Inferring tumor evolution from single-cell genomes

Series on Tumor Evolution

Everything is better if you do it with a Nested Effects Model – even inferring tumor evolution.

Let me introduce to you Oncogenetic Nested Effects Models, or for short OncoNEMs, which we just published in the new Single Cell collection of Genome Biology (see here). They exploit the fact that tumors accumulate mutations while they evolve, which leads to (noisy) subset relations between clones – exactly the type of pattern NEMs were made for.

We are interested in clonal evolution and the individual cells for us are just markers or representatives of the clones. To reconstruct a clonal phylogeny we thus need to cluster the cells into clones, while at the same time infer a tree connecting the clusters.

The major challenges to conquer on the way are the high noise levels in single cell data and the fact that some of the clones might not have been sampled, which means that the tree can contain populated and unpopulated nodes.

Here is an overview of our model:

Screen Shot 2016-04-18 at 15.57.48.png

The grey circles are clones, linked to each other in a phylogenetic tree. Each clone can contain cells (colored circles), with the top clone corresponding to normal cells (green) and all the others to cancer cells (red). The boxes at the bottom correspond to the observed mutations and they are linked to the clone in which they appear first.

So how do we do the inference? Look at these figures:

Screen Shot 2016-04-18 at 15.58.17.png

If we knew the model, we would know for each cell which mutations it contains: the ones from the clone it belongs to and from its parent clones all the way to the normal (B). We can thus write down the expected genotypes of all cells under the model (C). Comparing these expectations to the observed data we can count how many differences (false positives and false negatives) we have under this model (D). We compute the likelihood for each model analogous to a ‘traditional’ NEM and the model leading to the least amount of differences will win.

Simple enough. Now just add a search algorithm to walk through the space of possible models and we are all set.

You can find all the naughty details, simulations studies, etc here:



One thought on “Inferring tumor evolution from single-cell genomes

You gotta talk to me!

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s