Biology subjects

Dimitrova, N.

Publications and source records attributed to Dimitrova, N..

5 recordsLinked to original sources

bioRxiv2026

Beyond Segmentation Accuracy: Evaluating Biological Fidelity in Spatial Pathology

Multiplex immunofluorescence (MxIF) resolves cellular identity, state, and spatial organization within tumor tissue, enabling higher-order representations of cancer ecosystems including immune infiltration, multicellular neighborhoods, and tissue niches. Yet these analyses depend on a deceptively simple first step: accurately identifying individual cells. Segmentation errors can alter marker quantification, cellular phenotypes, cell-cell relationships, and ultimately the spatial organization inferred from an image. Consequently, improvements in conventional segmentation benchmarks matter biologically only if they propagate to more faithful representations of the tumor microenvironment. We test this relationship directly using an expert-curated benchmark of 31,924 expert-reviewed cells on three slides under a leave-slide-out protocol. We compare a domain-specific self-supervised encoder, adapted to multiplex tissue imaging, with supervised, microscopy-specific, prompted generalist, and general-purpose vision models, and pass every method's predicted cells through an identical downstream pipeline. We evaluate fidelity across four levels: cell segmentation, phenotype assignment, spatial neighborhood structure, and tissue organization, comprising niches and tumor-immune interactions. The resulting hierarchy reveals that these levels do not improve in lockstep. Cellular phenotype assignment is comparatively robust to segmentation differences, whereas missed cells increasingly disrupt neighborhood structure and tumor-immune interactions. As a stringent spatial-interaction test, tumor-CD8+ contact recovery is poorly predicted by conventional segmentation quality (Panoptic Quality: {rho} = 0.18), while errors specifically at tumor-immune interfaces are strongly associated with contact fidelity ({rho} = -0.80). Methods with nearly identical segmentation accuracy can therefore produce substantially different biological representations. These results argue that cell segmentation for spatial pathology should be evaluated not only by boundary and instance accuracy, but also by preservation of the biological relationships that downstream analyses are intended to measure.

bioinformatics↗

bioRxiv2026

Discovering Latent Scientific Concepts through Discrete Representation Learning

Scientific imaging is undergoing a fundamental transformation. Advances in multiplex immunofluorescence, computational pathology, and electron microscopy now enable observation of biological and physical systems at unprecedented spatial resolution. Rather than simply classifying known objects, these data offer the opportunity to discover previously unknown organizational principles. However, scientific image datasets are typically highly curated, information-rich, and limited in size, making reliable representation learning particularly challenging. We show that standard continuous self-supervised adaptation can become unstable in this regime, allowing teacher representations to drift and degrading downstream transfer. We introduce Kodiak, a discrete representation learning architecture that discovers latent scientific concepts through Sinkhorn-balanced codebook assignments. Instead of learning unconstrained continuous embeddings, Kodiak constructs a reusable vocabulary of discrete concepts that provides stable supervision for both masked-patch prediction and cross-view representation learning within a single-stage framework. Across five scientific imaging domains, including multiplex immunofluorescence for pancreatic cancer pathology and scanning electron microscopy, Kodiak consistently improves representation quality under both continued foundation-model adaptation and training from scratch. Beyond improved classification and transfer,the learned concepts are interpretable, reusable, and produce spatially coherent concept maps that support applications such as cell phenotyping, segmentation-free patch phenotyping, spatial niche discovery, tissue organization, and materials characterization. We view Kodiak as the representation-learning engine of our foundation model for spatial biology. More broadly, our results suggest that discrete representation learning provides a principled foundation for discovering reusable latent scientific concepts that can accelerate scientific interpretation across imaging domains.

cancer biology↗

bioRxiv2024

Metastatic site influences driver gene function in pancreatic cancer

Driver gene mutations can increase the metastatic potential of the primary tumor1-3, but their role in sustaining tumor growth at metastatic sites is poorly understood. A paradigm of such mutations is inactivation of SMAD4 - a transcriptional effector of TGF{beta} signaling - which is a hallmark of multiple gastrointestinal malignancies4,5. SMAD4 inactivation mediates TGF{beta}s remarkable anti-to pro-tumorigenic switch during cancer progression and can thus influence both tumor initiation and metastasis6-14. To determine whether metastatic tumors remain dependent on SMAD4 inactivation, we developed a mouse model of pancreatic ductal adenocarcinoma (PDAC) that enables Smad4 depletion in the pre-malignant pancreas and subsequent Smad4 reactivation in established metastases. As expected, Smad4 inactivation facilitated the formation of primary tumors that eventually colonized the liver and lungs. By contrast, Smad4 reactivation in metastatic disease had strikingly opposite effects depending on the tumors organ of residence: suppression of liver metastases and promotion of lung metastases. Integrative multiomic analysis revealed organ-specific differences in the tumor cells epigenomic state, whereby the liver and lungs harbored chromatin programs respectively dominated by the KLF and RUNX developmental transcription factors, with Klf4 depletion being sufficient to reverse Smad4s tumor-suppressive activity in liver metastases. Our results show how epigenetic states favored by the organ of residence can influence the function of driver genes in metastatic tumors. This organ-specific gene-chromatin interplay invites consideration of anatomical site in the interpretation of tumor genetics, with implications for the therapeutic targeting of metastatic disease.

cancer biology↗

bioRxiv2023

Overexpressed Malat1 Drives Metastasis through Inflammatory Reprogramming of Lung Adenocarcinoma Microenvironment

Metastasis is the main cause of cancer deaths but the molecular events leading to metastatic dissemination remain incompletely understood. Despite reports linking aberrant expression of long noncoding RNAs (lncRNAs) with increased metastatic incidence, in vivo evidence establishing driver roles for lncRNAs in metastatic progression is lacking. Here, we report that overexpression of the metastasis-associated lncRNA Malat1 (metastasis-associated lung adenocarcinoma transcript 1) in the autochthonous K-ras/p53 mouse model of lung adenocarcinoma (LUAD) is sufficient to drive cancer progression and metastatic dissemination. We show that increased expression of endogenous Malat1 RNA cooperates with p53 loss to promote widespread LUAD progression to a poorly differentiated, invasive, and metastatic disease. Mechanistically, we observe that Malat1 overexpression leads to the inappropriate transcription and paracrine secretion of the inflammatory cytokine, Ccl2, to augment the mobility of tumor and stromal cells in vitro and to trigger inflammatory responses in the tumor microenvironment in vivo. Notably, Ccl2 blockade fully reverses cellular and organismal phenotypes of Malat1 overexpression. We propose that Malat1 overexpression in advanced tumors activates Ccl2 signaling to reprogram the tumor microenvironment to an inflammatory and pro-metastatic state.

cancer biology↗

bioRxiv2020

Global shortfalls in extinction risk assessments for endemic flora

The Global Strategy for Plant Conservation (GSPC) ambitiously calls for an assessment of extinction risk for all recognised plant taxa by 20201. It is now clear that this target will not be met in the short-term; only 21-26% of known plant species have been assessed2 - a monumental shortfall in anticipated knowledge. Yet the need for risk assessments has never been more urgent. Plants are rapidly going extinct3,4 and face threats such as climate change5 and permanent deforestation6. Extinction risk assessments continue to provide the critical foundation to inform protection, management and recovery of plant species7,8, the loss of which will have clear consequences for maintaining planetary systems and human well-being9. Here, we rank countries of the world based on progress towards assessing the extinction risk to their endemic flora. Overall, 67% of country-based endemic species do not have an extinction risk assessment completed (143,294 species). We show that some of the worlds wealthiest nations, which also have relatively strong species protections, are failing to protect their unique flora by not systematically assessing risks to their endemic species.

ecology↗