The rise of antibody-drug conjugates (ADCs) marks a turning point in the treatment of solid tumors, offering new hope through targeted delivery of chemotherapy. Yet, many patients develop resistance or experience serious toxicities, underscoring the urgent need for more personalized approaches. The European OASIS project addresses this challenge by combining advanced biomedical technologies with artificial intelligence to better understand resistance mechanisms and guide treatment decisions. Coordinated by Gustave Roussy and uniting multiple European research and industrial partners, this collaborative initiative leverages diverse expertise across institutions. Our research develops AI-powered models for multimodal molecular profiling, integrating bulk omics, spatial omics, and imaging data. By capturing information across biological scales—from molecular alterations to tissue architecture—we aim to identify predictive biomarkers that can support clinical decisions and accelerate the development of next-generation ADCs. This work advances precision oncology by guiding treatment selection through a data-driven understanding of tumor biology. By leveraging machine learning to analyze heterogeneous data, we help clinicians match patients with appropriate therapies, improve outcomes, and minimize side effects. This interdisciplinary approach connects AI, cancer biology, pathology, and clinical research to deliver practical tools—such as the OASIS Score—that translate complex molecular data into actionable insights for personalized cancer care.
Recent breakthroughs in artificial intelligence are transforming biomedical research and healthcare through deep learning models that extract clinically actionable insights from complex, high-dimensional data. This project develops novel deep learning approaches tailored to biological data analysis across multiple resolution levels—from bulk multi-omics to single-cell and spatially-resolved profiles—to advance precision medicine. At the bulk omics level, we will identify molecular signatures associated with disease subtypes, resistance mechanisms, and prognosis. For single-cell analysis, our deep learning models will annotate cellular populations and correct for batch effects across datasets and technologies, capturing the heterogeneity underlying treatment response. At the spatial level, we will integrate spatial transcriptomics with histopathology images to uncover tissue architecture, spatial domains, and cellular interactions. This approach examines how deep learning can drive discovery and clinical insight across multiple scales rather than focusing on a single modality. Applications include patient stratification, identification of predictive biomarkers, and improved understanding of disease mechanisms across cancers and other complex diseases. The project aims to bridge the gap between cutting-edge computational methods and real-world clinical impact, supporting the development of personalized therapeutic strategies.
The emergence of circulating tumor DNA (ctDNA) as a non-invasive biomarker is reshaping the landscape of cancer prognosis and monitoring. However, ctDNA analysis remains challenging due to its low abundance, fragmentation patterns, and the high background of non-tumor DNA. This project develops novel, explainable, and clinically applicable deep learning models for ctDNA analysis to enable disease progression assessment, minimal residual disease tracking, and personalized treatment monitoring across various cancer types. By utilizing attention-based models, representation learning, and generative approaches, we will design algorithms capable of detecting weak tumor-derived signals from shallow or noisy sequencing data. These models will incorporate multi-dimensional features—including mutation patterns, fragmentomics, and epigenetic signatures—to enhance sensitivity and specificity, even at ultra-low variant allele frequencies. This work combines machine learning, liquid biopsy technologies, and oncology to translate complex sequencing signals into actionable insights for prognostic evaluation and improved patient outcomes. Through collaboration with clinical and molecular teams, we will develop tools that address real-world needs and can integrate into prognostic and monitoring workflows. This project advances data-driven precision medicine through AI-powered molecular profiling to support timely and individualized care.