We are driven by the belief that the spatial organization of tissue provides a powerful window into cell-cell interactions, a crucial component of disease progression and response. Yet, technical challenges have left this valuable source of information untapped: human comprehension can be overwhelmed by the complexity and size of individual tissue samples while purely automated approaches are hindered by the relative scarcity of independent tissue samples. By building approaches that leverage recent advances in machine learning within the framework established by biology, we hope to establish tissue organization itself as a predictive and investigative biomarker.
Research
1. Kidney Cancer
Kidney cancer is the poster child for intra-tumor heterogeneity. By applying machine learning to tumor tissue morphology we aim to understand the molecular basis of the heterogeneity, its evolutionary implications, and to design morphogenomic biomarkers of drug response that can overcome this heterogeneity.
2. Neurodegeneration
Protein aggregation is a hallmark of neurodegeneration, yet our understanding of what processes lead to their formation remains poor. We aim to understand how the morphology and location of protein aggregates in the brain can further our understanding of these diseases, connect their classical histopathological taxonomy to the emerging one based on protein conformations and ultimately lead to better stratification of patients.
3. Radiation Oncology
The standard of care for locally advanced rectal cancer typically consists of neoadjuvant radiotherapy (RT), chemotherapy and surgery, with the latter being the primary form of treatment. Recent evidence suggests that patients that demonstrate a complete response to the neoadjuvant RT can avoid surgery permanently. This work is motivated by the hypothesis that the analysis of early post-treatment tissue morphology, and specifically the change induced by neoadjuvant RT, could lead to the discovery of biomarkers that are predictive of treatment response.
4. Deep Learning for Histopathology
We develop novel deep learning algorithms to overcome machine learning challenges unique to histopathology: massive image size, widespread heterogeneity, lack of interpretability and non-biological staining variation.
Lab Members
JOIn US!
We believe that it is a question of when, not if, machine learning revolutionizes pathology. To get there, there are several, biological, experimental and computational challenges to be overcome. We are constantly on the lookout for smart and motivated scientists (graduate students or postdocs) to join us in this effort.
- Unique multi-disciplinary training: We are a machine learning lab at one of the top academic medical centers in the world, working on problems at the cutting edge of computation and biology. Do you want to learn how to leverage your machine learning background to tackle hard problems in biology? Or maybe you want to be trained in deep learning approaches that would scale up your pathology skills? Come talk to us.
- Flexible startup-up like environment: We are a young lab in a new department committed to a fast pace. We are not afraid to break the old rules of academia to in search of what works best for our science. We've adopted open office plans with no reserved offices to maximize interactions. We have the ability to hire scientific programmers so that we can focus on algorithms+biology.
- World-class computational facilities: the BioHPC and Bioinformatics Core Facility in our department provide the strong computational and analytical infrastructure necessary to build your projects.
- High-standard of living: We offer competitive compensation that rewards industry experience and fellowships in a city with low cost of living.
For a more formal description of what we are looking for please look at our Jobs page.