Biomedical science has reached the era of ‘Big Data’ now allowing to better shape some of the main challenges that still limits the success of many clinical approaches in the oncology field: e.g. role of the microenvironment in tumor progression, patient heterogeneity, mode-of-action of immunotherapy or resistance to treatment. For these concepts understanding spatial-temporal organization of the tumors is key and thus requires of visualization techniques at single cell level: (live) microscopy. Modern advances in optical technologies now allow to generate complex imaging data, however the lack of appropriate tools or experts to process or analyze it, often leaves it unexplored. In our lab (imAIgene-lab) we utilize modern Artificial Intelligence tools from distinct fields of computer vision, single cell technologies and data analytics to transform (dynamic) imaging data into meaningful information that can be exploited to study e.g. drug response prediction, combinatorial treatment in immune-oncology and personalized artificial-intelligence based diagnostics and treatment.
Our computational developments are directed to solving three challenges of oncology:
- Understanding the mode-of-action of T cell immunotherapy against solid tumors to overcome resistance.
Despite the revolutionary success of cancer T-Cell Immunotherapies their efficacy in treating solid tumours, is still very limited. Cancer heterogeneity between patients, and even within the same tumour, challenges the achievement of consistent clinical outcomes and has triggered the development of a multitude of T cell-based therapies using sophisticated engineering approaches: CART, bi-specific antibodies, TCR based, metabolome sensing. However, these widely exploited ‘living’ drugs are inherently dynamic and utilize different cellular strategies to achieve tumour targeting that cannot be analysed through a single snapshot readout. Dynamic live imaging can capture complex cellular function that spans overtime, explore functional differences between T cells, and define cellular events leading to tumour killing. In this research line we exploit computer vision and analytical tools to understand how T cell achieve successful killing, or different modes of disfunction and their relation to tumor resistance.
- Unravelling the role of the tumor microenvironment in tumor invasive behavior and resistance to immunotherapy in Diffuse Midline Glioma.
Modern technologies allow to study the distinct aspects involved in tumor progression from different angles, such as in vivo and ex vivo time-lapse imaging (e.g. intravital microscopy) for tumor cell invasion studies; multiplexed imaging for TME characterization or sc transcriptomics to quantitively dissect tumor heterogeneity and complexity. Separately each of these single omic approaches can mainly reflect one aspect of tumor biology, thus computational integrative approaches are necessary to acquire systematic understanding of how these different aspects interact define predictive paths to invasion and provide workflows to bridge phenotypic, molecular and contextual networks driving this process. To identify tumor intrinsic and extrinsic drivers of tumor invasion in this research line we develop a computational framework for multi-omic integration of three complementary layers: tumor cell dynamics; TME and tumor transcriptome. With this framework we aim to get a better understanding on the biology of Diffuse Midline Glioma, an invasive pediatric brain tumor with a very dire prognosis.
- Increase the depth of tumor spatial phenotyping with virtual multiplexing
Single-cell resolution volumetric imaging is an uprising technique in the biomedical research as it permits to generate cellular maps of human healthy and tumoral tissues that can widen our knowledge of a disease. One of the main limitations of volumetric imaging is the amount of markers that can be imaged simultaneously, a desired outcome to fully assess the composition of a tissue. To unravel the highly complex spatial cellular organization within (cancer) tissue it is necessary to further expand the amount of cell types that can be resolved with imaging approaches. In this research line we make use of the most recent advances in deep learning computer vision methods to perform virtual multiplexing, even with most simple microscope systems that are usually limited to 3-4 markers.