Stefanie Warnat‐Herresthal

Abstract Fast and reliable detection of patients with severe and heterogeneous illnesses is a major goal of precision medicine 1,2 . Patients with leukaemia can be identified using machine learning on the basis of their blood transcriptomes 3 . However, there is an increasing divide between what is technically possible and what is allowed, because of privacy legislation 4,5 . Here, to facilitate the integration of any medical data from any data owner worldwide without violating privacy laws, we introduce Swarm Learning—a decentralized machine-learning approach that unites edge computing, blockchain-based peer-to-peer networking and coordination while maintaining confidentiality without the need for a central coordinator, thereby going beyond federated learning. To illustrate the feasibility of using Swarm Learning to develop disease classifiers using distributed data, we chose four use cases of heterogeneous diseases (COVID-19, tuberculosis, leukaemia and lung pathologies). With more than 16,400 blood transcriptomes derived from 127 clinical studies with non-uniform distributions of cases and controls and substantial study biases, as well as more than 95,000 chest X-ray images, we show that Swarm Learning classifiers outperform those developed at individual sites. In addition, Swarm Learning completely fulfils local confidentiality regulations by design. We believe that this approach will notably accelerate the introduction of precision medicine.

Myeloid cells are a major cellular compartment of the immune system comprising monocytes, dendritic cells, tissue macrophages, and granulocytes. Models of cellular ontogeny, activation, differentiation, and tissue-specific functions of myeloid cells have been revisited during the last years with surprising results; for example, most tissue macrophages are yolk sac derived, monocytes and macrophages follow a multidimensional model of activation, and tissue signals have a significant impact on the functionality of all these cells. While these exciting results have brought these cells back to center stage, their enormous plasticity and heterogeneity, during both homeostasis and disease, are far from understood. At the same time, the ongoing revolution in single-cell genomics, with single-cell RNA sequencing (scRNA-seq) leading the way, promises to change this. Prevailing models of hematopoiesis with distinct intermediates are challenged by scRNA-seq data suggesting more continuous developmental trajectories in the myeloid cell compartment. Cell subset structures previously defined by protein marker expression need to be revised based on unbiased analyses of scRNA-seq data. Particularly in inflammatory conditions, myeloid cells exhibit substantially vaster heterogeneity than previously anticipated, and work performed within large international projects, such as the Human Cell Atlas, has already revealed novel tissue macrophage subsets. Based on these exciting developments, we propose the next steps to a full understanding of the myeloid cell compartment in health and diseases.

Acute myeloid leukemia (AML) is a severe, mostly fatal hematopoietic malignancy. We were interested in whether transcriptomic-based machine learning could predict AML status without requiring expert input. Using 12,029 samples from 105 different studies, we present a large-scale study of machine learning-based prediction of AML in which we address key questions relating to the combination of machine learning and transcriptomics and their practical use. We find data-driven, high-dimensional approaches-in which multivariate signatures are learned directly from genome-wide data with no prior knowledge-to be accurate and robust. Importantly, these approaches are highly scalable with low marginal cost, essentially matching human expert annotation in a near-automated workflow. Our results support the notion that transcriptomics combined with machine learning could be used as part of an integrated -omics approach wherein risk prediction, differential diagnosis, and subclassification of AML are achieved by genomics while diagnosis could be assisted by transcriptomic-based machine learning.

BACKGROUND AND AIMS: Treatment with tumor necrosis factor α (TNFα) antagonists in IBD patients suffers from primary non-response rates of up to 40%. Biomarkers for early prediction of therapy success are missing. We investigated the dynamics of gene expression and DNA methylation in blood samples of IBD patients treated with the TNF antagonist infliximab and analyzed the predictive potential regarding therapy outcome. METHODS: We performed a longitudinal, blood-based multi-omics study in two prospective IBD patient cohorts receiving first-time infliximab therapy (discovery: 14 patients, replication: 23 patients). Samples were collected at up to 7 time points (from baseline to 14 weeks after therapy induction). RNA-sequencing and genome-wide DNA methylation data were analyzed and correlated with clinical remission at week 14 as a primary endpoint. RESULTS: We found no consistent ex ante predictive signature across the two cohorts. Longitudinally upregulated transcripts in the non-remitter group comprised TH2- and eosinophil-related genes including ALOX15, FCER1A, and OLIG2. Network construction identified transcript modules that were coherently expressed at baseline and in non-remitting patients but were disrupted at early time points in remitting patients. These modules reflected processes such as interferon signaling, erythropoiesis, and platelet aggregation. DNA methylation analysis identified remission-specific temporal changes, which partially overlapped with transcriptomic signals. Machine learning approaches identified features from differentially expressed genes cis-linked to DNA methylation changes at week 2 as a robust predictor of therapy outcome at week 14, which was validated in a publicly available dataset of 20 infliximab-treated CD patients. CONCLUSIONS: Integrative multi-omics analysis reveals early shifts of gene expression and DNA methylation as predictors for efficient response to anti-TNF treatment. Lack of such signatures might be used to identify patients with IBD unlikely to benefit from TNF antagonists at an early time point.

Background Immune cells play a major role in the pathogenesis of COPD. Changes in the distribution and cellular functions of major immune cells, such as alveolar macrophages (AMs) and neutrophils are well known; however, their transcriptional reprogramming and contribution to the pathophysiology of COPD are still not fully understood. Method To determine changes in transcriptional reprogramming and lipid metabolism in the major immune cell type within bronchoalveolar lavage fluid, we analysed whole transcriptomes and lipidomes of sorted CD45 + Lin − HLA-DR + CD66b − Autofluorescence hi AMs from controls and COPD patients. Results We observed global transcriptional reprogramming featuring a spectrum of activation states, including pro- and anti-inflammatory signatures. We further detected significant changes between COPD patients and controls in genes involved in lipid metabolism, such as fatty acid biosynthesis in GOLD2 patients. Based on these findings, assessment of a total of 202 lipid species in sorted AMs revealed changes of cholesteryl esters, monoacylglycerols and phospholipids in a disease grade-dependent manner. Conclusions Transcriptome and lipidome profiling of COPD AMs revealed GOLD grade-dependent changes, such as in cholesterol metabolism and interferon-α and γ responses.