Home

Benchmarked approaches for cell lineage reconstructions from in vitro dividing cells and in silico models of Caenorhabditis elegans and Mus musculus development.

The Allen Institute lineage reconstruction DREAM challenge aimed to provide a new perspective on lineage inference by enabling participants from diverse fields to submit their reconstruction of trees for which the ground truth, i.e. the actual lineage, was not provided. It consisted of three challenges with lineages of increasing numbers of cells.

• The first challenge leveraged a then unpublished experimental dataset of 106 trees recorded with intMEMOIR in mouse embryonic stem cell colonies of less than 100 cells. This technique was chosen as it has the key advantage of readout by imaging which can be coupled with a time-lapse movie of the cells as they divide to provide a ground truth lineage tree.
• In the second challenge participants had to reconstruct an in silico tree of 1,000 cells, whose topology was derived from the Caenorhabditis elegans developmental cell lineage tree by removing a few clades in order to mask its identity to the participants. The resulting topology was used to simulate mutations in a recording array using a general framework for simulation of CRISPR-based lineage recording.
• In the third challenge, participants had to infer the lineage of cells in a simulated tree of ~10,000 cells representing 11 different cell types after one year of M. musculus development. Simulating such a large tree was possible by applying the Environment-dependent Stochastic tree Grammars (eSTGt), a programming and simulation environment for population dynamics adapted to simulate cell lineages . While the size of the actual simulated tree is estimated to be about 1012 or a trillion cells, the final sub-sampled lineage stored information for only 10,000 cells.

Best performing methods

• AMbeRland
• Cassiopeia
• DCLEAR
• FastTree
• Guan's method
• Liu's method
• RNI labs' method

Generating in silico datasets:

• C. elegans challenge
• M. musculus challenge

Data availability:

The training and testing data for three challenges can be accessed here (sign-in required).

Scoring:

We applied two widely used metrics for tree comparison: the Robinson-Foulds distance and the triplets distance. While both metrics are applied to assess tree similarities there is no clear agreement as to which one is more relevant for lineage trees. We decided to use both metrics as a way of evaluating their correlation and the insight they provide about the lineage relationships.

https://github.com/Lineage-Reconstruction-DREAM-Challenge/hub/wiki/Scoring

Others:

• Performance on c. elegans challenge's training data
• Running time