Biology subjects

Durand, E. Y.

Publications and source records attributed to Durand, E. Y..

9 recordsLinked to original sources

bioRxiv2025

Multiple instance learning with spatial transcriptomics for interpretable patient-level predictions: application in glioblastoma

Accurate prediction of patient outcomes remains a major challenge in oncology. While recent machine learning (ML) approaches often rely on bulk omics lacking spatial resolution or histology-based multiple instance learning (MIL), spatial transcriptomics (SpT) provides a unique opportunity to capture both molecular content and tissue architecture. However, no generalizable ML framework has yet been established to exploit SpT for patient-level outcomes. We present SpaMIL, a flexible and interpretable MIL framework designed for SpT, with a distillation strategy that enables deployment for hematoxylin and eosin (H&E) slides alone. We evaluate the framework by predicting survival from glioblastoma (GBM) patients, a clinically compelling setting given its aggressiveness with a median survival of only 15 months and the lack of prognostic clinical variables. We analyzed 76 GBM cases from the MOSAIC dataset: 43 with matched SpT, H&E, single-nucleus RNA-seq (scRNA-seq), bulk RNA-seq, and clinical variables, and 33 with H&E for external validation. We developed two main architectures: abMIL, tailored to SpTs spatial molecular structure, and MabMIL, which distills SpT-derived representations into H&E. Model interpretability was achieved through a Shapley-based framework linking prognostic predictions to cell-type compositions via SpT deconvolution. In benchmarking across the five GBM MOSAIC modalities, SpT-based abMIL achieved unprecedented prognostic accuracy (median C-index: 0.72, standard deviation: 0.04), outperforming all other modalities, including established clinical predictors. PCA and deconvolution-based SpT representations surpassed recent foundation models, suggesting the need for further research on SpT foundation models. Our interpretability analysis highlighted malignant and non-malignant cell subpopulations associated with favorable or poor prognosis, consistent with recent reports. Finally, MabMIL maintained strong performance while enabling H&E-only deployment, with improved condorance index over H&E-only baselines in both internal (0.59 vs. 0.57) and external (0.62 vs. 0.55) cohorts.

cancer biology↗

bioRxiv2025

Histology and spatial transcriptomic integration revealed infiltration zone with specific cell composition as a prognostic hotspot in glioblastoma

BackgroundGlioblastoma (GBM), the most aggressive primary brain tumor, has a median survival of approximately 15 months. Twenty percent of patients survive beyond three years, but known clinical factors like age, performance status, resection extent, and MGMT promoter methylation status do not fully explain the observed outcomes. ObjectiveOur objective was to identify novel histology derived biomarkers associated with end-of-spectrum overall survival (OS) to provide novel biological insight with a translational potential. MethodsWe analyzed a total of 748 GBM patients from 3 different cohorts, uniquely enriched in long survivors (n=98 with overall survival (OS) > 5y including n=196 with OS[≥]3y), with clinical data and H&E slides obtained from the primary tumor at baseline. We propose an interpretable machine learning (ML) methodology for the discovery of histological biomarkers. Our method learned to segment each H&E slide into three distinct regions associated with long-term survival, short-term survival, and non-informative tissue. We characterized these regions by integrating unsupervised learning, nuclei segmentation, blood vessels detection, pathologist annotations, and multimodal data including spatial transcriptomics from n=31 patients of the GBM MOSAIC dataset to discover fully interpretable biomarkers. ResultsOur OS prediction model using histology and clinical data as input achieved an area under the curve (AUC) of 0.85 for the classification of patients between OS<2 and OS[≥]3y in external cohort validation, outperforming significantly models trained on clinical data or on histology alone (AUC of 0.76; 0.73, respectively). Two novel biomarkers were predicting poor survival: the presence of regions of lowly infiltrated white matter enriched in malignant cells with a mesenchymal-like phenotype, and lower levels of angiogenesis associated with higher hypoxia response in the main tumor regions. We also found that a subtype of immunosuppressive tumor macrophages - defined by high PLIN2 expression and lipid accumulation- is consistently enriched in histological areas predictive of poor prognosis. ConclusionOur interpretable ML methodology identified a novel prognostic impact of biological processes and cell types according to distinct tumor regions of GBM. These results pave the way for spatially-informed biomarkers to improve risk stratification and for personalized spatially-targeted therapeutic strategies. Key highlightsO_LIOur ML model identified histological biomarkers predicting prognosis independently from known clinical factors C_LIO_LIThe region of lowly infiltrated white matter enriched in malignant cells including a mesenchymal-like phenotype is predictive of poor prognosis C_LIO_LIAngiogenesis is increased in areas predictive of long survival in main non-necrotic tumor regions. C_LIO_LIThe subtype of macrophages expressing PLIN2 and associated with increased lipid metabolism was associated with poor prognosis in all GBM regions. C_LI Highlights O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=116 SRC="FIGDIR/small/681087v4_ufig1.gif" ALT="Figure 1"> View larger version (43K): org.highwire.dtl.DTLVardef@45e619org.highwire.dtl.DTLVardef@105a4b0org.highwire.dtl.DTLVardef@17f257dorg.highwire.dtl.DTLVardef@7669b4_HPS_FORMAT_FIGEXP M_FIG C_FIG

cancer biology↗

bioRxiv2024

Transcriptome Analysis of Archived Tumor Tissues by Visium, GeoMx DSP, and Chromium Methods Reveals Inter- and Intra-Patient Heterogeneity

Recent advancements in probe-based, full-transcriptome, high-resolution technologies for Formalin-Fixed Paraffin-Embedded (FFPE) tissues, such as Visium CytAssist, Chromium Flex (10X Genomics), and GeoMx DSP (Nanostring), have opened new opportunities for studying decades-old archival samples in biobanks, facilitating the generation of data from extensive cohorts. However, the experimental protocols can be labor-intensive and costly; therefore, it is thus essential for researchers to carefully evaluate the strengths and limitations of each technology in relation to their specific research objectives. Here, we report the results of a comparative analysis of the three methods mentioned above on FFPE archival tumor samples from four non-small cell lung cancer, four breast cancer and six diffuse large B-cell lymphoma. We highlight some relative advantages and disadvantages of each method in the context of operational challenges, bioinformatic analysis and biological discovery. Our results show that: 1) all three methods yielded good-quality, highly reproducible transcriptomic data from serial sections of the same FFPE block; 2) GeoMx data contained mixtures of cell types, even when pre-selecting areas with cell type-specific markers; 3) high-throughput spot-level (Visium) or cell-level (Chromium) data enabled the identification of tumor heterogeneity within and between patients, which could be used to identify targeted therapies. Our data support the use of Visium and Chromium for high-throughput and discovery-driven projects, while the GeoMx platform could be suited for addressing specialized questions on targeted regions. All data generated from this study, including GeoMx, Visium, Chromium, H&E, and expert annotations are publicly available.

bioinformatics↗

bioRxiv2024

A deep learning-based multiscale integration of spatial omics with tumor morphology.

Spatial Transcriptomics (spTx) offers unprecedented insights into the spatial arrangement of the tumor microenvironment, tumor initiation/progression and identification of new therapeutic target candidates. However, spTx remains complex and unlikely to be routinely used in the near future. Hematoxylin and eosin (H&E) stained histological slides, on the other hand, are routinely generated for a large fraction of cancer patients. Here, we present a novel deep learning-based approach for multiscale integration of spTx with tumor morphology (MISO). We trained MISO to predict spTx from H&E on a new unpublished dataset of 72 10X Genomics Visium samples, and derived a novel estimate of the upper bound on the achievable performance. We demonstrate that MISO enables near single-cell-resolution, spatially-resolved gene expression prediction from H&E. In addition, MISO provides an effective patient representation framework that enables downstream predictive tasks such as molecular phenotyping or MSI prediction.

bioinformatics↗

bioRxiv2024

Deep learning uncovers histological patterns of YAP1/TEAD activity related to disease aggressiveness in cancer patients.

Over the last decade, Hippo signaling has emerged as a major tumor-suppressing pathway. Its dysregulation is associated with abnormal expression of YAP1 and TEAD-family genes. Recent works have highlighted the role of YAP1/TEAD activity in several cancers and its potential therapeutic implications. Therefore, identifying patients with a dysregulated Hippo pathway is key to enhancing treatment impact. Although recent studies have derived RNAseq-based signatures, there remains a need for a reproducible and cost-effective method to measure the pathway activation. In recent years, deep learning applied to histology slides have emerged as an effective way to predict molecular information from a data modality available in clinical routine. Here, we trained models to predict YAP1/TEAD activity from H&E-stained histology slides in multiple cancers. The robustness of our approach was assessed in seven independent validation cohorts. Finally, we showed that histological markers of disease aggressiveness were associated with dysfunctional Hippo signaling.

bioinformatics↗

bioRxiv2024

Robust Evaluation of Deep Learning-based Representation Methods for Survival and Gene Essentiality Prediction on Bulk RNA-seq Data

Deep learning (DL) has shown potential to provide powerful representations of bulk RNA-seq data in cancer research. However, there is no consensus regarding the impact of design choices of DL approaches on the performance of the learned representation, including the model architecture, the training methodology and the various hyperparameters. To address this problem, we evaluate the performance of various design choices of DL representation learning methods using TCGA and DepMap pan-cancer datasets, and assess their predictive power for survival and gene essentiality predictions. We demonstrate that non DL-based baseline methods achieve comparable or superior performance compared to more complex models on survival predictions tasks. DL representation methods, however, are the most efficient to predict the gene essentiality of cell lines. We show that auto-encoders (AE) are consistently improved by techniques such as masking and multi-head training. Our results suggest that the impact of DL representations and of pre-training are highly task- and architecture-dependent, highlighting the need for adopting rigorous evaluation guidelines. These guidelines for robust evaluation are implemented in a pipeline made available to the research community.

bioinformatics↗

bioRxiv2022

Biologically relevant integration of transcriptomics profiles from cancer cell lines, patient-derived xenografts and clinical tumors using deep neural networks

Cell lines and patient-derived xenografts are essential to cancer research, however, the results derived from such models often lack clinical translatability, as these models do not fully recapitulate the complex cancer biology. It is critically important to better understand the systematic differences between cell lines, xenografts and clinical tumors, and to be able to identify pre-clinical models that sufficiently resemble the biological characteristics of clinical tumors across different cancers. On another side, direct comparison of transcriptional profiles from pre-clinical models and clinical tumors is infeasible due to the mixture of technical artifacts and inherent biological signals. To address these challenges, we developed MOBER, Multi-Origin Batch Effect Remover method, to simultaneously extract biologically meaningful embeddings and remove batch effects from transcriptomic datasets of different origin. MOBER consists of two neural networks: conditional variational autoencoder and source discriminator neural network that is trained in adversarial fashion. We applied MOBER on transcriptional profiles from 932 cancer cell lines, 434 patient-derived tumor xenografts and 11159 clinical tumors and identified pre-clinical models with greatest transcriptional fidelity to clinical tumors, and models that are transcriptionally unrepresentative of their respective clinical tumors. MOBER can conserve the biological signals from the original datasets, while generating embeddings that do not encode confounder information. In addition, it allows for transformation of transcriptional profiles of pre-clinical models to resemble the ones of clinical tumors, and therefore can be used to improve the clinical translation of insights gained from pre-clinical models. As a batch effect removal method, MOBER can be applied widely to transcriptomics datasets of different origin, allowing for integration of multiple datasets simultaneously.

bioinformatics↗

bioRxiv2021

A scalable pipeline for local ancestry inference using tens of thousands of reference haplotypes

Ancestry deconvolution is the task of identifying the ancestral origins of chromosomal segments of admixed individuals. It has important applications, from mapping disease genes to identifying loci potentially under natural selection. However, most existing methods are limited to a small number of ancestral populations and are unsuitable for large-scale applications. In this article, we describe Ancestry Composition, a modular pipeline for accurate and efficient ancestry deconvolution. In the first stage, a string-kernel support-vector-machines classifier assigns provisional ancestry labels to short statistically phased genomic segments. In the second stage, an autoregressive pair hidden Markov model corrects phasing errors, smooths local ancestry estimates, and computes confidence scores. Using publicly available datasets and more than 12,000 individuals from the customer database of the personal genetics company, 23andMe, Inc., we have constructed a reference panel containing more than 14,000 unrelated individuals of unadmixed ancestry. We used principal components analysis (PCA) and uniform manifold approximation and projection (UMAP) to identify genetic clusters and define 45 distinct reference populations upon which to train our method. In cross-validation experiments, Ancestry Composition achieves high precision and recall.

genomics↗

bioRxiv2020

Fully unsupervised deep mode of action learning for phenotyping high-content cellular images

The identification and discovery of phenotypes from high content screening (HCS) images is a challenging task. Earlier works use image analysis pipelines to extract biological features, supervised training methods or generate features with neural networks pretrained on non-cellular images. We introduce a novel fully unsupervised deep learning algorithm to cluster cellular images with similar Mode-of-Action together using only the images pixel intensity values as input. The method outperforms existing approaches on the labelled subset of the BBBC021 dataset and achieves an accuracy of 97.09% for correctly classifying the Mode-of-Action (MOA) by nearest neighbors matching. One unique aspect of the approach is that it is able to perform training on the entire unannotated dataset, to correctly cluster similar treatments beyond the annotated subset of the dataset and can be used for novel MOA discovery.

bioinformatics↗