Biology subjects

Frossard, P.

Publications and source records attributed to Frossard, P..

6 recordsLinked to original sources

bioRxiv2026

Fine-Grained and Reliable Cell Classification from Pathology Images

Molecular assays such as spatial transcriptomics facilitate fine-grained identification of cell types for analyzing tumor microenvironments. However, they are expensive to acquire in routine clinical practice compared to hematoxylin and eosin (H&E) images. With molecular guidance, such images can distinguish cell lineages, yet often fail to resolve closely related subtypes. To overcome this, we introduce CellBridge, a multi-modal contrastive learning framework pretrained on 16M cells with paired H&E and spatial profiles, for transferring molecular information into morphological representations. CellBridge further incorporates a Cell-Ontology-aware conformal prediction module that assigns labels at different levels of specificity: leaf-level cell types when morphology alone provides sufficient evidence, and broader ancestor categories when distinguishing among more specific cell types would require molecular information. Across in-domain and out-of-domain benchmarks, CellBridge outperforms leading pathology foundation models in macro F1 by up to 16 percentage points, with the largest gains on the most fine-grained label spaces. It consistently achieves smaller prediction sets, while reducing class-wise coverage gaps. We envision CellBridge as a step toward extracting cellular information that currently requires costly molecular assays from routinely acquired pathology images, while explicitly identifying where such assays remain necessary. Our code is available at https://github.com/Lotfollahi-lab/cellbridge.

bioinformatics↗

bioRxiv2026

Resolving context-specific protein-protein interactomes forbiological discovery and therapeutic target prioritisation

Protein function is shaped by cellular context, yet most protein representations and interaction maps remain context-agnostic. Here we present ProtScape, a multiscale graph-learning framework integrating global protein interactions, cell-type gene expression and protein language models to learn context-specific representations and infer interactomes across more than 200 cell types. ProtScape substantially outperforms existing approaches in interaction reconstruction, increasing the area under the precision-recall curve by 40 percentage points. Its predicted interactions were supported by held-out continuous STRING global evidence, while its representations recovered higher-order protein organisation. In patient-derived amyotrophic lateral sclerosis motor neurons, ProtScape revealed stage-specific network changes implicating RAB-dependent trafficking as a candidate early disease mechanism. In Parkinson's disease, it recovered clinically supported therapeutic targets from a proteome-wide search space 16-fold smaller than that required by competing representations. Together, ProtScape provides a scalable framework for translating context-specific interactome organisation into experimentally testable disease mechanisms and therapeutic hypotheses.

bioinformatics↗

bioRxiv2026

Topology-aware reconstruction of cellular state landscapes from microscopy using self-supervised learning

Morphology and spatial organisation provide complementary readouts of cellular state. However, reconstructing continuous cellular state landscapes from imaging data remains challenging, particularly in dense biological cultures. Here we present SI-SimCLR, a spatially informed self-supervised learning framework that learns biologically informative representations directly from fluorescence microscopy images without requiring segmentation or manual annotation. Combined with a graph-based partial optimal transport framework, SI-SimCLR enables reconstruction of cellular phenotypic landscapes from static imaging data, revealing how phenotypic substates are organised and connected. To establish and validate this framework, we generated a multimodal dataset of human iPSC-derived astrocytes using high-content imaging and matched bulk transcriptomics. SI-SimCLR resolved distinct interconnected astrocyte substates associated with disease and inflammatory states. ALS astrocytes occupied constrained regions of the morphological landscape. Strikingly, morphology and transcriptomics captured distinct and complementary aspects of astrocyte state variation.Together, our framework establishes a scalable and annotation-free strategy for reconstructing cellular phenotypic landscapes from microscopy data, enabling analysis of cellular heterogeneity, landscape connectivity and phenotypic responses across biological systems.

bioinformatics↗

bioRxiv2025

Extended pre-training of histopathology foundation models uncovers co-existing breast cancer archetypes characterized by RNA splicing or TGF-β dysregulation

In recent years, histopathology foundation models (hFM) have rapidly advanced in size and complexity, achieving excellent performance in tasks such as cancer diagnosis and biomarker discovery. Here, we reveal novel capabilities of these models by specializing hFMs, originally trained on diverse tissue types, specifically to invasive tumor tissue. It enables unprecedented discrimination of visually similar yet molecularly distinct tumor regions, that were previously indistinguishable by baseline models, which eventually leads to uncovering new biological insights into breast cancer. Our contributions are threefold. First, to the best of our knowledge, this is the first study to systematically evaluate the biological concepts encoded within hFM representations across multiple scales. Second, we explore extended pre-training to identify optimal conditions that enhance the models ability to encode richer, tumor tissue-specific biological concepts. We show that this refinement strategy transforms generalist models into a specialist one capable of resolving subtle, recurrent tumor regions with distinct morphological and molecular identities, called tumor archetypes. Finally, leveraging this specialized model, we uncover two dominant tumor archetypes in invasive breast cancer characterized by distinct aberrant gene expression signatures, notably RNA metabolism dysregulation and TGF-{beta} signaling. Strikingly, these archetypes coexist within the same tumors as spatially distinct regions with varying densities and patterns, and are recurrent across patients, highlighting their universality and potential clinical relevance for patient stratification. Altogether our study demonstrates how extended pre-training of state-of-the-art hFM with specific tumor tissues can unlock rich molecular and morphological information encoded in H&E images. By providing a more accessible approach to investigating tumor heterogeneity, this work opens new avenues for precision oncology, using routine histopathology slides and low computational resources. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=128 SRC="FIGDIR/small/645192v1_ufig1.gif" ALT="Figure 1"> View larger version (30K): org.highwire.dtl.DTLVardef@38e660org.highwire.dtl.DTLVardef@19ce335org.highwire.dtl.DTLVardef@108e8e6org.highwire.dtl.DTLVardef@1f25bc0_HPS_FORMAT_FIGEXP M_FIG Graphical abstract C_FIG

cancer biology↗

bioRxiv2025

Tissue reassembly with generative AI

The spatial arrangement of cells is fundamental to their function, but single-cell RNA sequencing loses spatial context by dissociating cells from tissues. We present LUNA, a generative AI model that reassembles dissociated cells into tissue structures solely from gene expression by learning spatial priors from existing spatially resolved datasets. We apply and validate LUNA across multiple technologies including reconstructing the MERFISH whole mouse brain atlas with over 1.2 million cells, de novo reassembly of mouse central nervous system scRNA-seq atlas and inference of the spatial locations of nuclei lost during Slide-tags profiling. Furthermore, we show that LUNA generalizes to unseen cell types and to pathological samples in the Parkinsons disease mouse model, identifying regions of change due to pathology. We envision that AI-driven tissue reassembly can help to overcome current technological limitations and advance our understanding of tissue organization and function.

bioinformatics↗

bioRxiv2021

Dynamics of Functional Network Organization Through Graph Mixture Learning

Understanding the organizational principles of human brain activity at the systems level remains a major challenge in network neuroscience. Here, we introduce a fully data-driven approach based on graph learning to extract meaningful repeating network patterns from regionally-averaged time-courses. We use the Graph Laplacian Mixture Model (GLMM), a generative model that treats functional data as a collection of signals expressed on multiple underlying graphs. By exploiting covariance between activity of brain regions, these graphs can be learned without resorting to structural information. To validate the proposed technique, we first apply it to task fMRI with a known experimental paradigm. The probability of each graph to occur at each time-point is found to be consistent with the task timing, while the spatial patterns associated to each epoch of the task are in line with previously established activation patterns using classical regression analysis. We further on apply the technique to resting state data, which leads to extracted graphs that correspond to well-known brain functional activation patterns. The GLMM allows to learn graphs entirely from the functional activity that, in practice, turn out to reveal high degrees of similarity to the structural connectome. We compared similarity of the default mode network estimated from different task data and comparing them to each other and to structure. Using different metrics, a similar distinction between high- and low-level cognitive tasks arises. Overall, this method allows us to infer relevant functional brain networks without the need of structural connectome information. Moreover, we find that these networks correspond better to structure compared to traditional methods.

neuroscience↗