Spatial transcriptomics makes it possible to determine the full transcriptome,
that this, the expression of all messanger RNAs of cells, at defined spatial
location in tissue sections. This revolutionary technology effectively unites the
century-old tradition of tissue section observation under the microscope with
XXI st century genomics. We are now poised to understand the biological
processes that generate the microscopic morphologies that define organs in
health and disease. Indeed, most biological functions and tissue states can be
inferred from transcriptomic profiles.

While transcriptomic data analysis has yield major discoveries in the
last two decade, apprehending it in the context of space and tissue
morphology is completely new. Moreover, how to characterize quantitatively
tissue morphology is also an open topic with implication for both fundamental
research and diagnostic. The thesis will address both problems in the context
of cancer research.

First, using deep learning auto-encoders thousands of images patches
will be clustered in a low dimension latent space that abstracts-out irrelevant
noise and features such as tissue orientation, etc. Second, another deep
network will be trained to associate these latent space representations with
the corresponding transcriptomic profiles.
This interdisciplinary work will be supervised by researchers from LISA
(image analysis) and IRIBHM (bioinformatics and oncology, project initiator).

contacts:

Olivier Debeir, LISA (Olivier.Debeir@ulb.ac.be)
Vincent Detours, IRIBHM (vdetours@ulb.ac.be)