Biology subjects

Claassen, M.

Publications and source records attributed to Claassen, M..

At least 19 recordsLinked to original sources

bioRxiv2026

Mechanistic gene-regulation-based trajectory inference with bowside

While single-cell RNA sequencing enables the in-depth study of differentiation processes, simultaneous trajectory inference and gene regulatory network reconstruction remain challenging. We propose bowside, a shallow neural stochastic differentiation model that mechanistically models gene regulation. Our method successfully recovers gene expression trends and the underlying regulatory interactions. We demonstrate bowside on hematopoiesis and induced pluripotent stem cell reprogramming data, retrieving known fate-determining network motifs.

bioinformatics↗

bioRxiv2026

G-LATO: Inference of Spatial Latent Ordering via Deep Gaussian Processes

Spatial transcriptomics enables the study of cells within their native tissue context, yet identifying gradients of cellular development remains challenging. We introduce a deep Gaussian process model to address this gap. Our method recovers spatially smooth gradients explaining observed gene expression. We illustrate our method on healthy liver and glioblastoma data in reconstructing known spatial organisation and uncovering new pathological gradients, thus providing robust inference for spatial biology.

bioinformatics↗

bioRxiv2026

t2pmhc: A Structure-Informed Graph Neural Network to predict TCR-pMHC Binding

Mapping of T cell receptors (TCRs) to their cognate MHC-presented peptides (pMHC) is central for the development of precision immunotherapies and vaccine design. However, accurate prediction of TCR affinity to peptide antigens remains an open challenge. Most approaches rely solely on sequence information, although increasing evidence suggests that TCR-pMHC binding is primarily determined by three-dimensional structural interactions within the entire TCR-pMHC complex. Consequently, sequence-based methods often fail to generalize to peptides not included in the training data (unseen peptides). Here we introduce t2pmhc, a structure-based graph neural network framework for predicting TCR-pMHC binding using predicted structures of the entire TCR-pMHC complex. We evaluated a Graph Convolutional Network (GCN) and a Graph Attention Network, both demonstrating improved generalization to unseen peptides compared to state-of-the-art models across a variety of public datasets. Evaluation with crystallographic structures yields high-confidence predictions, indicating that current limitations of structure-based models are largely driven by the accuracy of structure prediction. Analysis of node attention patterns in t2pmhc-GCN reveals biologically consistent patterns, assigning high attention to the peptide and the CDR3 regions. Within the peptide sequence, canonical MHC anchor residues are consistently downweighted, whereas potential TCR-binding residues are upweighted. These findings establish t2pmhc as a structure-informed framework for robust TCR-pMHC binding prediction, enabling improved generalization to unseen antigens and providing a foundation for integrating TCR repertoire sequencing into vaccine design and immunotherapy.

bioinformatics↗

bioRxiv2025

velotest: Statistical assessment of RNA velocity embeddings reveals quality differences for reliable trajectory visualizations

Contemporary single-cell RNA-sequencing studies typically use RNA velocity analyses to investigate cellular dynamics. Commonly, RNA velocities are assessed with two-dimensional projections. Although it is well established that nonlinear embeddings of gene expression data cannot comprehensively preserve the structure of the original high-dimensional data across scales, there is no tool that systematically assesses and quantifies the faithfulness of velocity embeddings. This situation potentially leads to misleading velocity visualizations and putative cell state transitions. To this end, we introduce velotest, a cell-level statistical test that quantifies the faithfulness of a velocity embedding with respect to its high-dimensional counterpart. Applying velotest across multiple datasets and embedding methods, we find strong differences in the embedding qualities between embedding methods, datasets, and regions in individual datasets. velotest thus provides an essential validation step and enables researchers to select a suitable visualization of high-dimensional velocities, ensuring that hypotheses drawn from velocity projections are grounded in accurately represented dynamics.

bioinformatics↗

bioRxiv2025

Accurate quantification of spliced and unspliced transcripts for single-cell RNA sequencing with tidesurf

MotivationSingle-cell RNA sequencing (scRNA-seq) allows for the detailed analysis of dynamic cellular processes. In particular, this has been enabled by the estimation of RNA velocity, the derivative of gene expression, from separate count matrices for different splice states, which provides information about a cells immediate future even in snapshot data. Useful velocity estimates strongly depend on accurate counts for spliced and unspliced transcripts. Velocyto remains the standard tool for spliced and unspliced mRNA molecule quantification. However, despite considerable advances in scRNA-seq protocols, velocyto has not been updated to account for peculiarities of new protocols, such as popular approaches based on 5 chemistry. ResultsTo address this shortcoming, we present tidesurf, a command line tool for the quantification of spliced and unspliced transcript molecules from scRNA-seq libraries. Employing it on four different publicly available 10x Genomics Chromium datasets, we show the accuracy on various datasets generated with either 3 or 5 chemistry, whereas velocytos results are highly erroneous for the latter. Considering broader applicability, our results highlight tidesurf as a potential replacement for velocyto. Availability and implementationA Python implementation of tidesurf is available from PyPI and at github.com/janschleicher/tidesurf. Code for reproducing the analyses is available at github.com/janschleicher/tidesurf projects.

bioinformatics↗

bioRxiv2025

Quantification of spliced and unspliced transcripts by velocyto is inaccurate for 5'-sequencing data

RNA velocity allows for the prediction of future states of individual cells based on their current transcriptional activity. The technique depends on the separate quantification of spliced and unspliced transcripts in single-cell RNA-sequencing data, first introduced by velocyto in 2018. Since its introduction, significant advances have been made in the field, including new protocols by 10x Genomics that enable sequencing from the 5-end of mRNA molecules instead of the 3-end. Despite these advances, velocyto has not been updated since its release but is still commonly used with these new protocols. In this study, we demonstrate that velocyto cannot accurately detect the reversed direction of transcripts, leading to incorrect count assignments. By comparing velocyto to alevin-fry, a quantification method compatible with 5-sequencing data, we show that this limitation can result in substantial deviations in inferred velocities and differing interpretations. Therefore, we do not recommend the use of velocyto with 5-sequencing data.

bioinformatics↗

bioRxiv2023

Factorial state-space modelling for kinetic clustering and lineage inference

Single-cell RNA sequencing (scRNAseq) protocols measure the abundance of expressed transcripts for single cells. Gene expression profiles of cells (cell-states) represent the functional properties of the cell and are used to cluster cell-states that have a common functional identity (cell-type). Standard clustering methods for scRNAseq data perform hard clustering based on KNN graphs. This approach implicitly assumes that variation among cell-states within a cluster does not correspond to changes in functional properties. Differentiation is a directed process of transitions between cell-types via gradual changes in cell-states over the course of the process. We propose a latent state-space Markov model that utilises cell-state transitions derived from RNA velocity to model differentiation as a sequence of latent state transitions and to perform soft kinetic clustering of cell-states that accommodates the transitional nature of cells in a differentiation process. We applied this model to the differentiation of Radial-glia cells into mature neurons and demonstrate the utility of our method in discriminating between functional and transitional cell-states.

bioinformatics↗

bioRxiv2023

S3-CIMA: Supervised spatial single-cell image analysis for the identification of disease-associated cell type compositions in tissue

The spatial organization of various cell types within the tissue microenvironment is a key element for the formation of physiological and pathological processes, including cancer and autoimmune diseases. Here, we present S3-CIMA, a weakly supervised convolutional neural network model that enables the detection of disease-specific microenvironment compositions from high-dimensional proteomic imaging data. We demonstrate the utility of this approach by determining cancer outcome- and cellular signaling-specific spatial cell state compositions in highly multiplexed fluorescence microscopy data of the tumor microenvironment in colorectal cancer. Moreover, we use S3-CIMA to identify disease onset-specific changes of the pancreatic tissue microenvironment in type 1 diabetes using imaging mass cytometry data. We evaluated S3-CIMA as a powerful tool to discover novel disease-associated spatial cellular interactions from currently available and future spatial biology datasets.

bioinformatics↗

bioRxiv2022

Phase-specific signatures of wound fibroblasts and matrix patterns define cancer-associated fibroblast subtypes

Healing wounds and cancers present remarkable cellular and molecular parallels, but the specific roles of the healing phases are largely unknown. We developed a bioinformatics pipeline to identify genes and pathways that define distinct phases across the time course of healing. Their comparison to cancer transcriptomes revealed that a resolution-phase wound signature is associated with increased severity in skin cancer and enriches for extracellular matrix-related pathways. Comparisons of transcriptomes of early- and late-phase wound fibroblasts vs skin cancer-associated fibroblasts (CAFs) identified an "early-wound" CAF subtype, which localizes to the inner tumor stroma and expresses collagen-related genes that are controlled by the RUNX2 transcription factor. A "late-wound" CAF subtype localizes to the outer tumor stroma and expresses elastin-related genes. Matrix imaging of primary melanoma tissue microarrays validated these matrix signatures and identified collagen- vs elastin-rich niches within the tumor microenvironment, whose spatial organization predicts survival and recurrence. These results identify wound-regulated genes and matrix patterns with prognostic potential in skin cancer.

genomics↗

bioRxiv2022

An assessment of Nano-RECall: Interpretation of Oxford Nanopore sequence data for HIV-1 drug resistance testing

IntroductionOxford Nanopore Technologies (ONT) offer sequencing with low-capital-layout sequencing options, which could assist in expanding HIV drug resistance testing to resource limited settings. However, sequence analysis remains time time-consuming and reliant on skilled personnel. Moreover, current ONT bioinformatic pipelines provide a single consensus sequence that is not equivalent to Sanger sequencing, as drug resistance is often detected in mixed populations. We have therefore investigated an integrated bioinformatic pipeline, Nano-RECall, for seamless drug resistance of low read coverage ONT sequence data from affordable Flongle or MinION flow cells. MethodsWe compared Sanger sequencing to ONT sequencing of the same HIV-1 subtype C polymerase chain reaction (PCR) amplicons, respectively using RECall and the novel Nano-RECall bioinformatics pipelines. Amplicons were from separate assays a) Applied Biosystems HIV-1 Genotyping Kit (ThermoFisher) spanning protease (PR) to reverse transcriptase (RT) (PR-RT) (n=46) and b) homebrew integrase (IN) (n=21). We investigated optimal read-depth by assessing the coefficient of variation (CV) of nucleotide proportions for various read-depths; and between replicates of 400 reads. The agreement between Sanger sequences and ONT sequences were assessed at nucleotide level, and at codon level for Stanford HIV drug resistance database mutations. ResultsThe coefficient of variation of ONT minority variants plateaued after a read depth of 400-fold implying limited benefit of additional depth and replicates of 400 reads showed a CV of [~]6 % for a representative position. The average sequence similarity between ONT and Sanger sequences was 99.3% (95% CI: 99.1-99.4%) for PR-RT and 99.6% (95% CI: 99.4-99.7%) for INT. Drug resistance mutations did not differ for 21 IN sequences; 16 mutations were detected by both ONT- and Sanger sequencing. For the 46 PR and RT sequences, 245 mutations were detected by either ONT or Sanger, of these 238 (97.1%) were detected by both. ConclusionsThe Nano-RECall pipeline, freely available as a downloadable application on a Windows computer, provides Sanger-equivalent HIV drug resistance interpretation. This novel pipeline combined with a simple workflow and multiplexing samples on ONT flow-cells would contribute to making HIV drug resistance sequencing feasible for resource limited settings.

microbiology↗

bioRxiv2022

Privacy-Preserving Federated Neural Network Learning for Disease-Associated Cell Classification

Training accurate and robust machine learning models requires a large amount of data that is usually scattered across data-silos. Sharing or centralizing the data of different healthcare institutions is, however, unfeasible or prohibitively difficult due to privacy regulations. In this work, we address this problem by using a novel privacy-preserving federated learning-based approach, PriCell, for complex machine learning models such as convolutional neural networks. PriCell relies on multiparty homomorphic encryption and enables the collaborative training of encrypted neural networks with multiple healthcare institutions. We preserve the confidentiality of each institutions input data, of any intermediate values, and of the trained model parameters. We efficiently replicate the training of a published state-of-the-art convolutional neural network architecture in a decentralized and privacy-preserving manner. Our solution achieves an accuracy comparable to the one obtained with the centralized solution, with an improvement of at least one-order-of-magnitude in execution time with respect to prior secure solutions. Our work guarantees patient privacy and ensures data utility for efficient multi-center studies involving complex healthcare data.

cell biology↗

bioRxiv2020

Cytopath: Simulation based inference of differentiation trajectories from RNA velocity fields

AO_SCPLOWBSTRACTC_SCPLOWTrajectory inference from single-cell RNA sequencing data bears the potential to systematically reconstruct complex differentiation processes, however inferring trajectories that accurately model the biological characteristics of varied processes continues to be a challenge, notwithstanding the many available solutions. In general, trajectory and pseudotime inference methods have so far suffered from the ambiguity of static single-cell transcriptome snapshots lacking a concept of directionality and rate of transcriptional activity. We report Cytopath, a method for trajectory inference that takes advantage of transcriptional activity information from RNA velocity of single-cells to perform trajectory inference. Cytopath performs this task by defining a Markov chain model, simulating an ensemble of possible differentiation trajectories and constructs a consensus trajectory. We show that Cytopath can recapitulate the topological and molecular characteristics of the differentiation process under study. In our analysis we include differentiation trajectories with varying bifurcated, circular, convergent and mixed topology studied in single-snapshot as well as time-series single-cell RNA sequencing experiments. We demonstrate superior and enabling capability to reconstruct differentiation trajectories in comparison to state-of-the art trajectory inference approaches.

bioinformatics↗

bioRxiv2020

Fate trajectories of CD8+ T cells in chronic LCMV infection

AO_SCPLOWBSTRACTC_SCPLOWIn chronic infections CD8+ T cells acquire a state termed "exhaustion" which is characterized by impaired effector functions and expression of co-inhibitory receptors as response to continuous TCR stimulation. Recently, the pool of exhausted T cells has been shown to harbor multiple functionally distinct populations with memory-like and effector-like features, though differentiation and lineage relations between these are unclear. In this work we present a comprehensive scRNAseq time-series analysis from beginning of infection to established exhaustion in CD8 T cells. We apply lineage inference using informed cell transitions derived from RNA velocity to identify differential start and end states and connections between them. We identify a branch region early during chronic infection where pre-committed cells separate into an exhausted and a memory-like lineage and discovered molecular markers demarcating this branch event. Adoptive transfer experiments confirmed fate-commitment of cells only after this branch point. We additionally linked the progression along developmental lineages to antigenic TCR stimulation.

immunology↗

bioRxiv2020

Genome-wide CRISPR and small-molecule screens uncover targetable dependencies in ATRT

Atypical teratoid rhabdoid tumors (ATRT) are incurable high-grade pediatric brain tumors. Concepts for molecular-driven therapies in ATRTs lag behind, mainly due to the absence of actionable genetic alterations. We performed genome-wide CRISPR/Cas9 knockout screens in six human ATRT cell lines and identified a total of 671 context-specific essential genes. Based on these genetic dependencies, we constructed a library of small-molecule inhibitors that we found to preferentially inhibit growth of ATRT cells. CDK4/6 inhibitors, among the most potent drugs in our library, are capable of inhibiting tumor growth due to mutual exclusive dependency of ATRTs on CDK4 or CDK6. These distinct dependencies drive heterogeneity in response to CDK4/6 inhibitors in ATRTs. Our approach might serve as a blueprint for fostering the identification of functionally-instructed therapeutic strategies in other incurable diseases beyond ATRT, whose genomic profiles also lack actionable alterations so far.

cancer biology↗

bioRxiv2020

Deterministic scRNA-seq of individual intestinal organoids reveals new subtypes and coexisting distinct stem cell pools

Single-cell RNA-sequencing (scRNA-seq) has transformed our ability to resolve cellular properties across systems. However, current scRNA-seq platforms are one-size-fits-all approaches that are tailored toward large cell inputs (> 1,000 cells), rendering them inefficient and costly when processing small, individual tissue samples. This important drawback tends to be resolved by loading bulk samples, but this yields confounded mosaic cell population read-outs. To overcome these technological limitations, we developed a deterministic, mRNA-capture bead and cell co-encapsulation dropleting system, DisCo. We demonstrate that DisCo enables precise particle and cell positioning and droplet sorting control through combined machine-vision and multilayer microfluidics. In comparison to other microfluidics systems, the active flow control driving DisCo, enables continuous operation and processing of low-input samples (< 100 cells) at high capture efficiency (> 70%). To underscore the unique capabilities of our approach, we analyzed intestinal organoid development by "DisCo-ing" 31 individual organoids at varying developmental stages. This revealed extensive organoid heterogeneity, identifying distinct subtypes including a regenerative fetal-like Ly6a+ stem cell population which persists as symmetrical cysts even under differentiation conditions. Furthermore, we uncovered a so far uncharacterized "gobloid" subtype consisting predominantly of precursor and mature (Muc2+) goblet cells. These findings demonstrate the unique power of DisCo in providing high-resolution snapshots of cellular heterogeneity among small, individual tissues.

bioengineering↗

bioRxiv2020

Positive feedback induces switch between distributive and processive phosphorylation of Hog1

Cellular decision making often builds on ultrasensitive MAPK pathways. The phosphorylation mechanism of MAP kinase has so far been described as either distributive or processive, with distributive mechanisms generating ultrasensitivity in theoretical analyses. However, the in vivo mechanism of MAP kinase phosphorylation and its regulation by feedback loops remain unclear. We thus characterized the regulation of the MAP kinase Hog1 in Saccharomyces cerevisiae, which is transiently activated in response to hyperosmolarity. Specifically, we combined Hog1 activation data from different modalities and multiple conditions. We constructed ODE models with different pathway topologies, which were then assessed via parameter estimation and model selection. Interestingly, our best fitting model switches between distributive and processive phosphorylation behavior via a positive feedback loop targeting the MAP kinase-kinase Pbs2. Simulations further suggest that this mixed mechanism is required not only for full sensitivity to stimuli, but also to ensure robustness to different perturbations.

systems biology↗

bioRxiv2019

psupertime: supervised pseudotime inference for single cell RNA-seq data with sequential labels

Single cell RNA-seq has been successfully combined with pseudotime inference methods to investigate biological processes which have sequential labels, such as time series studies of development and differentiation. Pseudotime methods developed to date ignore the labels, and where there is substantial variation in the data not associated with the labels (such as cell cycle variation or batch effects), they can fail to find relevant genes. We introduce psupertime, a supervised pseudotime approach which outperforms benchmark pseudotime methods by explicitly using the sequential labels as input. psupertime uses a simple, regression-based model, which by acknowledging the labels assures that genes relevant to the process, rather than to major drivers of variation, are found. psupertime is applicable to the wide range of single cell RNA-seq datasets with sequential labels, derived from either experimental design or user-selected cell cluster sequences, and provides a tool for targeted identification of genes regulated along biological processes.

genomics↗

bioRxiv2018

Dynamic Distribution Decomposition for Single-Cell Snapshot Time Series Identifies Subpopulations and Trajectories during iPSC Differentiation

Recent high-dimensional single-cell technologies such as mass cytometry are enabling time series experiments to monitor the temporal evolution of cell state distributions and to identify dynamically important cell states, such as fate decision states in differentiation. However, these technologies are destructive, and require analysis approaches that temporally map between cell state distributions across time points. Current approaches to approximate the single-cell time series as a dynamical system suffer from too restrictive assumptions about the type of kinetics, or link together pairs of sequential measurements in a discontinuous fashion.\n\nWe propose Dynamic Distribution Decomposition (DDD), an operator approximation approach to infer a continuous distribution map between time points. On the basis of single-cell snapshot time series data, DDD approximates the continuous time Perron-Frobenius operator by means of a finite set of basis functions. This procedure can be interpreted as a continuous time Markov chain over a continuum of states. By only assuming a memoryless Markov (autonomous) process, the types of dynamics represented are more general than those represented by other common models, e.g., chemical reaction networks, stochastic differential equations. Additionally, the continuity assumption ensures that the same dynamical system maps between all time points, not arbitrarily changing at each time point. We demonstrate the ability of DDD to reconstruct dynamically important cell states and their transitions both on synthetic data, as well as on mass cytometry time series of iPSC reprogramming of a fibroblast system. We use DDD to find previously identified subpopulations of cells and to visualize differentiation trajectories.\n\nDynamic Distribution Decomposition allows interpreting high-dimensional snapshot time series data as a low-dimensional Markov process, thereby enabling an interpretable dynamics analysis for a variety of biological processes by means of identifying their dynamically important cell states.\n\nAuthor summaryHigh-dimensional single-cell snapshot measurements are now increasingly utilized to study dynamic processes. Such measurements enable us to evaluate cell population distributions and their evolution over time. However, it is not trivial to map these distribution across time and to identify dynamically important cell states, i.e. bottleneck regions of state space exhibiting a high degree of change. We present Dynamic Distribution Decomposition (DDD) achieving this task by encoding single-cell measurements as linear combination of basis function distributions and evolving these as a linear system. We demonstrate reconstruction of dynamically important states for synthetic data of a bifurcated diffusion process and mass cytometry data for iPSC reprogramming.

systems biology↗