fixed setup.py to avoid conda build issue

This release does not involve any changes to decifer functionality and simply fixes an issue involved in the conda installation.

silhouette score for clustering, bugfix

This release has a variety of changes including:
• Pass the argument --silhouette to DeCiFer to have it automatically determine the best number of clusters using the silhouette score, this is in contrast to the current default method of using the elbow criteria which, in practice, requires adjusting the sensitivity via the --elbow parameter for different samples in order to get good results
• bug fixes that enhance clustering and enable more accurate assignment of genotype trees to SNVs
• when analyzing multiple samples per patient, DeCiFer creates a cluster for SNVs that are truncal in all samples (cluster 1) in addition to one cluster per sample to capture sample-specific SNVs (clusters 2 through n, where n is the number of samples). However, these sample-specific clusters are disabled when analyzing just a single sample from a tumor.

fixed bug when calculating CIs for clusters with no assigned mutations

DeCiFer now outputs "NaN" for confidence intervals (CIs) for clusters to which no mutations were assigned. Previously, DeCiFer would attempt to compute these CIs and throw an error. This may only have briefly been an issue since version 2.1.0, where we introduced a feature in which DeCiFer filters out outlier mutations that poorly fit the observed data. Thus, it was possible for a cluster to have all of its mutations filtered out (and printed to _Outliers_output.tsv).

bug fix changing how genotype trees are stored internally

A bug in the code was fixed that, in rare cases, caused DeCiFer to erroneously issue a "Skipping mutation" warning for SNVs that did not have genotype trees for certain copy number combinations. Genotype trees are now internally stored such that the order of copy number states in the input file doesn't matter.

Outlier mutations printed to separate file

For each SNV, DeCiFer selects one of K clusters and the genotype tree that has the highest likelihood. However, even the best-fitting model still produces an expected VAF that is far away from an SNVs observed VAF. For these SNVs, we print the results to a separate output file with suffix _Outliers_output.tsv. How many standard deviations the observed VAF of an SNV needs to be from the expected VAF (computing using the DCF/CCF cluster center) in order to be classified as an outlier and printed to this separate file can be changed using the new --vafdevfilter option, which has a default value of 1.5.

BETA option to compute DCF/CCF cluster CIs more conservatively.

Beta: use the --conservativeCIs flag to have DeCiFer compute CIs that are wider and more conservative. This approach uses the DCF point values of the mutations assigned to a cluster to compute the cluster's CIs. Specifically, we compute the median of the distribution of DCF point values (one for each mutation assigned to the cluster) and use bootstrap resampling to calculate CIs. To be conservative, we use the minimum and maximum observed median across 10,000 bootstrap replicates.

Fit beta-binomial parameters as DeCiFer option

Users can now pass an option to DeCiFer to use beta-binomial distributions to find clusters and assign mutations (default is binomial). We have updated the documentation to describe this procedure and the additional input file requirement containing the number of germline mutation counts in each sample.

In addition, we have also

1.) fixed a bug related to rescaling results by purity so that DCF/CCF values range from 0 to 1 for each sample, as opposed to ranging from 0 to the sample purity (as in v1*)

2.) changed the default number of restarts to 20 to make DeCiFer have decreased runtimes, we have noticed similar results with a wide range of restart values

Print output for all specified values of K

Given a range of K clusters specified by the user, DeCiFer tries to find the best value using model selection and the --elbow parameter. However, it can be useful to see the output for all K values (not just the one chosen via model selection) to explore how your results may vary, especially since one value for the --elbow criteria may not be appropriate for all datasets (e.g. single sample vs. many samples from same tumor).

CMM CCF rescaled by purity

In this release we have also rescaled the calculation of Constant Mutation Multiplicity (CMM) CCFs by purity so that they are comparable with CCF/DCF values. Moreover, we also fixed an assert error that may rarely occur.

DCF analyses scaled by purity

We have scaled all analyses by purity such that DCF values have the same interpretation in every sample, regardless of the sample's purity. This change has also enhanced the clustering analysis. Because we have also migrated the source code from python 2 to python 3, we bumped the version to 2.