥150 cell/μL) on ICS; higher stratum EOS not on ICS; lower stratum EOS (<150 cells/μL) on ICS; and lower stratum EOS not on ICS. In addition, an incident ICS cohort was used to investigate the rate of FEV1 change by EOS and incident ICS use. Mixed-effects linear regression was used to compare rates of FEV1 change in mL/year.Results: A total of 26,675 COPD patients met our inclusion criteria (median age 69, 46% female). The median duration of follow up was 4.2 years. The rate of FEV1 change in prevalent ICS users was slower than non-ICS users (−12.6 mL/year vs −21.1 mL/year; P =0.001). The rate of FEV1 change was not significantly different when stratified by EOS level. The rate of FEV1 change in incident ICS users increased (+4.2 mL/year) vs −21.2 mL/year loss in non-ICS users; P<0.001. In patients with high EOS, inci

To elucidate the proteomic features of aging in plasma, the subproteome targeted by the SOMAscan assay was profiled in blood samples from 202 females from the TwinsUK cohort. Findings were replicated in 677 independent individuals from the AddNeuroMed, Alzheimer's Research UK, and Dementia Case Registry cohorts. Results were further validated using RNAseq data from whole blood in TwinsUK and the most significant proteins were tested for association with aging-related phenotypes after adjustment for age. Eleven proteins were associated with chronological age and were replicated at protein level in an independent population. These were further investigated at gene expression level in 384 females from the TwinsUK cohort. The two most strongly associated proteins were chordin-like protein 1 (meta-analysis β [SE] = 0.013 [0.001], p = 3.66 × 10−46) and pleiotrophin (0.012 [0.005], p = 3.88 × 10−41). Chordin-like protein 1 was also significantly correlated with birthweight (0.06 [0.02], p = 0.005) and with the individual Framingham 10-years cardiovascular risk scores in TwinsUK (0.71 [0.18], p = 9.9 × 10−5). Pleiotrophin is a secreted growth factor with a plethora of functions in multiple tissues and known to be a marker for cardiovascular risk and osteoporosis. Our study highlights the importance of proteomics to identify some molecular mechanisms involved in human health and aging