Cells can be altered by diseases or genetic mutations. However, the possibilities for finding out how cells are affected by various influences through laboratory experiments are limited. Due to the sheer number of possible combinations of treatment and disease conditions, expanding these data to characterize disease and disease treatment in traditional life science laboratories is labor intensive and costly and, hence, not scalable.
First machine learning model for predicting cell behavior
It is therefore a central goal of computational biology to use mathematical models to describe cellular responses to perturbations to make them accessible to numerical methods. Previous models were based on statistical and mechanistic approaches. With scGen, Fabian Theis, a professor of Mathematical Modelling of Biological Systems at TUM and the director of the Institute of Computational Biology at the Helmholtz Zentrum München, and his team have developed the first machine learning-based software tool in this area. The tool can be used to study high-dimensional phenomena for which no observations exist. It utilizes ideas from image, sequence and language processing to create a computer model of a cell's behavior.
Applying data from mice to human cells
Consequently, if scGen is trained with data capturing the effect of perturbations on a given system, the model can generate reliable predictions for another system of interest. The tool therefore harbors the potential for new approaches to the study and treatment of diseases. "For the first time, we have the opportunity to use data generated in one model system such as mouse and use the data to predict disease or therapy response in human patients," says Mohammad Lotfollahi, a PhD candidate at the TUM School of Life Sciences Weihenstephan.
scGen software to be expanded to address more complex problems
The participating researchers plan to improve the predictive power of scGen through further development work. This will also make it possible to compute combinations of factors affecting cells. "We can now start optimizing scGen to answer more and more complex questions about diseases," says Prof. Theis.
Theis, F.; Lotfollahi, M.; Wolf A.: scGen predicts single-cell perturbation responses. In: Nature Methods 16; 715-721 (2019). DOI 10.1038/s41592-019-0494-8
Contacts to this article:
Prof. Fabian Theis
Technical University of Munich
Chair of Mathematical Modeling of Biological Systems
Tel: +49 (89) 289 18386