Entering edit mode
5.8 years ago
Kevin Blighe
87k
Just adding my own analysis protocol for CyTOF data here: https://github.com/kevinblighe/cytofNet
This simply requires FCS files as input and was developed independently of the main packages that have since been released, but its functionality is similar.
Currently unpublished but will be pushing to get it out.
cytofNet
Unbiased identification of ideal clusters, k, in mass cytometry time-of-flight (CyTOF) data, with interrogation of clusters via network analysis.
Key points
- data is downsampled based on low variance
- unbiased clustering to identify ideal number of centers, k, is performed via bootstrapped partitioning around medoids (PAM) and 3 metrics: Gap statistic (Tibshirani et al., 2001); silhouette coefficient; elbow method
- extra checks for ideal cluster segregation based on pairwise statistical measures between clusters
- medoids from PAM are used to infer low/high marker expression
- network plot construction used to show relationships between identified clusters