Flimma: a federated and privacy-aware tool for differential gene expression analysis

Aggregating transcriptomics data across hospitals can increase sensitivity and robustness of differential expression analyses, yielding deeper clinical insights. As data exchange is often restricted by privacy legislation, meta-analyses are frequently employed to pool local results. However, the accuracy might drop if class labels are inhomogeneously distributed among cohorts. Flimma (https://exbio.wzw.tum.de/flimma/) addresses this issue by implementing the state-of-the-art workflow limma voom in a federated manner, i.e., patient data never leaves its source site. Flimma results are identical to those generated by limma voom on aggregated datasets even in imbalanced scenarios where meta-analysis approaches fail.

Biased Data, Biased AI: Deep Networks Predict the Acquisition Site of TCGA Images

Deep learning models applied to healthcare applications including digital pathology have been increasing their scope and importance in recent years. Many of these models have been trained on The Cancer Genome Atlas (TCGA) atlas of digital images, or use it as a validation source. This study shows that there are tissue source site (tss) specific patterns of TCGA images that could be used to identify contributing institutions without any explicit training. Furthermore, it was observed that a model trained for cancer subtype classification has discovered such tss specific patterns within digital slides to classify cancer types. Digital scanner configuration and noise, tissue stain variation and artifacts, and source site patient demographics are among factors that likely account for the observed bias. Therefore, researchers should be cautious of such bias when using histopathology datasets for developing and training deep networks.

Examining Batch Effect in Histopathology as a Distributionally Robust Optimization Problem

Computer vision (CV) approaches applied to digital pathology have informed biological discovery and development of tools to help inform clinical decision-making. However, batch effects in the images represent a major challenge to effective analysis and interpretation of these data. The standard methods to circumvent learning such confounders include (i) application of image augmentation techniques and (ii) examination of the learning process by evaluating through external validation (e.g., unseen data coming from a comparable dataset collected at another hospital). Here, we show that the source site of a histopathology slide can be learned from the image using CV algorithms in spite of image augmentation, and we explore these source site predictions using interpretability tools. A CV model trained using Empirical Risk Minimization (ERM) risks learning this signal as a spurious correlate in the weak-label regime, which we abate by using a Distributionally Robust Optimization (DRO) method with abstention. We find that the model trained using DRO outperforms a model trained using ERM by 9.9, 13 and 15% in identifying tumor versus normal tissue in Lung Adenocarcinoma, Gleason score in Prostate Adenocarcinoma, and tumor tissue grade in clear cell renal cell carcinoma. Further, by examining the areas abstained by the model, we find that the model trained using a DRO method is more robust to heterogeneity and artifacts in the tissue. We believe that a DRO method trained with abstention may offer novel insights into relevant areas of the tissue contributing to a particular phenotype. Together, we suggest using data augmentation methods that help mitigate a digital pathology model's reliance on spurious visual features, as well as selecting models that are more robust to spurious features for translational discovery and clinical decision support.

Simulation of near-fault ground motions for randomized directivity parameters

The Dabaghi and Der Kiureghian stochastic near-fault ground motion model requires information about the source, site, and source-to-site geometry, including directivity parameters. Directivity parameters entail often unavailable knowledge of the rupture geometry and hypocenter location. This article presents methods to randomize the directivity parameters required to simulate near-fault ground motions. A first procedure is proposed where only the contributing fault, earthquake magnitude, and site location are known. Possible rupture directivity conditions are accounted for by randomizing the rupture geometry and hypocenter location. For this purpose, new predictive models of the rupture geometry parameters are developed for shallow crustal earthquakes with magnitudes between 5.2 and 7.9. To allow validation of synthetic motions with NGA-West2 models, a second procedure randomizes the rupture geometry and both hypocenter and site locations. Results show a general agreement between the two methods.

Non-topographic current contrast in scanning field emission microscopy

In scanning field emission microscopy (SFEM), a tip (the source) is approached to few (or a few tens of) nanometres distance from a surface (the collector) and biased to field-emit electrons. In a previous study (Zanin et al. 2016 Proc. R. Soc. A 472 , 20160475. ( doi:10.1098/rspa.2016.0475 )), the field-emitted current was found to change by approximately 1% at a monatomic surface step (approx. 200 pm thick). Here we prepare surface domains of adjacent different materials that, in some instances, have a topographic contrast smaller than 15 pm. Nevertheless, we observe a contrast in the field-emitted current as high as 10%. This non-topographic collector material dependence is a yet unexplored degree of freedom calling for a new understanding of the quantum mechanical tunnelling barrier at the source site that takes into account the properties of the material at the collector site.

Combining genotypic and phenotypic variation in a geospatial framework to identify sources of mussels in northern New Zealand

The New Zealand green-lipped mussel aquaculture industry is largely dependent on the supply of young mussels that wash up on Ninety Mile Beach (so-called Kaitaia spat), which are collected and trucked to aquaculture farms. The locations of source populations of Kaitaia spat are unknown and this lack of knowledge represents a major problem because spat supply may be irregular. We combined genotypic (microsatellite) and phenotypic (shell geochemistry) data in a geospatial framework to determine if this new approach can help identify source populations of mussels collected from two spat-collecting and four non-spat-collecting sites further south. Genetic analyses resolved differentiated clusters (mostly three clusters), but no obvious source populations. Shell geochemistry analyses resolved six differentiated clusters, as did the combined genotypic and phenotypic data. Analyses revealed high levels of spatial and temporal variability in the geochemistry signal. Whilst we have not been able to identify the source site(s) of Kaitaia spat our analyses indicate that geospatial testing using combined genotypic and phenotypic data is a powerful approach. Next steps should employ analyses of single nucleotide polymorphism markers with shell geochemistry and in conjunction with high resolution physical oceanographic modelling to resolve the longstanding question of the origin of Kaitaia spat.

An open-source site database of strong-motion stations in Japan: K-NET and KiK-net (v1.0.0)

This article describes an open-source site database for a total number of 1742 earthquake recording sites in the K-NET (Kyoshin network) and KiK-net (Kiban Kyoshin network) networks in Japan. This database contains site characterization parameters directly derived from available velocity profiles, including average wave velocities, bedrock depths, and velocity contrast. Meanwhile, it also consists of earthquake horizontal-to-vertical spectral ratio (HVSR) and peak parameters, for example, peak frequency, amplitude, width, and prominence. In addition, the site database also comprises topographic and geological proxies inferred from regional models or maps. Each parameter is derived in a consistent manner for all sites. This site database can benefit the application of machine learning techniques in studies on site amplification. Besides, it can facilitate, for instances, the search of the optimal site parameter(s) for the prediction of site amplification, the development and testing of ground-motion models or methodologies, as well as investigations on spatial or regional variability in site response. All resources (the site database, earthquake HVSR data at all sites, and the MATLAB script for peak identification) can be freely accessed via: https://doi.org/10.5880/GFZ.2.1.2020.006