Background

Revolutions in technology have exponentially increased our ability to generate genomic information, resulting in a wealth of data waiting to be explored. Machine learning can be used to uncover valuable information from these datasets. This allows us to gain insights into the connections between genetic and epigenetic changes on molecular traits and has wide-ranging implications in health and disease. However, the predictive models that are used can widely vary, existing in diverse formats across varying locations. The lack of a model zoo (central repository) makes it difficult for the research community to utilize and build on existing models.

Project

Integrative machine learning to decipher genome function

A collaboration between the Stegle lab at EMBL, the Kundaje lab at Stanford and the Gagneur lab in Munich, we have recently established Kipoi – the first repository of machine learning models in regulatory genomics. Within Kipoi, models can readily be applied to interpret human disease variations and to address basic biological questions. Users can apply existing models to their own datasets or use them to derive new models and share these with the rest of the Kipoi community.  Our collaborative effort between EMBL and Stanford aims to advance the state of the art in this field by leveraging the joint expertise at both sites, as well as the community as a whole.
 

Publications:

^{The Kipoi repository accelerates community exchange and reuse of predictive models for genomics. Z Avsec, R Kreuzhuber, J Israeli, N Xu, J Cheng, A Shrikumar, A Banerjee, DS Kim, T Beier, L Urban, A Kundaje, O Stegle, JGagneur. Nature biotechnology (2019).}

Find out more:

Interested in finding out more about the machine learning and genomics? Do you want to contribute to this great resource for scientists? Get in touch, we would love to hear from you!