bayesian gaussian mixture modeling  (MathWorks Inc)

Journal: PLoS Computational Biology

Article Title: SciClone: Inferring Clonal Architecture and Tracking the Spatial and Temporal Patterns of Tumor Evolution

doi: 10.1371/journal.pcbi.1003665

Figure Lengend Snippet: (a) A fraction of the ∼800 variants from Fig. 3 were randomly sampled and the resulting number of clusters was inferred using beta mixture modeling. Error bars represent standard deviation ( ). (b) Mutations from clusters one and two from the AML relapse sample were used to assess the limits of cluster separability. As the distance between the two mutation groups was varied, the resulting clusters were assessed for overlap (the fraction of the data within a single standard deviation of both clusters) and accuracy (the fraction of items that were correctly assigned to a second cluster). (c) Consensus clustering of the AML data set (Fig. 3) for number of initial clusters varied from six to 15 and clustering method varied across beta, Gaussian, and binomial mixture models for a total of 30 runs. consensus matrix holds all variants across both rows and columns and has been reordered so that variants belonging to the same cluster are adjacent to one another. Matrix entry , is the fraction of runs in which variant and were co-clustered; entry corresponds to the top-left of the matrix heat map. The narrowest neutral-colored band corresponds to a single variant alternatively classified by Gaussian mixture modeling . The larger neutral-colored band corresponds to variants alternatively classified as a sixth cluster by binomial mixture modeling .

Article Snippet: To do so, SciClone employs variational Bayesian mixture modeling of beta, binomial, and Gaussian distributions.

Techniques: Standard Deviation, Mutagenesis, Variant Assay