Ekaterina Krymova

Lipids are fundamental components of the brain, crucial for synaptic transmission and signal propagation. Altered brain lipid composition is associated with common and rare neuropathologies, yet the spatial organization of the mammalian brain lipidome remains insufficiently characterized compared with other modalities ^1–6 . Here we mapped the membrane-lipid architecture of the adult mouse brain at micrometric scale, across sexes and during pregnancy. This lipid brain atlas reveals that lipids describe a fine-grained biochemical structure that aligns with functional anatomy. Membrane-lipid spatial heterogeneity clusters into territories, which we termed ‘lipizones’. Lipizones partially mirror cell-type territories, but also capture distal axon terminals. Through lipizones, we (1) reveal the organizing principles of the grey matter lipidome, related to connectivity and cytoarchitecture; (2) discover a new axis of oligodendrocyte heterogeneity in the white matter; and (3) find biochemical zonation in the choroid plexus and in the ventricular walls. We show that this lipidomic architecture can adapt to changing physiological needs. In the brain of pregnant female mice, the white matter is metabolically activated and the cortex undergoes a lipizone-specific remodelling that is particularly pronounced in layer 4. These results are a foundational resource ( https://lbae-v2.epfl.ch/ ) poised to reshape the understanding of lipids in brain development, physiology and pathology.

Structured interferences from substrates, matrices, and instrumental artifacts in mid-infrared (IR) spectra often overlap with analyte features and introduce errors in quantitative analysis. Existing correction methods rely on heuristic parameter selection and frequently perform suboptimally when applied across a diverse set of samples. We present a probabilistic framework, Variational-Inference-based Background Elimination in Spectroscopy (VIBES), for automated estimation of regularization hyperparameters from the data that enables removal of such interferences. Background contributions are modeled using covariance structures learned from analytical blank measurements, analyte signals as residual components from background removal under appropriate probabilistic assumptions that promote robustness and hyperparameters calibrated by maximizing a variational approximation to the model evidence. VIBES is demonstrated on two types of transmission-mode measurements with different types of background variations and analyte features, and shows capability to isolate the analyte absorption spectrum with high correlation (r > 0.95) between absorbance and mass loading across sharp and broad peak spectra. Interference-corrected spectra are furthermore shown to be consistent with absorption coefficients derived from complex refractive indices acquired from other measurement modalities. The possibility for quantitative extraction of analyte spectra in the presence of various interferences suggests a path toward development of universal calibrations for mid-IR spectroscopy.

Mineral desert dust is a major contributor to total atmospheric particulate matter ¹ . Desert dust outbreaks degrade air quality and can pose adverse health effects ² , including asthma exacerbation ³ and increased mortality ⁴ . At some European locations, there has been a rise in the intensity and frequency of transported dust outbreaks from deserts in recent decades ^5–9 . However, it remains unclear whether this increase is consistent across Europe and whether desertification and aridity or shifts in atmospheric circulation are the main drivers behind this rise. Here we compile a database of daily dust metal concentrations from European sites, establishing robust elemental ratios for transported dust. Using this database, we develop a machine learning model to estimate daily PM ₁₀ (particulate matter smaller than 10 μm) dust concentrations from 2012 to 2021, ranging from 2.09 ± 1.05 μg m ⁻³ across northern and central Europe to 5.28 ± 2.65 μg m ⁻³ across the south. In southern Europe, residents are exposed to transported dust events averaging 9.68 ± 4.85 μg m ⁻³ , linked to a 0.67 ± 0.02% rise in daily mortality. Intensified dust intrusions over the past decade are linked to shifts in atmospheric circulation. Data from an Alpine ice core record shows a 110% increase in dust concentrations since pre-industrial times, mostly associated with North African desertification. As climate change accelerates land degradation and affects weather patterns, worsening dust pollution may pose increasing risks to public health and air quality goals.

Enhancer-promoter interactions (EPIs) play a central role in gene regulation, but experimental techniques such as Hi-C for mapping these interactions remain costly and labor-intensive. Computational methods have been developed to predict EPIs in silico from DNA sequence and chromatin information; however, there are major challenges with the generalizability and accuracy of predictions by existing methods across cell types and conditions unseen during model training. We developed and validated UniversalEPI, an attention-based deep ensemble model that predicts EPIs up to 2 Mb apart using only DNA sequence and chromatin accessibility (ATAC-seq) data. Unlike models that reconstruct full Hi-C contact maps, UniversalEPI focuses on biologically relevant, sparse chromatin interactions between accessible regulatory elements. It generalizes across both bulk and single-cell ATAC-seq-derived pseudo-bulk datasets, delivering state-of-the-art performance while using fewer input modalities than existing approaches. By modeling predictive uncertainty, UniversalEPI enables statistically robust differential analysis of chromatin interactions across conditions. We demonstrate its utility by tracking dynamic EPIs during human macrophage activation and identifying regulatory differences between cancer cell states in esophageal adenocarcinoma. By providing precalculated Hi-C predictions for 157 ENCODE datasets, UniversalEPI expands the scope and applicability of in silico 3D genome modeling for studying gene regulation in development and disease.

Fine-grained prediction of chromatin accessibility from DNA sequence is a foundational step in modeling gene expression changes resulting from sequence variants. Yet, few methods operate at the resolution necessary to capture subtle effects of single-nucleotide changes. Furthermore, it remains unclear which architectural components, such as residual connections, normalization strategies, or attention mechanisms, drive performance in these high-resolution predictions. To address these knowledge gaps, we systematically evaluate classic architectural choices and introduce ConvNeXt V2 blocks, originally developed for computer vision, as high-resolution feature extractors in deep learning models for genomic data. Integrated into diverse architectures such as convoluted neural networks (CNNs), long short-term memory (LSTM), dilated CNNs, and transformers, ConvNeXt V2 blocks consistently improve performance, leading to similar prediction accuracy across these different model types. This reveals that early feature extraction, rather than downstream architecture, is the primary determinant of prediction accuracy. A comprehensive evaluation of these models on ATAC-seq signal prediction at 4-bp resolution in a cell type–specific manner identifies the ConvNeXt-based dilated CNN as the most robust performer, better preserving the signal’s shape. Our codebase and benchmarks provide practical tools for high-resolution chromatin modeling.

Fine particulate matter (PM) poses a major threat to public health, with organic aerosol (OA) being a key component. Major OA sources, hydrocarbon-like OA (HOA), biomass burning OA (BBOA), and oxygenated OA (OOA), have distinct health and environmental impacts. However, OA source apportionment via positive matrix factorization (PMF) applied to aerosol mass spectrometry (AMS) or aerosol chemical speciation monitoring (ACSM) data is costly and limited to a few supersites, leaving over 80% of OA data uncategorized in global monitoring networks. To address this gap, we trained machine learning models to predict HOA, BBOA, and OOA using limited OA source apportionment data and widely available organic carbon (OC) measurements across Europe (2010–2019). Our best performing model expanded the OA source data set 4-fold, yielding 85 000 daily apportionment values across 180 sites. Results show that HOA and BBOA peak in winter, particularly in urban areas, while OOA, consistently the dominant fraction, is more regionally distributed with less seasonal variability. This study provides a significantly expanded OA source data set, enabling better identification of pollution hotspots and supporting high-resolution exposure assessments.

Lipids are fundamental components of the brain, crucial for synaptic transmission and signal propagation. Altered brain lipid composition is associated with common and rare neuropathologies, yet, the spatial organization of the mammalian brain lipidome remains insufficiently characterized compared to other modalities. Here, we mapped the membrane lipid architecture of the adult mouse brain at micrometric scale, across sexes, and during pregnancy. This Lipid Brain Atlas reveals that lipids define a fine-grained biochemical structure that aligns with functional anatomy. Membrane lipid spatial heterogeneity clusters into territories, which we termed lipizones. Lipizones partially mirror cell type territories, but also capture distal axon terminals. Through lipizones, (i) we reveal the organizing principles of the gray matter lipidome, related to connectivity and cytoarchitecture; (ii) we discover a new axis of oligodendrocyte heterogeneity in the white matter; (iii) and we find biochemical zonation in the choroid plexus and in the ventricular walls. We show that this lipidomic architecture can adapt to changing physiological needs. In the brain of pregnant females, the white matter is metabolically activated and the outer cortex is reorganized. These results are a foundational resource (https://lbae-v2.epfl.ch/), poised to reshape our understanding of lipids in brain development, physiology, and pathology.

As the global shift towards renewable energy accelerates, achieving stability in power systems is crucial. Hydropower accounts for approximately 17% of energy produced worldwide, and with its capacity for active and reactive power regulation, is well-suited to provide necessary ancillary services. However, as demand for these services rises, hydropower systems must adapt to handle rapid dynamic changes and off-design conditions. Fatigue damage in hydraulic machines, driven by fluctuating loads and varying mechanical stresses, is especially prominent during the transient start-up of the machine. In this study, we introduce a data-driven approach to identify transient start-up trajectories that minimize fatigue damage. We optimize the trajectory by leveraging a machine learning model, trained on experimental stress data of reduced-scale model turbines. Numerical and experimental results confirm that our optimized trajectory significantly reduces start-up damage, representing a meaningful advancement in hydropower operations, maintenance, and the safe transition to higher operational flexibility.