Biology subjects

Tejada-Lapuerta, A.

Publications and source records attributed to Tejada-Lapuerta, A..

5 recordsLinked to original sources

bioRxiv2025

scConcept: Contrastive pretraining for technology-agnostic single-cell representations beyond reconstruction

Recent large-scale single-cell foundation models have shown promise for exploring cellular states, yet they often underperform compared to simpler, domain-specific methods, raising concerns about their broader applicability. A key limitation lies in their reliance on masked language modeling, which is well suited for generative language tasks but poorly aligned with learning rich cell-level embeddings required in single-cell biology. Moreover, the proliferation of transcriptomic technologies--from whole transcriptome dissociated assays to image-based targeted profiling--poses a major challenge for cross-platform generalization. Here, we align with recent advances in machine learning to move beyond reconstruction metrics, which often do not capture important sample variation. We present scConcept ("contrastive cell pre-training"), a transformer-based contrastive learning framework that directly optimizes cell embeddings by contrasting multiple views of cells. By replacing gene-level reconstruction with a cell-level identification task, scConcept learns robust representations that are invariant to count distributions and gene panel selection, across diverse assays and technologies. To highlight the capability of the proposed framework, we pretrain scConcept on a similar corpus of over 30 million single-cell RNA-seq profiles as recent foundation models. Our approach demonstrates superior performance not only compared to state-of-the-art pretrained foundation models but also domain-specific methods in various downstream tasks, including cell-type annotation, technology integration, dissociated to spatial cell-type transfer, spatial imputation, gene panel optimization, and mapping new technologies on already existing atlases. Our results highlight contrastive pretraining as a powerful alternative to reconstruction-based strategies for single-cell modeling, providing a path toward general-purpose, technology-agnostic cell representations.

bioinformatics↗

bioRxiv2025

CellFlow enables generative single-cell phenotype modeling with flow matching

High-content phenotypic screens provide a powerful strategy for studying biological systems, but the scale of possible perturbations and cell states makes exhaustive experiments unfeasible. Computational models that are trained on existing data and extrapolate to correctly predict outcomes in unseen contexts have the potential to accelerate biological discovery. Here, we present CellFlow, a flexible framework based on flow matching that can model single cell phenotypes induced by complex perturbations. We apply CellFlow to various phenotypic screens, accurately predicting expression responses to a wide range of perturbations, including cytokine stimulation, drug treatments and gene knockouts. CellFlow successfully modeled developmental perturbations at the whole-embryo scale and guided cell fate and organoid engineering by predicting heterogeneous cell populations arising from combinatorial morphogen treatments and by performing a virtual organoid protocol screen. Taken together, CellFlow has the potential to accelerate discovery from phenotypic screens by learning from existing data and generating phenotypes induced by unseen conditions.

bioinformatics↗

bioRxiv2024

Exploring the Boundaries of Medulloblastoma Subgroups with Synthetic Data Generation

Medulloblastoma is a childhood brain tumor traditionally classified into four molecular subgroups. Recent evidence suggests that Groups 3 and 4 represent a biological continuum rather than distinct entities, a paradigm shift with significant implications for understanding disease biology and treatment strategies. Nevertheless, assessing this hypothesis is challenging mainly due to data scarcity. In this study, we analyze the largest available transcriptomics dataset to provide compelling evidence for the existence of an intermediate subgroup between Groups 3 and 4, characterized by distinct molecular features. To overcome limitations posed by data scarcity, we employ synthetic data generation using a Variational Autoencoder and apply explainability techniques to identify key relationships between gene expression and disease subgroups. Furthermore, by incorporating Machine Learning Fairness approaches, we demonstrate that overlooking this intermediate subgroup can result in treatment disparities. Our findings are further supported by both existing and newly proposed studies using diverse datasets and methodologies, including graphbased analyses and multi-scale simulations, underscoring the robustness and reproducibility of our results. This study demonstrates the potential of synthetic data generation to refine rare disease subtyping and advance our understanding of the underlying biological mechanisms. Keywords: Medulloblastoma, pediatric cancer, representation learning, autoencoder, synthetic data

bioinformatics↗

bioRxiv2024

Scalable and universal prediction of cellular phenotypes

Biological systems can be interrogated by perturbing individual components and observing the consequences across molecular, cellular, and phenotypic levels. The vast combinatorial space of possible perturbations and responses makes exhaustive experimentation infeasible. Recent advances in machine learning have shown that training on diverse datasets enables transfer learning across tasks, capturing patterns that generalize and improving performance on previously unseen problems. Inspired by this principle, we present Prophet, a transformer-based model pretrained on a vast, heterogeneous collection of perturbation experiments. This pretraining allows Prophet to predict the outcomes of untested genetic or chemical perturbations in novel cellular contexts, spanning phenotypes such as gene expression, viability, and morphology. By leveraging shared structure across apparently disconnected assays, Prophet provides a scalable framework for large-scale virtual screening and prioritization of informative experiments. Prophet consistently outperforms baseline models, including those trained on single phenotypes, showing that transfer learning between phenotypes not only is possible but improves predictive accuracy. Its capabilities extends to in vivo developmental systems, where it recapitulates known lineage biology and proposes new candidates. In a large-scale in silico screen for melanoma, Prophet identified and experimentally validated compounds with selective activity that mirrored clinically approved therapies, demonstrating its ability to transform perturbation biology into a predictive and scalable engine for therapeutic discovery.

cell biology↗

bioRxiv2024

Nicheformer: a foundation model for single-cell and spatial omics

Tissue makeup relies fundamentally on the cellular microenvironment. Spatial single-cell genomics allows probing the underlying cellular interactions in an unbiased, scalable fashion. To learn a unified cell representation that accounts for local dependencies in the cellular microenvironment, we propose Nicheformer, a transformer-based foundation model that combines human and mouse dissociated single-cell and targeted spatial transcriptomics data. Pretrained on over 57 million dissociated and 53 million spatially resolved cells across 73 tissues on cellular reconstruction, the model is fine-tuned on spatial tasks for spatial omics data to decode spatially resolved cellular information. Nicheformer excels in linear-probing and fine-tuning scenarios for a novel set of downstream tasks, in particular spatial composition prediction and spatial label prediction. We further show that existing foundation models trained on dissociated single-cell data alone are not capable of recapitulating the spatial complexity of cells in their microenvironments, indicating that multiscale models are required to understand complex local dependencies at scale. Nicheformer enables the prediction of the spatial context of dissociated cells, allowing the transfer of rich spatial information to scRNA-seq datasets. Overall, Nicheformer sets the stage for the next generation of machine-learning models in spatial single-cell analysis. Extended AbstractTissue makeup and the corresponding orchestration of vital biological activities, ranging from development and differentiation to immune response and regeneration, rely fundamentally on the cellular microenvironment and the interactions between cells. Spatial single-cell genomics allows probing such interactions in an unbiased and, increasingly, scalable fashion. To learn a unified cell representation that accounts for local dependencies in the cellular microenvironment and the underlying cell interactions, we propose to generalize recent foundation modeling approaches for disassociated single-cell transcriptomics to the spatial omics setting. Our model, Nicheformer, is a transformer-based foundation model that combines human and mouse dissociated single-cell and targeted spatial transcriptomics data to learn a cellular representation useful for a large variety of downstream tasks. Nicheformer is pretrained on over 57 million dissociated and 53 million spatially resolved cells across 73 tissues from both human and mouse. Subsequently, the model is fine-tuned on spatial tasks for spatial omics data to decode spatially resolved cellular information. We demonstrate the usefulness of Nicheformer in both linear-probing as well as fine-tuning scenarios on a novel set of spatially-relevant downstream tasks such as spatial density prediction or niche and region label prediction. In particular, we show that Nicheformer enables the prediction of the spatial context of dissociated cells, allowing the transfer of rich spatial information to scRNA-seq datasets. We define a series of novel spatial prediction problems and observe consistent top performance of Nicheformer, demonstrating the advantage of the improved model capacity of the underlying transformer. Additionally, we benchmarked Nicheformer in these tasks against scGPT1, Geneformer2, scVI3 and PCA and show that the Nicheformer architecture excels in these tasks. Altogether, our large-scale resource of more than 110 million cells in a partial spatial context, together with the set of novel spatial learning tasks and the Nicheformer model itself, will pave the way for the next generation of machine-learning models for spatial single-cell analysis.

bioinformatics↗