TreeMineR 1.1.0

• Fixed Monte Carlo P-value calculation in TreeMineR(). Simulated maximum log-likelihood ratios (LLRs) equal to the observed LLR are now counted as at least as extreme as the observed value (>= instead of >), as required for a valid Monte Carlo P-value. Previously, ties were excluded, which inflated the family-wise type-I error rate, particularly for sparse count data with many possible ties (#9).

• Fixed calc_llr() so that nodes in which every individual is exposed (or every individual is unexposed) no longer get an undefined (NaN) log-likelihood ratio. Such nodes were previously dropped from both the simulated reference distribution and the reported results, which could bias P-values (#9).

• Added a check to TreeMineR() which now throws an error if p is not greater than 0 and smaller than 1. p = 0 or p = 1 made the log-likelihood ratio degenerate (Inf) for all nodes containing exposed, respectively unexposed, individuals.

• Monte Carlo simulations in TreeMineR() now preserve the correlation between related cuts. Exposure is simulated once per individual and re-aggregated through the tree, rather than resampling every cut independently, so a parent cut's simulated count is now mechanically consistent with its children's, as it is in the observed data (#10).

• Added a check which now throws an error if data does not contain any rows, and if tree contains leafs that are duplicated across different branches. Both previously caused an uninformative error from internal data.table operations.