I am interested in using SciClone for tumor heterogeneity analysis, and was testing out my ability to properly run it by recreating Figs 3 and 5 from Miller and White et al. 2014 using the data and scripts available at github.com/genome/sciclone-meta. While my recreation of Fig. 5 appears identical to that published, my analysis (please see image below) of the data available for Fig. 3 using the provided run.R script differs from that in the publication, in that I am getting an extra and quite small cluster (Cluster 2 in image below). Has anyone else noticed this, and do you know what might be causing it?
I can recreate this. Changes to the codebase have resulted in slightly different results, especially in difficult cases like the one displayed. Rolling back to version 1.0.6, which was the version released with the paper, still partitions them into 5 clusters.
Those three points don't neatly tuck into cluster1, and are probably a little offset due to technical artifacts that skew their VAF. If we were to take a hard look at the alignments in IGV, my guess is that they would be real variants in a difficult-to-align region, and could probably be added as points to exclude from clustering. Regardless, they're an edge case and I could see the argument either way, based on looking at the plot. (Of course, based on the biology and evolutionary relationships of the clusters, a distinct cluster there absolutely does not make sense!).
Bottom line, I think the changes that were made improve the algorithm for the vast majority of cases, but unfortunately, also affect this example negatively. Best of luck!
What's the version number of the sciClone and bmm packages that you're using? You can find this with library(sciClone);sessionInfo()
sciClone_1.1.0 bmm_0.3.1