(Previous | Contents | Next)

Limitations of LAMARC

This documentation has a lot to say about what LAMARC can do. Here is a concise guide to what it cannot do, with some suggestions for other approaches.

Some combinations of analyses are not possible.

Due to program or mathematical limitations, some combinations of analyses are not possible and will be rejected if attempted:

  • Gamma-distributed variation in mutation rate among regions is not compatible with estimation of growth, nor can it currently be used in a Baysian run.
  • Newick trees cannot be written out if migration or recombination are allowed.
  • Migration rate estimation requires at least two populations.
  • Recombination estimation requires at least two linked markers.
  • Mapping requires recombination.
  • Divergence can only be inferred in a Bayesian run, not a likelihood run.
  • Some cases cannot be well modelled by LAMARC.

    LAMARC has a wide range of evolutionary models but not all possible ones by any means. Some significant omissions:
  • Samples from multiple time points in a fast-evolving population. Consider the BEAST program of Drummond and Rambaut for this.
  • Multiple population divergence cases where the population tree is not known. Consider the *BEAST program of Drummond and Rambaut when it is available.
  • RFLP, AFLP, or insertion/deletion data. You may be able to use the ARLEQUIN program of Excoffier; we know of no full coalescent likelihood or Bayesian analysis which can handle these data.
  • Growth models other than simple exponential growth or decline. Consider the BEAST program of Drummond and Rambaut.
  • Geographic isolation as a function of distance, rather than via separation into distinct subpopulations. Consider the Dancing Trees algorithm of Baird, if an implementation becomes available.
  • Combining data which have recombination and data which do not (i.e. nuclear and mitochondrial DNA) in the same recombination-aware analysis. We know of no alternative. You may want to do two separate analyses with LAMARC.
  • Sequences from multi-gene families. The underlying coalescent model in LAMARC is not correct for such data. Consider the gene-families ML algorithm of Dubb, if an implementation becomes available.
  • Recombination rates which vary among regions or across the sequence. Consider the LDHAT program of McVean.
  • (Previous | Contents | Next)