C³ONTEXT: A Common Consensus on Convective OrgaNizaTion during the EUREC⁴A eXperimenT

The C3ONTEXT (A Common Consensus on Convective OrgaNizaTion during the EUREC4A eXperimenT) dataset is presented as an overview about the meso-scale cloud patterns identified during the EUREC4A field campaign in early 2020. Based on infrared and visible satellite images, 50 researchers of the EUREC4A science team manually identified the four prevailing meso-scale patterns of shallow convection observed by Stevens et al. (2020). The common consensus on the observed meso-scale cloud patterns emerging from these manual classifications is presented. It builds the basis for future studies and reduces the subjective nature of these visually defined cloud patterns. This consensus makes it possible to contextualize the measurements of the EUREC4A field campaign and interpret them in the meso-scale setting. Commonly used approaches to capture the meso-scale patterns are computed for comparison and show good agreement with the manual classifications. All four patterns as classified by Stevens et al. (2020) were present in January–February 2020 although not all were dominant during the observing period of EUREC4A. The full dataset including postprocessed datasets for easier usage are openly available at the Zenodo archive at https://doi. org/10.5281/zenodo.5724585 (Schulz, 2021b).

Phenomics-Assisted Selection for Herbage Accumulation in Alfalfa (Medicago sativa L.)

The application of remote sensing in plant breeding is becoming a routine method for fast and non-destructive high-throughput phenotyping (HTP) using unmanned aerial vehicles (UAVs) equipped with sensors. Alfalfa (Medicago sativa L.) is a perennial forage legume grown in more than 30 million hectares worldwide. Breeding alfalfa for herbage accumulation (HA) requires frequent and multiple phenotyping efforts, which is laborious and costly. The objective of this study was to assess the efficiency of UAV-based imagery and spatial analysis in the selection of alfalfa for HA. The alfalfa breeding population was composed of 145 full-sib and 34 half-sib families, and the experimental design was a row-column with augmented representation of controls. The experiment was established in November 2017, and HA was harvested four times between August 2018 and January 2019. A UAV equipped with a multispectral camera was used for HTP before each harvest. Four vegetation indices (VIs) were calculated from the UAV-based images: NDVI, NDRE, GNDVI, and GRVI. All VIs showed a high correlation with HA, and VIs predicted HA with moderate accuracy. HA and NDVI were used for further analyses to calculate the genetic parameters using linear mixed models. The spatial analysis had a significant effect in both dimensions (rows and columns) for HA and NDVI, resulting in improvements in the estimation of genetic parameters. Univariate models for NDVI and HA, and bivariate models, were fit to predict family performance for scenarios with various levels of HA data (simulated in silico by assigning missing values to full dataset). The bivariate models provided higher correlation among predicted values, higher coincidence for selection, and higher genetic gain even for scenarios with only 30% of HA data. Hence, HTP is a reliable and efficient method to aid alfalfa phenotyping to improve HA. Additionally, the use of spatial analysis can also improve the accuracy of selection in breeding trials.

Detailed patient-individual reporting of lymph node involvement in oropharyngeal squamous cell carcinoma with an online interface

Purpose/ObjectiveWhereas the prevalence of lymph node level (LNL) involvement in head & neck squamous cell carcinomas (HNSCC) has been reported, the details of lymphatic progression patterns are insufficiently quantified. In this study, we investigate how the risk of metastases in each LNL depends on the involvement of upstream LNLs, T-category, HPV status and other risk factors.Materials/MethodsWe retrospectively analyzed patients with newly diagnosed oropharyngeal HNSCC treated at a single institution, resulting in a dataset of 287 patients. For all patients, involvement of LNLs I-VII was recorded individually based on available diagnostic modalities (PET, MR, CT, FNA) together with clinicpathological factors. To analyze the dataset, a web-based graphical user interface (GUI) was developed, which allows querying the number of patients with a certain combination of co-involved LNLs and tumor characteristics.ResultsThe full dataset and GUI is part of the publication. Selected findings are: Ipsilateral level IV was involved in 27% of patients with level II and III involvement, but only in 2% of patients with level II but not III involvement. Prevalence of involvement of ipsilateral levels II, III, IV, V was 79%, 34%, 7%, 3% for early T-category patients (T1/T2) and 85%, 50%, 17%, 9% for late T-category (T3/T4), quantifying increasing involvement with T-category. Contralateral levels II, III, IV were involved in 41%, 19%, 4% and 12%, 3%, 2% for tumors for tumors with and without midline extension, respectively. T-stage dependence of LNL involvement was more pronounced in HPV negative than positive tumors, but overall involvement was similar. Ipsilateral level VII was involved in 14% and 6% of patients with primary tumors in the tonsil and the base of tongue, respectively.ConclusionsDetailed quantification of LNL involvement in HNSCC depending on involvement of upstream LNLs and clinicopathological factors may allow for further personalization of CTV-N definition in the future.

JLAN: medical code prediction via joint learning attention networks and denoising mechanism

Background Clinical notes are documents that contain detailed information about the health status of patients. Medical codes generally accompany them. However, the manual diagnosis is costly and error-prone. Moreover, large datasets in clinical diagnosis are susceptible to noise labels because of erroneous manual annotation. Therefore, machine learning has been utilized to perform automatic diagnoses. Previous state-of-the-art (SOTA) models used convolutional neural networks to build document representations for predicting medical codes. However, the clinical notes are usually long-tailed. Moreover, most models fail to deal with the noise during code allocation. Therefore, denoising mechanism and long-tailed classification are the keys to automated coding at scale. Results In this paper, a new joint learning model is proposed to extend our attention model for predicting medical codes from clinical notes. On the MIMIC-III-50 dataset, our model outperforms all the baselines and SOTA models in all quantitative metrics. On the MIMIC-III-full dataset, our model outperforms in the macro-F1, micro-F1, macro-AUC, and precision at eight compared to the most advanced models. In addition, after introducing the denoising mechanism, the convergence speed of the model becomes faster, and the loss of the model is reduced overall. Conclusions The innovations of our model are threefold: firstly, the code-specific representation can be identified by adopted the self-attention mechanism and the label attention mechanism. Secondly, the performance of the long-tailed distributions can be boosted by introducing the joint learning mechanism. Thirdly, the denoising mechanism is suitable for reducing the noise effects in medical code prediction. Finally, we evaluate the effectiveness of our model on the widely-used MIMIC-III datasets and achieve new SOTA results.

Adaptive market hypothesis: An empirical analysis of the Wine Market

We test the nature of weak form informational efficiency present in the wine market using daily return of LIV-EX 50 index from 1/1/2010 to 12/6/2020. First, we employ a number of statistical tests including variance ratio tests, tests for linear and non-linear dependence and Hurst coefficient. The tests are applied on the full dataset and on four non overlapping sub-samples of equal length. The variance ratio tests provide a mixed regarding informational efficiency. Evidence of non-linear dependence in the return series was found. The Hurst coefficient values confirm the presence of long run persistence in the wine market. Based on the mixed evidence, we test the possibility of adaptive nature of the wine market. We employ the newly proposed Adaptive Index (AI) to quantify the degree of information inefficiency in the wine market at any instance. Our results confirm that wine market is adaptive and periodically shifts between states of efficiency and inefficiency. The wine market is found to be relatively free from the Covid-19 induced shock and the safe haven property of wine is thus confirmed. Finally, impact of various macroeconomic and financial events on wine market efficiency is identified by using AI.

Using Machine Learning to estimate the technical potential of shallow ground-source heat pumps with thermal interference

The increasing use of ground-source heat pumps (GSHPs) for heating and cooling of buildings raises questions regarding the technical potential of GSHPs and their impact on the temperature in the shallow subsurface. In this paper, we develop a method using Machine Learning to estimate the technical potential of shallow GSHPs, which enables such an estimation for Switzerland with limited data and computational resources. A training dataset is constructed based on meteorological and geological data across Switzerland. We analyse correlations and the importance of each of the input data for estimating the GSHP potential and compare different input feature sets and Machine Learning models. The Random Forest algorithm, trained on the full dataset, provides the best performance to estimate the GSHP potential. The resulting model yields an R2 score of 0.95 for the annual energy potential, 0.86 for the heat extraction rate, and 0.82 for the potential number of boreholes per GSHP system.

43 Single and Multiple-breed Genomic Predictions for Conformation Traits of Canadian Dairy Goats

The use of multiple-breed models can increase the accuracy of estimated breeding values (EBV) when few phenotypes are available for a trait. However, pooling breeds is not always beneficial for genomic evaluations due to the low consistency of gametic phase between individual breeds. The objective of this study was to compare the expected gain in accuracy of single-step genomic breeding values (GEBV) for conformation traits of Canadian Alpine and Saanen goats predicted using single and multiple-breed models. The traits considered were body capacity, dairy character, feet and legs, fore udder, general appearance, rear udder, suspensory ligament, and teats, all recorded by trained classifiers, using a 1 to 9 scale. The full datasets included a total of 7,500 phenotypes for each trait (5,158 Alpine and 2,342 Saanen) and 1,707 50K genotypes (833 Alpine, 874 Saanen). Standard errors of prediction (SEP) were obtained for EBV and GEBV predicted using single-trait animal models on full or validation datasets. Breed difference was accounted for as a fixed effect in the multiple-breed models. Average theoretical accuracies were calculated from the SEP. For Saanen, with fewer records, expected accuracies of EBV and GEBV for the validation animals (selection candidates) were consistently higher for the multiple-breed models. Trait specific gains in theoretical accuracy of GEBV relative to EBV for the selection candidates ranged from 30 to 48% for Alpine and 41 to 61% for Saanen. Averaged across all traits, GEBV predicted from the full dataset were 32 to 38% more accurate than EBV for genotyped animals and the largest gains were found for does without conformation records (49 to 55%) and bucks without daughter records (56 to 82%). Overall, the implementation of genomic selection would substantially increase selection accuracy for young breeding candidates and, consequently, the rate of genetic improvement for conformation traits in Canadian dairy goats.

Characterization of non-refractory (NR) PM₁ and source apportionment of organic aerosol in Kraków, Poland

Kraków is routinely affected by very high air pollution levels, especially during the winter months. Although a lot of effort has been made to characterize ambient aerosol, there is a lack of online and long-term measurements of non-refractory aerosol. Our measurements at the AGH University of Science and Technology provide the online long-term chemical composition of ambient submicron particulate matter (PM1) between January 2018 and April 2019. Here we report the chemical characterization of non-refractory submicron aerosol and source apportionment of the organic fraction by positive matrix factorization (PMF). In contrast to other long-term source apportionment studies, we let a small PMF window roll over the dataset instead of performing PMF over the full dataset or on separate seasons. In this way, the seasonal variation in the source profiles can be captured. The uncertainties in the PMF solutions are addressed by the bootstrap resampling strategy and the random a-value approach for constrained factors. We observe clear seasonal patterns in the concentration and composition of PM1, with high concentrations during the winter months and lower concentrations during the summer months. Organics are the dominant species throughout the campaign. Five organic aerosol (OA) factors are resolved, of which three are of a primary nature (hydrocarbon-like OA (HOA), biomass burning OA (BBOA) and coal combustion OA (CCOA)) and two are of a secondary nature (more oxidized oxygenated OA (MO-OOA) and less oxidized oxygenated OA (LO-OOA)). While HOA contributes on average 8.6 % ± 2.3 % throughout the campaign, the solid-fuel-combustion-related BBOA and CCOA show a clear seasonal trend with average contributions of 10.4 % ± 2.7 % and 14.1 %, ±2.1 %, respectively. Not only BBOA but also CCOA is associated with residential heating because of the pronounced yearly cycle where the highest contributions are observed during wintertime. Throughout the campaign, the OOA can be separated into MO-OOA and LO-OOA with average contributions of 38.4 % ± 8.4 % and 28.5 % ± 11.2 %, respectively.

Estimation of accuracy loss by training a deep-learning-based cell organelle recognition software using full dataset and a reduced dataset containing only subviral particle distribution information

In collaboration with the Institute of Virology, Philipps University, Marburg, a deep-learning-based method that recognizes and classifies cell organelles based on the distribution of subviral particles in fluorescence microscopy images of virus-infected cells has been further developed. In this work a method to recognize cell organelles by means of partial image information is extended. The focus is on investigating loss of accuracy by only providing information about subviral particles and not all cell organelles to an adopted Mask-R convolutional neural network. Our results show that the subviral particle distribution holds information about the cell morphology, thus making it possible to use it for cell organelle-labelling.