Skip to main content

Gaussian process models of pseudotime and branching in single-cell gene expression experiments

Speaker:  Magnus Rattray, Manchester


In single-cell gene expressions experiments each cell may be at a different point in some dynamic process but the time information for each cell is not available. Pseudotime methods seek to infer time from these high-dimensional and noisy data points. We are developing methods for inference of pseudotime and branching dynamics in single-cell gene expression data. We use Gaussian processes, which allow for uncertainty in pseudotime inference and provide a natural prior over branching processes. To make inference tractable we have implemented methods using the GPflow/Tensorflow package, which allows for efficient inference through gradient-based optimisation of variational marginal likelihoods.


Sumon Ahmed, Magnus Rattray, Alexis Boukouvalas "GrandPrix: Scaling up the Bayesian GPLVM for single-cell data” bioRxiv 227843; doi:

Alexis Boukouvalas, James Hensman, Magnus Rattray "BGP: Branched Gaussian processes for identifying gene-specific branching dynamics in single cell data” bioRxiv 166868; doi:

And the dates of future statistics seminars/events:

16 February (1pm): Elisa Bellotti, Manchester
2 March (1pm): Sara Wade, Warwick

7 June (10-5pm): Workshop on Statistical Network Science. The webpage is now up and you can register (for free) at: