(Previous | Contents | Next)
BUG FIX: corrected major divergence bugs. Attempts to infer epoch times produced markedly wrong results. All runs inferring epoch times should be re-run.
BUG FIX: spurious migration blocks in single population converter runs. The lam_conv file converter no longer creates dysfunctional migration-enabled lamarc XML files when there is only one population in the input data.
BUG FIX: corrected major migration bug A bug in data summarization for migration parameters reversed the direction of migration in some senses bug not others. All runs modeling migration parameters not held constant should be re-run.
BUG FIX: improved handling of extreme negative growth A bug in tree generation under extreme negative growth resulted in trees with default branch length being accepted. If you have seen results (especially in Likelihood runs) where growth rates oscillated regularly from negative to positive, that was likely this bug. All runs modeling growth should be re-run unless they were Bayesian runs with exclusively non-negative priors.
New feature: Divergence: LAMARC now models multiple populations diverging from common ancestors.
New feature: SNP panel corrections: LAMARC can now correct for the loss of recent variation induced by using SNP panel chips.
Enhancement: faster convergence in Likelihood runs: Method for checking convergence in LIKELIHOOD runs has been improved, yielding faster convergence.
BUG FIX: corrected bug in data likelihood in presence of recombination A bug in the data likelihood calculation for recombinant trees caused increasingly inaccurate data likelihoods as the number of non-recombinant sub trees grew. This led to LAMARC preferring trees with fewer recombinations. Recombinant analyses run with LAMARC versions 2.1.2 through 2.1.5 should be re-run.
BUG FIX: corrected random number generator Corrected a bug in which random values which were supposed to be in the open interval (0,1) were sometimes returning 0 or 1. The bug was rare, resulting in unexplained crashes every 4 billion or so random number draws. There is no need to re-run analyses which did not crash. We also updated our random number generators to use the Boost Mersenne twister (boost::mt19937) random number generators.
BUG FIX: data uncertainty model with SNP data (beta test) Data likelihood for the invariant base pairs in SNP data now incorporates the per-base error rate. Analyses using SNP data with the per-base error rate model should be re-run.
data uncertainty model (beta test)
improved output reports
compiles with g++ 4.3.3 on Linux This release updates the code base to compile with g++ 4.3.3. Earlier compilers should still work.
minor user experience improvements Several error messages from the converter have been improved. Converter can now read input files missing the end-of-line character. Lamarc now records the random seed used. This is useful when debugging problems.
Bug fix for haplotype rearranging code This release fixes a bug introduced into the code used to guess haplotype resolution. Analyses using this feature in LAMARC versions 2.1.2 and 2.1.2b should be re-run. Additional improvements include:
Minor changes affecting user experience only Removed requirement for user to "press enter to quit" in batch mode. Restored missing icons to MSW distribution.
Limitations for Recombination relaxed. LAMARC is now 'final coalescent' aware, meaning that individual sites that have coalesced no longer induce recombination events. This means that recombination can be estimated for much longer distances--version 2.1.1 had problems when theta * r * sequence length (or 4NCl) was any higher than about 5. 2.1.2 can now handle values of 4NCl up to ~100. In humans, this translates to about .2 centimorgans, or 200 kilobases. We're working on expanding this even further--if you have a data set that needs longer recombination lengths, let us know (we're particularly interested in people using LAMARC over long distances for trait mapping).
Tracer output now includes the probabilities of the sequence data on the current genealogy in Bayesian runs as well as in Likelihood runs.
Unknown microsatellite data are now recognized by the converter.
Maximization has been made somewhat more efficient in some cases, particularly in runs that estimate variable mutation rates over regions drawn from an unknown gamma distribution.
Bayesian Tracer files now omit parameters set to be 'invalid'.
Bayesian analyses now handle runs with too few unique sampled parameter values a bit more robustly, and warn a bit more sternly.
Trait mapping. Trait data (such as disease status) can be mapped, modelling trait changes as K-Allele data, with arbitrary models for penetrance. Mapping can be performed using two approaches, one that includes the trait data when rearranging trees, and one that analyzes the trees after they are produced and collects the likelihoods of the data being produced at each site. (See the mapping documentation for more information.)
Multiple data types within a linked genomic region. LAMARC is now able to correctly analyze a genomic region which contains, for example, several microsatellite markers and a stretch of single-copy DNA. The researcher will need to provide the expected relative mutation rate of each type of data.
GUI Converter. The GUI file conversion utility included with this release has been significantly updated, replacing the beta version originally released with version 2.0.
Batch versions. LAMARC could previously be compiled in such a way as to produce a 'batch' version. Now, that capability has been extended to the normal version: If you execute lamarc with a '-b' (or '--batch') command line option, it will run through and produce output without further interaction from the user. The converter may be run the same way, also with a '-b' command-line flag (see the converter documentation).
Input file setting from the command line. You may now specify a LAMARC input file to use at the command line with a command like "lamarc new_infile.xml". This is particularly helpful with the '-b' option, above, as it means you can use a different input file than the default 'infile.xml'.
Bayesian runs with different effective population sizes for different regions were producing erroneous output due to a bug in tree rearrangement. (This would most easily manifest as abnormally low acceptance rates for theta values in regions with an effective population size other than 1.0.) This has been fixed.
Phase tags in XML input file. As part of enabling multiple data segments of different types within a linked genomic region, we have changed the numbering system used to indicate which sites in an individual are phase-unknown to match the numbering system used for other purposes. As a result, any previous XML input files which explicitly listed the phase-unknown sites for each individual (rather than indicating that no sites were phase-known or phase-unknown) will need to be hand-edited to use the new scheme. LAMARC will generally be able to detect this problem and issue an error message. If you find you have this problem with an old infile, you can either recreate the lamarc input with the converter, or edit the infile. If you have all phase-unknown data, the simplest method is to change the default (and now incorrect) tags:
<phase type="unknown"> 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 [...] </phase>
<phase type="known"> </phase>
The latter will set all of your markers to phase unknown.
Note: version 2.0.3 was not released to the general public, but was made available for the 2006 Workshop on Molecular Evolution at Woods Hole.
Gamma rates among regions. Mutation rates can now vary among genomic regions according to a gamma distribution; the program will attempt to estimate the gamma shape parameter. This is particularly useful for collections of microsatellite regions where little is known about the relative mutation rates of each region. See the document "Combining data with different mutation rates" for more information.
Tracer compatibility. LAMARC now automatically writes files that can be read by the Tracer utility of Drummond and Rambaut. In a Bayesian run Tracer can monitor convergence of the parameter estimates. In a likelihood run it can monitor the data likelihood of the genealogies. In both cases, it is useful in determining whether the program has been run long enough. See the "Using Tracer with LAMARC" documentation for more information.
Newick tree. LAMARC can now write out the tree of highest data likelihood it finds for each region, as a Newick format tree, in cases which do not have migration or recombination.
Several limitations of the Stepwise and Mixed-KS microsatellite models have been relaxed, allowing runs with widely divergent microsatellite counts to run to completion instead of halting partway through the program run. (If an analysis using an earlier version of LAMARC ran to completion, it was not affected by this problem; affected runs would crash.)
The mixing parameter of the Mixed-KS model could previously be set to adjust during the run, but did not actually do so. Now it does.
Summary file reading and writing can now be used with Bayesian as well as likelihood analysis.
Maximizer fine-tuning allows likelihood maximization and profiling to succeed in some cases where they previously would have failed.
Bayesian multi-region analysis was incorrect in the previous version; probability curves representing multiple regions were added rather than multiplied. The resulting MPEs are correct but the confidence limits are unnecessarily wide. All Bayesian runs with multiple unlinked genomic regions should be redone.
Likelihood multi-replicate analysis was incorrect in the previous version. Neither MLEs nor confidence limits are accurate. All likelihood runs done using replication from version 2.0 (not from earlier versions) should be redone.
Bayesian analysis. Lamarc can now make a Bayesian estimation of population parameters as an alternative to the original maximum-likelihood estimation. Linear and logarithmic priors with user-specified upper and lower bounds are available. Users are strongly encouraged to set appropriate priors. In our limited experience, the results of Bayesian analysis are quite similar to those of likelihood analysis, but the Bayesian approach may be superior for estimating parameter values near zero.
As an adjunct to Bayesian analysis, we offer a new genealogy-search strategy which reconsiders only branch lengths. This may allow the search to more rapidly react to newly proposed values of Theta. It can also be used in likelihood-based analysis. It is currently enabled by default for all lamarc runs, so attempts to exactly replicate previous results will first need to disable this strategy.
Parameter constraints. Individual population parameters (such as migration or growth rates) may now be constrained to a user-specified value, or groups of them may be constrained to be equal. We especially recommend use of constraints to reduce the number of parameters in cases with many subpopulations.
Different Ne and mu among genetic regions. It is now possible to set the relative Ne (effective population size) and relative mu (neutral mutation rate) of each genetic region independently, allowing a correct joint estimate over unlike regions such as autosomal and sex-chromosome samples or DNA and microsatellite samples.
New data types and models. Data with multiple alleles among which no particular relationship is implied, such as elecrophoretic alleles, can now be coded as "K-Allele data" and analyzed via a K-Allele model. This model is also available for microsatellites as an alternative to the stepwise mutation models; furthermore, a mixed model which attempts to optimize the ratio of stepwise changes and K-allele changes can be used for microsatellite data.
Graphical user interface for file conversion utility. File conversion from PHYLIP or MIGRATE format files can now be done using a GUI interface which is significantly easier than the text-based form. The text-based converter is still available.
Multiple replicates are now correctly implemented using the method of Geyer.
Inference on phase-unknown DNA or SNP data had a serious bug in version 1.2 which is fixed in this version. Previous analyses of this type should be repeated as the bug did not cause a crash, but led to inaccurate results.
Effectiveness of the maximizer has been greatly improved; it finds correct maxima in a much larger proportion of cases. (The maximizer is the part of the program that searches the n-dimensional likelihood surface for the maximum height, which is the maximum likelihood.)
Older XML files which use the following tag will need to be modified. The tag <map_position> (with an underscore) has become <map-position> (with a dash) for consistency with the other tags.
Multiple replicates. Regrettably, we have disabled the ability to do multiple replicates of each chain as an accuracy improving measure. This had not been implemented correctly; it produced approximately correct maximum likelihood estimates but too-narrow confidence intervals. We will re-enable this feature as soon as we have correct algorithms for combining results over replicates. (This feature has been corrected and re-enabled in version 2.0.)
Previous multiple-replicate runs may well have too-narrow confidence intervals and should be redone. We regret this problem.Additions
Growth. The program can now estimate an exponential growth rate for a single population or for several subpopulations. This duplicates the functionality of FLUCTUATE except that LAMARC can estimate growth in the presence of recombination and/or migration as well.
General Time-Reversible mutational model. For DNA, RNA or SNP data, the program can now use a fully-specified form of the GTR mutational model. It is not able to optimize the parameters of this model, but other tools such as PAUP*/Modeltest can be used to develop an optimal model to be applied by LAMARC.
Adaptive heating. When using the MC^3 or "heated chains" strategy to improve searching, the program can now adjust the temperatures of the heated chains automatically in an attempt to improve efficiency, rather than relying on user-specified fixed temperatures.
Menu revision. The menu has been extensively revised and now has the capacity to undo multiple changes. In addition, a few options on the menu have been moved to theoretically more reasonable spots.
Saving menu options. The program automatically writes a file, "menusettings_infile.xml", which contains the user's original infile updated with the results of all changes made via the menu. This greatly simplifies re-running a complicated case.
Saving sampled genealogies. The program is now capable of writing a file containing summaries of its sampled genealogies, and can read that file back in and resume a run. This is useful in recovering a run that has crashed, and can also be used to do more complex analyses of the same genealogies. For example, you may wish to do a quick run with no profiling in order to find the best run parameters, and then re-analyze those genealogies with profiling if they are satisfactory.
No-menu option. The program can now be compiled in a no-menu form which takes all of its input from the XML infile. This is useful in designing large simulation studies and other batch runs.
Output The tables of data have been transposed so that what used to be displayed in rows is now displayed in columns. This puts all modified values of the same parameter in the same column, which should make it easier to follow the changes.
File converter. When input data was presented in Phylip "interleaved" format it was truncated in the file converter. Also, if multiple input sequences had the same Phylip-truncated name, the converter would silently discard the duplicates.
Maximizer accuracy The maximization routines which generated the maximum likelihood estimates, confidence limits and profiles for population parameters were sometimes unsuccessful in finding the true maxima, leading to incorrect estimates and inconsistent profiles. While we cannot guarantee that the new routines will succeed in all cases, they are greatly improved and also provide more feedback when they fail.
Microsatellite data. Both a stepwise mutational model and a Brownian model are provided. Variable rates at different microsatellite regions can be accommodated with a Felsenstein-Churchill Hidden Markov model. Warning: the stepwise model is very slow, and so has not been as thoroughly tested as the others.
SNP data. We implement the "reconstituted DNA" model of Kuhner et al. 2000. The user must provide map information showing the location of the SNPs relative to each other in order to estimate recombination rate; unmapped SNPs are usable for population size and migration rate estimation only.
Genotypic data. The program can now use data for which the haplotypes are unknown. It searches among many different haplotype resolutions. Be sure to use heating if you use this option, as otherwise the search tends to become stuck.
Speed. The version 1.0 release still contained some debugging code which slowed it down substantially (all likelihoods were calculated twice). Version 1.1 should be quite a bit faster.
Lost data in converter. The file conversion program silently lost the last nucleotide of each sequence. This will have had a slight effect on the results. If your sequences are very short you may wish to re-run previous analyses.
Incorrect converter output. Using the file converter for multiple population cases produced defective LAMARC input files which could not run successfully.
File converter flexibility. The file converter is now able to deal with a much wider array of input data.
(Previous | Contents | Next)