Project Ex Vivo

Project Ex Vivo is a joint cancer research collaboration between Microsoft and the Broad Institute, with support from the Dana-Farber Cancer Institute. By synergizing each other’s world-leading expertise in AI, high-throughput biology, and clinical oncology, we are building a true end-to-end framework for precision oncology: from computation, to wet-lab experimentation, to the clinic, and back again.

Understanding and treating cancer precisely demands a new way of thinking

Cancer is understood primarily as a disease of DNA alterations and often treated using a reductionist approach, with each cancer distilled to a single metric such as histologic type or harboring a specific molecular abnormality. As a result, the current paradigm for precision oncology consists primarily of using genetic alterations to direct therapy (e.g., targeting tumor mutations). Strikingly, few patients receive prolonged benefit from this approach.

Project Ex Vivo envisions a new, constructionist paradigm for precision oncology, one powered by the bottom-up integration of experimentation and computation. We see cancers as complex (eco)systems, beyond just mutational variation, that necessitate systems-level understanding and intervention. Our ultimate objective is to more effectively model cancer ex vivo – outside the body – in a patient-specific manner. Doing so will unlock the ability to more effectively stratify patients and identify therapies that target diverse aspects of human cancers.

Defining, engineering, and targeting cell states

Learning cell state and tumor microenvironment archetypes

We are developing AI models to map the expression states and spatial interactions of the in vivo tumor microenvironment. To this end, we are generating new single-cell and spatial transcriptomics datasets of tumor specimens from a variety of cancer subtypes.

Optimizing ex vivo models

Today’s lab model systems of cancer, including cell lines and patient-derived spheroid or organoid models, are incomplete and often unfaithful representations of in vivo biology. We are building new experimental methods to scale, engineer, and test ex vivo models that more faithfully simulate in vivo tumor biology. By measuring and mapping how cell states across these models relate to function, we are creating first-in-class datasets that will provide a basis for optimization of faithful, personalized ex vivo models of cancer.

Designing state-specific precision therapies

A learned map of in vivo and ex vivo cell states provides a new therapeutic search space. Using our data on how cell states map to function across cancers, we are developing computational methods to infer state-specific vulnerabilities that could be targeted therapeutically. This enables us to nominate new therapeutic approaches that target specific cell states. Our ultimate vision is to work towards a new paradigm that integrates genetic and non-genetic attributes for therapy selection in precision oncology.

Our vision for clinical impact

Our initial work in pancreatic ductal adenocarcinoma (opens in new tab) (PDAC) uncovered limitations in current cancer modeling pipelines and revealed fundamental insights into the therapeutic significance of cell state and its microenvironmental drivers.

Project Ex Vivo will codify new pipelines and workflows to establish non-genetic attributes (e.g., cell state) as targetable features in cancer, with an initial focus on pancreatic cancer and eventually extending to additional malignancies. We develop AI models to learn accurate representations of each tumor and cellular model by integrating genetic markers, expression state, and tumor microenvironmental interactions. We will optimize model generation workflows for greater in vivo fidelity, nominate new candidates for therapeutic development, and establish a new paradigm integrating genetic and non-genetic attributes to guide therapy selection in precision oncology. We are developing rigorous experimental, computational, and automation pipelines that can be applied across a variety of cancer, non-cancer, and clinical contexts. By credentialing cell state and its relation to the local environment as a biomarker for both prognostication and therapy selection, the success of this project may improve patient stratification for diverse clinical trials (chemotherapy, immunotherapy, small molecules, etc.), thereby enhancing the likelihood of therapeutic efficacy, potentially lowering costs, and ultimately improving patient outcomes.