Search bioRxiv⌕ Search

Biology subjects

Wählby, C.

Publications and source records attributed to Wählby, C..

12 recordsLinked to original sources
bioRxiv2026

Label-free identification of bacterial species using adapted vision transformers

Antibiotic resistance is an increasingly serious problem worldwide. Identifying the bacterial species can help guide antibiotic selection and improve patient outcomes. In sepsis, where every minute counts, rapid identification is particularly important, and new approaches are needed. In this study, deep learning models were trained to identify bacterial species from phase-contrast time-lapse images of cells growing in microfluidic traps, without fluorescent labels. We demonstrated this approach in two previous studies using laboratory strains. Here, the previous work is extended to clinical bacterial isolates, evaluating performance on isolates not seen during training. Furthermore, it was tested whether autofluorescence could aid species identification. Two validation sets were acquired in separate experiments from the training data, the first containing clinical-isolate traps with five species (Escherichia coli, Enterococcus faecalis, Klebsiella pneumoniae, Staphylococcus aureus and Enterobacter cloacae), imaged every two minutes, and the second containing clinical isolates of two species (E. coli and K. pneumoniae) recovered from spiked donor blood, imaged every minute. Training with the best previous CNN method achieved unsatisfactory results. Therefore, vision transformers were explored, and a self-supervised adaptation procedure was developed for pretrained DINOv3 using a frozen teacher, followed by supervised classification using time-lapse montages. Mean macro-F1 was 93.6 {+/-} 1.3% for isolates precultured in Mueller-Hinton medium using 60 frames (118 minutes) and 87.2 {+/-} 2.7% for isolates precultured in blood using 55 frames (54 minutes) (mean {+/-} sample standard deviation across training seeds). Performance was evaluated over time by increasing the number of frames available to the model during testing, and Fourier-domain filtering approximated reduced numerical aperture. The results are promising and suggest that the aperture can be decreased while retaining good results. However, further evaluation on independent isolates, experiments and patient samples is needed to establish generalization. Together with previous findings, these results bring us closer to rapid image-based diagnostics that could transform the initial treatment of acute bacterial infections by enabling better-informed antibiotic selection.

microbiology↗
bioRxiv2026

SFUMATO: Bayesian probabilistic clustering for uncertainty-aware spatialtranscriptomics analysis and mapping

In situ sequencing methods provide subcellular-resolution gene expression data while preserving spatial context, enabling the integrated analysis of genomics and tissue morphology. However, many clustering workflows represent spatial transcriptomic structure through hard labels, imposing sharp boundaries between transcriptional domains and limiting the visualisation of gradual transitions, mixed signals, and uncertainty in cluster assignment, resulting in potentially misleading interpretations of tissue organisation. Here we present SFUMATO (Segmentation-Free Uncertainty Mapping and Analysis of Transcriptomic Organisation), a GPU-accelerated Bayesian probabilistic clustering and visualisation framework for spatial transcriptomics. SFUMATO builds on segmentation-free, multi-scale transcript binning and applies a Bayesian Gaussian mixture model to infer posterior probabilities over transcriptional components at each spatial location. These posterior probabilities provide a unified representation for hierarchical clustering, uncertainty-aware visualisation, and quantitative estimation of transcript-defined area fractions. SFUMATO assigns related colours to transcriptionally related components and blends colours according to posterior probabilities, producing maps in which sharp boundaries, gradual transitions, and ambiguous regions are represented directly in the visualisation. Classical hard labels remain recoverable from the same posterior distributions, preserving compatibility with downstream analyses that require discrete clusters. We evaluate SFUMATO on public 10x Genomics Xenium mouse brain and human breast cancer datasets. In mouse brain, SFUMATO preserves local transcriptional similarity in its colour encoding, maintains hierarchically consistent visualisations across different numbers of clusters, and recovers cell-associated spatial organisation without requiring prior cell segmentation. In breast cancer, SFUMATO-derived posterior maps can be converted into semantic masks concordant with pathology-associated regions, particularly invasive carcinoma, while highlighting the greater heterogeneity of DCIS-associated tissue. Together, these results show that SFUMATO provides an interpretable probabilistic representation of spatial transcriptomics that connects clustering, visualisation, and semantic segmentation across cell-associated and niche-level scales. More broadly, SFUMATO also provides a compact and continuous representation of highly multiplexed spatial transcriptomic information, offering a flexible input for future downstream analyses and multimodal integration.

bioinformatics↗
bioRxiv2024

Rapid identification of seven bacterial species using microfluidics, time-lapse phase-contrast microscopy, and deep learning

For effective treatment of bacterial infections, it is essential to identify the species causing the infection as early as possible. Current methods typically require hours of overnight culturing of a bacterial sample and a larger quantity of cells to function effectively. This study uses one-hour phase-contrast time-lapses of single-cell bacterial growth collected from microfluidic chip traps, also known as a "mother machine". These time-lapses are then used to train deep artificial neural networks (Convolutional Neural Networks and Vision Transformers) to identify the species. We have previously demonstrated this approach on four different species, which is now extended to seven common pathogens causing human infections: Pseudomonas aeruginosa, Escherichia coli, Klebsiella pneumoniae, Acinetobacter baumannii, Enterococcus faecalis, Proteus mirabilis, and Staphylococcus aureus. Furthermore, we expand upon our previous work by evaluating real-time performance as additional frames are captured during testing, and investigating the role of training set size, data quality, and data augmentation as well as the contribution of texture and morphology to performance. The experiments suggest that spatiotemporal features can be learned from video data of bacterial cell divisions, with both texture and morphology contributing to classifier decision. The method could be used simultaneously with phenotypic antibiotic susceptibility testing (AST) in the microfluidic chip. The best models attained an average precision of 93.5% and a recall of 94.7% (0.997 AUC) on a trap basis in a separate, unseen experiment with mixed species after around one hour. However, in a real-world scenario, one can assume many traps will contain the actual species causing the infection. Still, several challenges remain, such as isolating bacteria directly from blood and validating the method on diverse clinical isolates. This proof of principle study brings us closer to real-time diagnostics that could transform the initial treatment of acute infections.

bioinformatics↗
bioRxiv2023

Label-free deep learning-based species classification of bacteria imaged by phase-contrast microscopy

Reliable detection and classification of bacteria and other pathogens in the human body, animals, food, and water is crucial for improving and safeguarding public health. For instance, identifying the species and its antibiotic susceptibility is vital for effective bacterial infection treatment. Here we show that phase contrast time-lapse microscopy combined with deep learning is sufficient to discriminate four species of bacteria relevant to human health. The classification is performed on living bacteria and does not require fixation or staining, meaning that the bacterial species can be determined as the bacteria reproduce in a microfluidic device, enabling parallel determination of susceptibility to antibiotics. We assess the performance of convolutional neural networks and vision transformers, where the best model attained a class-average accuracy exceeding 98%. Our successful proof-of-principle results suggest that the methods should be challenged with data covering more species and clinically relevant isolates for future clinical use. Author SummaryBacterial infections are a leading cause of premature death worldwide, and growing antibiotic resistance is making treatment increasingly challenging. To effectively treat a patient with a bacterial infection, it is essential to quickly detect and identify the bacterial species and determine its susceptibility to different antibiotics. Prompt and effective treatment is crucial for the patients survival. A microfluidic device functions as a miniature "lab-on-chip" for manipulating and analyzing tiny amounts of fluids, such as blood or urine samples from patients. Microfluidic chips with chambers and channels have been designed for quickly testing bacterial susceptibility to different antibiotics by analyzing bacterial growth. Identifying bacterial species has previously relied on killing the bacteria and applying species-specific fluorescent probes. We introduce deep learning models as a fast and cost-effective method for identifying bacteria species directly from phase-contrast microscopy images of living bacteria simultaneously as growth is analyzed. We envision this method being employed concurrently with antibiotic susceptibility tests in future applications, significantly enhancing bacterial infection treatments.

bioinformatics↗
bioRxiv2022

Points2regions: Fast Interactive Clustering Of In Situ Transcriptomics Data

Imaging-based spatial transcriptomics techniques generate image data that, once processed, results in a set of spatial points with categorical labels for different mRNA species. A crucial part of analyzing downstream data involves the analysis of these point patterns. Here, biologically interesting patterns can be explored at different spatial scales. Molecular patterns on a cellular level would correspond to cell types, whereas patterns on a millimeter scale would correspond to tissue-level structures. Often, clustering methods are employed to identify and segment regions with distinct point-patterns. Traditional clustering techniques for such data are constrained by reliance on complementary data or extensive machine learning, limiting their applicability to tasks on a particular scale. This paper introduces Points2Regions, a practical tool for clustering spatial points with categorical labels. Its flexible and computationally efficient clustering approach enables pattern discovery across multiple scales, making it a powerful tool for exploratory analysis. Points2Regions has demonstrated efficient performance in various datasets, adeptly defining biologically relevant regions similar to those found by scale-specific methods. As a Python package integrated into TissUUmaps and a Napari plugin, it offers interactive clustering and visualization, significantly enhancing user experience in data exploration. In essence, Points2Regions presents a user-friendly and simple tool for exploratory analysis of spatial points with categorical labels.

bioinformatics↗
bioRxiv2022

TissUUmaps 3: Interactive visualization and quality assessment of large-scale spatial omics data

Background and ObjectivesSpatially resolved techniques for exploring the molecular landscape of tissue samples, such as spatial transcriptomics, often result in millions of data points and images too large to view on a regular desktop computer, limiting the possibilities in visual interactive data exploration. TissUUmaps is a free, open-source browser-based tool for GPU-accelerated visualization and interactive exploration of 107+ data points overlaying tissue samples. MethodsHerein we describe how TissUUmaps 3 provides instant multiresolution image viewing and can be customized, shared, and also integrated into Jupyter Notebooks. We introduce new modules where users can visualize markers and regions, explore spatial statistics, perform quantitative analyses of tissue morphology, and assess the quality of decoding in situ transcriptomics data. ResultsWe show that thanks to targeted optimizations the time and cost associated with interactive data exploration were reduced, enabling TissUUmaps 3 to handle the scale of todays spatial transcriptomics methods. ConclusionTissUUmaps 3 provides significantly improved performance for large multiplex datasets as compared to previous versions. We envision TissUUmaps to contribute to broader dissemination and flexible sharing of large-scale spatial omics data.

bioinformatics↗
bioRxiv2022

Developmental origins of cell heterogeneity in the human lung

The lung contains numerous specialized cell-types with distinct roles in tissue function and integrity. To clarify the origins and mechanisms generating cell heterogeneity, we created a first comprehensive topographic atlas of early human lung development. We report 83 cell states, several spatially-resolved developmental trajectories and predict cell interactions within defined tissue niches. We integrated scRNA-Seq and spatial transcriptomics into a web-based, open platform for interactive exploration. To illustrate the utility of our approach we show distinct states of secretory and neuroendocrine cells, largely overlapping with the programs activated either during lung fibrosis or small cell lung cancer progression. We define the origin of uncharacterized airway fibroblasts associated with airway smooth muscle in bronchovascular bundles, and describe a trajectory of Schwann cell progenitors to intrinsic parasympathetic neurons controlling bronchoconstriction. Our atlas provides a rich resource for further research and a reference for defining deviations from homeostatic and repair mechanisms leading to pulmonary diseases.

developmental biology↗
bioRxiv2021

ISTDECO: In Situ Transcriptomics Decoding by Deconvolution

In Situ Transcriptomics (IST) is a set of image-based transcriptomics approaches that enables localisation of gene expression directly in tissue samples. IST techniques produce multiplexed image series in which fluorescent spots are either present or absent across imaging rounds and colour channels. A spots presence and absence form a type of barcoded pattern that labels a particular type of mRNA. Therefore, the expression of a gene can be determined by localising the fluorescent spots and decode the barcode that they form. Existing IST algorithms usually do this in two separate steps: spot localisation and barcode decoding. Although these algorithms are efficient, they are limited by strictly separating the localisation and decoding steps. This limitation becomes apparent in regions with low signal-to-noise ratio or high spot densities. We argue that an improved gene expression decoding can be obtained by combining these two steps into a single algorithm. This allows for an efficient decoding that is less sensitive to noise and optical crowding. We present IST Decoding by Deconvolution (ISTDECO), a principled decoding approach combining spectral and spatial deconvolution into a single algorithm. We evaluate ISTDECO on simulated data, as well as on two real IST datasets, and compare with state-of-the-art. ISTDECO achieves state-of-the-art performance despite high spot densities and low signal-to-noise ratios. It is easily implemented and runs efficiently using a GPU. ISTDECO implementation, datasets and demos are available online at: github.com/axanderssonuu/istdeco

bioinformatics↗
bioRxiv2021

Machine learning for cell classification and neighborhood analysis in glioma tissue

Multiplexed and spatially resolved single-cell analyses that intend to study tissue heterogeneity and cell organization invariably face as a first step the challenge of cell classification. Accuracy and reproducibility are important for the down-stream process of counting cells, quantifying cell-cell interactions, and extracting information on disease-specific localized cell niches. Novel staining techniques make it possible to visualize and quantify large numbers of cell-specific molecular markers in parallel. However, due to variations in sample handling and artefacts from staining and scanning, cells of the same type may present different marker profiles both within and across samples. We address multiplexed immunofluorescence data from tissue microarrays of low grade gliomas and present a methodology using two different machine learning architectures and features insensitive to illumination to perform cell classification. The fully automated cell classification provides a measure of confidence for the decision and requires a comparably small annotated dataset for training, which can be created using freely available tools. Using the proposed method, we reached an accuracy of 83.1% on cell classification without the need for standardization of samples. Using our confidence measure, cells with low-confidence classifications could be excluded, pushing the classification accuracy to 94.5%. Next, we used the cell classification results to search for cell niches with an unsupervised learning approach based on graph neural networks. We show that the approach can re-detect specialized tissue niches in previously published data, and that our proposed cell classification leads to niche definitions that may be relevant for sub-groups of glioma, if applied to larger datasets.

bioinformatics↗
bioRxiv2020

Rapid development of cloud-native intelligent data pipelines for scientific data streams using the HASTE Toolkit

This paper introduces the HASTE Toolkit, a cloud-native software toolkit capable of partitioning data streams in order to prioritize usage of limited resources. This in turn enables more efficient data-intensive experiments. We propose a model that introduces automated, autonomous decision making in data pipelines, such that a stream of data can be partitioned into a tiered or ordered data hierarchy. Importantly, the partitioning is online and based on data content rather than a priori metadata. At the core of the model are interestingness functions and policies. Interestingness functions assign a quantitative measure of interestingness to a single data object in the stream, an interestingness score. Based on this score, a policy guides decisions on how to prioritize computational resource usage for a given object. The HASTE Toolkit is a collection of tools to adapt data stream processing to this pipeline model. The result is smart data pipelines capable of effective or even optimal use of e.g. storage, compute and network bandwidth, to support experiments involving rapid processing of scientific data characterized by large individual data object sizes. We demonstrate the proposed model and our toolkit through two microscopy imaging case studies, each with their own interestingness functions, policies, and data hierarchies. The first deals with a high content screening experiment, where images are analyzed in an on-premise container cloud with the goal of prioritizing the images for storage and subsequent computation. The second considers edge processing of images for upload into the public cloud for a real-time control loop for a transmission electron microscope. Key PointsO_LIWe propose a pipeline model for building intelligent pipelines for streams, accounting for actual information content in data rather than a priori metadata, and present the HASTE Toolkit, a cloud-native software toolkit for supporting rapid development according to the proposed model. C_LIO_LIWe demonstrate how the HASTE Toolkit enables intelligent resource optimization in two image analysis case studies based on a) high-content imaging and b) transmission electron microscopy. C_LIO_LIWe highlight the challenges of storage, processing and transfer in streamed high volume, high velocity scientific data for both cloud and cloud-edge use cases. C_LI

bioinformatics↗
bioRxiv2020

Deep learning models for lipid-nanoparticle-based drug delivery

Large-scale time-lapse microscopy experiments are useful to understand delivery and expression in RNA-based therapeutics. The resulting data has high dimensionality and high (but sparse) information content, making it challenging and costly to store and process. Early prediction of experimental outcome enables intelligent data management and decision making. We start from time-lapse data of HepG2 cells exposed to lipid-nanoparticles loaded with mRNA for expression of green fluorescent protein (GFP). We hypothesize that it is possible to predict if a cell will express GFP or not based on cell morphology at time-points prior to GFP expression. Here we present results on per-cell classification (GFP expression/no GFP expression) and regression (level of GFP expression) using three different approaches. In the first approach we use a convolutional neural network extracting per-cell features at each time point. We then utilize the same features combined with: a long-short-term memory (LSTM) network encoding temporal dynamics (approach 2); and time-series feature extraction using the python package tsfresh followed by principal component analysis and gradient boosting machines (approach 3), to reach a final classification or regression result. Application of the three approaches to a previously unanalyzed test set of cells showed good predictive performance of all three approaches but that accounting for the temporal dynamics via LSTMs or tsfresh led to significantly improved performance. The predictions made by the LSTM and tsfresh applications were not significantly different. The results highlight the benefit of accounting for temporal dynamics when studying drug delivery using high content imaging.

cell biology↗
bioRxiv2019

Identification of spatial compartments in tissue from in situ sequencing data

Spatial organization of tissue characterizes biological function, and spatially resolved gene expression has the power to reveal variations of features with high resolution. Here, we propose a novel graph-based in situ sequencing decoding approach that improves recall, enabling precise spatial gene expression analysis. We apply our method on in situ sequencing data from mouse brain sections, identify spatial compartments that correspond with known brain regions, and relate them with tissue morphology.

neuroscience↗