Benchmarking metrics

[GiulioSP]

Hi,
The Phyclone paper describes benchmarking metrics including Relationship Reconstruction Error (RRE) and Ancestor-Descendant F-score (AD F-score). Are there published scripts or packages for these metrics? RRE is well defined in the references but I can't find previous implementations of it, and AD F-score seems to be an original metric.

Thanks

[elhurtado]

Ancestor-Descendant F-score is conceptually pretty simple, since it's really just testing the validity of ancestor-descendant mutation pairs over the tree, (i.e. identifying true/false positives and negatives). While the F-Score portion is a convenient way to report on the precision and recall of the Ancestor-Descendant relationships in one metric.

I've attached scripts that'll compute the clustering and AD F-Score benchmarking metrics. The "compute_metrics.py" script requires that the tree being tested and the ground-truth tree are provided as networkx graphs saved in pickle (.pkl) format, where clonal SNVs are stored as sets within their respective clonal nodes under the key "snvs". You can use the "post_process_phyclone.py" script to convert PhyClone results (results table and clone tree newick file) into a networkx graph pickle as required by the "compute_metrics.py" script.

tree_metrics.zip

RRE is a bit more involved, for an example of its implementation: please see here

[GiulioSP]

Thank you kindly!

[github-actions]

This issue is stale because it has been open for 30 days with no activity.