RZooRoH 0.4.2

• Improved memory usage - in particular for parallel computing
• More efficient computation of allele frequencies (C functions)
• Improved management of convergence problems:
  • possibility to automatically restart analysis in case of convergence problems by providing multiple optim methods or starting values for mixing coefficients,
  • improved error message, information on problematic individuals (or pairs),
  • added @optimerr slot to kinship results
• Improved numerical properties for ‘Interval’ models (better stability with very low values)
• New modelling of the minimum rate in MixKL models (compatible with more optim methods)
• New options to provide map directly in cM or in physical position but with a cM-to-Mb ratio different from 1
• Added new accessor functions for kinship results (update and merge), extended update and merge functions to IBD analyses
• Added new accessor function to estimate inbreeding coefficients based on identified HBD segments
• Corrections of minor bugs
• Updated documentation according to changes and clarified links between mixing coefficients and inbreeding rates per layer, and between “rates” and rates of coancestry changes
• Updated documentation on conversions (using PLINK2 and dosage formats)
• Modifications to the vignette, including advice for memory management and handling convergence problems
• Set freqem to TRUE by default
• Updated citations

RZooRoH 0.4.3

• Optimization of the Viterbi algorithm: improved efficiency for identical results
• Replacement of sums over generations by exact closed forms (faster, same results)
• RecTable: new logarithmic grid (1 bp to 1 Gb) with a uniform error < 0.5% (the best value of the previous grid), without the former 100 bp floor
• Fixed handling of optim failures in zoorun() and zookin(): no more crash when optim returns an error, and results are now correctly kept when a run converges after an automatic relaunch (additional_zmix or multiple optim_methods). Failed runs are reported with zeros (optimerr = 99)
• Corrected bug when running Viterbi with the step function and without parameter estimation
• Fixed small bug of error reporting (in case of single error)
• Forward-backward: guard against division by zero when emission or initial (stationary) state probabilities are extremely low or zero
• RecTable: fixed incorrect indexing for marker spacings > 100 Mb (and > 1000 Mb)
• Mixing coefficients in mixkl and step_mixkl models are now bounded above by a fixed value of 0.999 (in addition to the lower bound minmix). This avoids numerical problems when a single class captures almost all autozygosity, with negligible impact on estimates in practice
• The transformation between optimized parameters and mixing coefficients now uses plogis()/qlogis(), which are numerically stable for extreme values (the previous form could return NaN)
• Fixed the @realized slot for step_mixkl models run without localhbd: it stored the full Forward-Backward output instead of the realized autozygosity proportions
• Improved numerical stability of the likelihood, Forward-Backward and Viterbi routines for layer models (“SingleRate” HBD classes)
• Emission probabilities are now bounded below by 1e-12 to avoid numerical failures when an observation is impossible under the model, e.g. a genotype incompatible with an allele frequency of zero, or extreme read depths with the “ad” format. The bound has no effect on realistic data
• Fix mixc length (rep(0, K2)) after optimization failure in Step models
• Internal code simplifications in lik_kl() (no change in results)
• Updated help pages, vignette and citations, notably regarding the trim_ad option, MAF filtering, minmix usage, starting values for interval models and the new rectable.