Kundaje Lab @ Stanford University
We are computational biologists at Stanford University ๐ฒ
๐งฌ Our overarching goal is to decipher the genetic basis of disease, particularly through the lens of gene regulation. We believe that a genome-wide and system-level understanding of gene regulation is essential to decipher the causal genetic and molecular bases of disease.
๐ป Our approach for achieving this is to develop statistical and machine learning models fit to large-scale genomic data. What makes our models powerful lies in the mindset that we apply when developing them. They are designed to be:
- ๐ mechanistic. Our models are designed to capture the underlying biological processes, rather than merely learning input-output relationships. This makes them causal, rather than correlative, meaning that we can use them to predict the effects of perturbations to the system.
- ๐ interpretable. To us, a biological model is not just a black box. We have developed interpretation techniques to interrogate the model regarding the biological rules that it has learned.
- โ๏ธ simple. We follow the principle of Occamโs razor, which states that the simplest explanation is usually the best. We strive to develop models that are as simple as possible, while still being able to explain the data. This makes our models more robust and easier to interpret.
- ๐ช hierarchical. Mirroring the hierarchical nature of biological systems, our models build up from simple rules to more complex ones. This allows us to capture the complexity of biological systems without sacrificing interpretability.
๐ฎ After training these models on genomic data, we use them as to generate and test novel biological hypotheses in silico. We also use these models to identify potential therapeutic targets for disease.
๐งช We collaborate extensively with experimental biologists within and outside Stanford to validate the hypotheses generated by our models and discover new biology.