Biology subjects

Staub, E.

Publications and source records attributed to Staub, E..

9 recordsLinked to original sources

bioRxiv2026

Modelling interpretable patient-level representationsfrom structured and simple multimodal data

Patient cohort profiling increasingly includes structured views for multiple modalities, such as single-cell RNA sequencing, spatial transcriptomics or proteomics, and histology, each providing multiple subobservations per patient, including single cells, spatial spots or patches. To model such data along with simple patient-level views, current multimodal integration methods typically rely on separately precomputed summaries and fail to fully leverage information in structured views. Here we present FACTMx, a variational framework that jointly models structured and simple views to learn interpretable patient-level representations. FACTMx couples latent patient factors with subobservation clustering and per-patient component proportions, enabling direct interpretation and downstream association analyses. The framework supports different structured-view mixture assumptions, including topic- and Gaussian-structured data, while retaining modular encoder-decoder parameterisations. In simulations spanning sparse and dense dependencies and multiple noise regimes, FACTMx improved reconstruction, integration and recovery of structured components relative to previous methods. Applied to non-small cell lung cancer cohorts, FACTMx captured survival-associated latent signals linked to immune microenvironments, gene expression pathways and spatially coherent histological patterns. In a longitudinal coronary syndrome cohort, FACTMx highlighted an outcome-associated axis connected to ejection-fraction change, immune cell states, soluble mediators and cardiac injury markers. These results support joint structured-simple modelling for interpretable multimodal patient stratification.

bioinformatics↗

bioRxiv2026

Cross-View Latent Integration via Nonparametric Gamma Shrinkage Factor Analysis

Factor analysis is a dominant paradigm for multi-omic heterogeneous data, but is challenged by partially redundant signals and noise across views and by an unknown true number of factors. We present CLING, an unsupervised multi-view factor model with hierarchical Bayesian sparsity priors: a product-of-Gammas prior inducing cumulative column-wise shrinkage (increasing with factor index) coupled with a Gamma-Gamma local-precision hierarchy on loadings yielding heavy-tailed marginals. This pairing enables automatic factor selection by adaptively deactivating unsupported factors while retaining active ones during inference, and induces selective sparsity that allows salient loadings to escape shrinkage while collapsing negligible ones. As a fully conjugate hierarchical model, CLING admits a scalable variational inference algorithm for multi-view data. Across synthetic benchmarks and multiomics datasets, CLING recovers more accurate factors and more informative loadings while explaining at least as much variance as competitive multi-view baselines; on glioblastoma gene expression and DNA methylation data, CLING identifies pathways linked to tumor subtype and patient age. Source code: https://github.com/szczurek-lab/CLING.

bioinformatics↗

bioRxiv2025

Cellohood: multi-granular discovery of cellular neighborhoods with a permutation-invariant set transformer auto-encoder

Discovering cellular neighborhoods and their roles in disease requires computational methods that consider the full breadth of data and offer multi-level granularity and interpretability. Here, we introduce Cellohood, a permutation-invariant set transformer auto-encoder equipped with a clinical association pipeline. Cellohood encodes full readouts for bags of spatially co-localized cells and supports multi-level analyses, providing interpretability by mapping latent dimensions to spatial tissue features. Our model surpasses current methods in accuracy of cellular neighborhood detection for spatial transcriptomics of the human cortex and CODEX spleen data measuring lupus progression. Applied to cancer data across three granularity levels, our method recovers neighborhoods along the expected immune-cold to immune-hot spectrum and further refines them into biologically and clinically meaningful subclasses, revealing spatial patterns linked to prognosis, histology, stage, and tumor mutational burden, and uncovering subgroups that transcend standard classifications. Overall, Cellohood enables in-depth analysis of complex tissues, revealing clinically informative spatial neighborhoods.

bioinformatics↗

bioRxiv2023

Stromal signals dominate gene expression signature scores that aim to describe cancer-intrinsic stemness or mesenchymality characteristics

PurposeEpithelial-to-mesenchymal transition (EMT) in cancer cells confers migratory ability, a crucial aspect of tumor metastasis that frequently leads to death. In multiple studies, authors proposed gene expression signatures for EMT, stemness, and mesenchymality (EMT-related) characteristics of tumors based on bulk tumor expression profiling. However, recent studies have suggested that non-cancerous cells in the tumor micro- or macroenvironment heavily influence individual signature profiles. Experimental DesignWe analyzed scores of 11 published and frequently referenced gene expression signatures in bulk, single cell, and pseudo bulk expression data across multiple cancer types. ResultsOur study strengthens and extends the influence of non-cancerous cells on signatures that were proposed to describe EMT-related (EMT, mesenchymal, or stemness) characteristics in various cancer types. The cell type composition, especially the amount of tumor cells, of a tumor sample frequently dominates EMT-related signature scores. Additionally, our analyses revealed that stromal cells, most often fibroblasts, are the main drivers of the EMT-related signature scores. ConclusionsWe call attention to the risk of false conclusions about tumor properties when interpreting EMT-related signatures, especially in a clinical setting: high patient scores of EMT-related signatures or calls of "stemness subtypes" often result from low tumor cell content in tumor biopsies rather than cancer cell-specific stemness or mesenchymality/EMT characteristics.

bioinformatics↗

bioRxiv2023

Immune cell type signature discovery and random forest classification for analysis of single cell gene expression datasets

BackgroundRobust immune cell gene expression signatures are central to the analysis of single cell studies. Nearly all known sets of immune cell signatures have been derived by making use of only single gene expression datasets. Utilizing the power of multiple integrated datasets could lead to high-quality immune cell signatures which could be used as superior inputs to machine learning-based cell type classification approaches. ResultsWe established a novel gene expression similarity-based workflow for the discovery of immune cell type signatures that leverages multiple datasets, here four single cell expression datasets from three different cancer types. We used our immune cell signatures to train random forest classifiers for immune cell type assignment of single-cell RNA-seq datasets. We obtained similar or better prediction results compared to commonly used methods for cell type assignment in two independent benchmarking datasets. Our gene signature set yields higher prediction scores than other published immune cell type gene sets in our random forest approach. Discussion and conclusionWe demonstrated the quality of our immune cell signatures and their strong performance in a random forest-based cell typing approach. We argue that classifying cells based on our comparably slim sets of genes accompanied by a random forest-based approach not only matches or outperforms widely used published approaches. It also facilitates unbiased downstream statistical analyses of differential gene expression between cell types for 90% of all genes whose expression profiles have not been used for cell type classification.

genomics↗

bioRxiv2022

SLIDE-VIP: a comprehensive, cell line- and patient-based framework for synthetic lethality prediction in DNA damage repair, chromatin remodeling and cell cycle

Discovering synthetic lethal (SL) gene partners of cancer genes is an important step in developing cancer therapies. However, identification of SL interactions is challenging, due to a large number of possible gene pairs, inherent noise and confounding factors in the observed signal. To discover robust SL interactions, we devised SLIDE-VIP, a novel framework combining eight statistical tests, including a new patient data-based test iSurvLRT. SLIDE-VIP leverages multi-omics data from four different sources: gene inactivation cell line screens, cancer patient data, drug screens and gene pathways. We applied SLIDE-VIP to discover SL interactions between genes involved in DNA damage repair, chromatin remodeling and cell cycle, and their potentially druggable partners. The top 883 ranking SL candidates had strong evidence in cell line and patient data, 250-fold reducing the initial space of 200K pairs. Drug screen and pathway tests provided additional corroboration and insights into these interactions. We rediscovered well-known SL pairs such as RB1 and E2F3 or PRKDC and ATM, and in addition, proposed strong novel SL candidates such as PTEN and PIK3CB. In summary, SLIDE-VIP opens the door to the discovery of SL interactions with clinical potential. All analysis and visualizations are available via the online SLIDE-VIP WebApp.

bioinformatics↗

bioRxiv2021

CONET: Copy number event tree model of evolutionary tumor history for single-cell data

Copy number alterations constitute important phenomena in tumor evolution. Whole genome single cell sequencing gives insight into copy number profiles of individual cells, but is highly noisy. Here, we propose CONET, a probabilistic model for joint inference of the evolutionary tree on copy number events and copy number calling. CONET employs an efficient MCMC procedure to search the space of possible model structures and parameters and utilizes both per-bin and per-breakpoint data. We introduce a range of model priors and penalties for efficient regularization. CONET achieves excellent performance on simulated data and for 260 cells from xenograft breast cancer sample.

bioinformatics↗

bioRxiv2021

Interpretable deep recommender system model for prediction of kinase inhibitor efficacy across cancer cell lines

Computational models for drug sensitivity prediction have the potential to revolutionise personalized cancer medicine. Drug sensitivity assays, as well as profiling of cancer cell lines and drugs becomes increasingly available for training such models. Machine learning methods for drug sensitivity prediction must be optimized for: (i) leveraging the wealth of information about both cancer cell lines and drugs, (ii) predictive performance and (iii) interpretability. Multiple methods were proposed for predicting drug sensitivity from cancer cell line features, some in a multi-task fashion. So far, no such model leveraged drug inhibition profiles. Recent neural network-based recommender systems arise as models capable of predicting cancer cell line response to drugs from their biological features with high prediction accuracy. These models, however, require a tailored approach to model interpretability. In this work, we develop a neural network recommender system for kinase inhibitor sensitivity prediction called DEERS. The model utilizes molecular features of the cancer cell lines and kinase inhibition profiles of the drugs. DEERS incorporates two autoencoders to project cell line and drug features into 10-dimensional hidden representations and a feed-forward neural network to combine them into response prediction. We propose a novel model interpretability approach offering the widest possible assessment of the specific genes and biological processes that underlie the action of the drugs on the cell lines. The approach considers also such genes and processes that were not included in the set of modeled features. Our approach outperforms simpler matrix factorization models, achieving R=0.82 correlation between true and predicted response for the unseen cell lines. Using the interpretability analysis, we evaluate correlation of all human genes with each of the hidden cell line dimensions. Subsequently, we identify 67 biological processes associated with these dimensions. Combined with drug response data, these associations point at the processes that drive the cell line sensitivity to particular compounds. Detailed case studies are shown for PHA-793887, XMD14-99 and Dabrafenib. Our framework provides an expressive, multitask neural network model with a custom interpretability approach for inferring underlying biological factors and explaining cancer cell response to drugs.

bioinformatics↗

bioRxiv2019

Feature selection strategies for drug sensitivity prediction

Drug sensitivity prediction constitutes one of the main challenges in personalized medicine. The major difficulty of this problem stems from the fact that the sensitivity of cancer cells to treatment depends on an unknown subset of a large number of biological features. Although feature selection is the key to interpretable results and identification of potential biomarkers, a comprehensive assessment of feature selection methods for drug sensitivity prediction has so far not been performed. We propose feature selection approaches driven by prior knowledge of drug targets, target pathways, and gene expression signatures. We asses these methodologies on Genomics of Drug Sensitivity in Cancer (GDSC) dataset, a panel of around 1000 cell lines screened against multiple anticancer compounds. We compare our results with a baseline model utilizing genome-wide gene expression features and common data-driven feature selection techniques. Together, 2484 unique models were evaluated, providing a comprehensive study of feature selection strategies for the drug response prediction problem. For 23 drugs, the models achieve better predictive performance when the features are selected according to prior knowledge of drug targets and pathways. The best correlation of observed and predicted response using the test set is achieved for Linifanib (r=0.75). Extending the drug-dependent features with gene expression signatures yields models that are most predictive of drug response for 60 drugs, with the best performing example of Dabrafenib. Examples of how pre-selection of features benefits the model interpretability are given for Dabrafenib, Linifanib and Quizartinib. Based on GDSC drug data, we find that feature selection driven by prior knowledge tends to yield better results for drugs targeting specific genes and pathways, while models with the genome-wide features perform better for drugs affecting general mechanisms such as metabolism and DNA replication. For a significant group of the compounds, even a very small number of features based on simple drug properties is often highly predictive of drug sensitivity, can explain the mechanism of drug action and be used as guidelines for their prescription. In general, choosing appropriate feature selection strategies has the potential to develop interpretable models that are indicative for therapy design.

bioinformatics↗