NEWS.md

In OchoaLab/ligera: LIght GEnetic Robust Association

ligera 1.0.0.9000 (2020-06-26)

• First GitHub release! Includes ligera function and scripts/ligera.R command line interface script, which supports plink BED/BIM/FAM files.

ligera 1.0.1.9000 (2020-07-09)

• Added ligera2, which implements full "BOLT trick" to avoid having to calculate kinship explicitly, and which scales as $O( m n \sqrt{\kappa} )$ instead of $O( mn^2 + n^2 \sqrt{\kappa} )$, which is far better as n increases.

ligera 1.0.2.9000 (2020-07-13)

• Added ligera2_bed, a version of ligera2 specific to plink BED files sped up with Rcpp code for critical components!
• Added covariate support to all the main functions (ligera, ligera2, and ligera2_bed).

ligera 1.0.3.9000 (2020-07-14)

• Added multiscan wrapper for all versions of LIGERA (ligera_multi, ligera2_multi, ligera2_bed_multi), which iterates the genetic association scans and adds significant loci to covariates (options for one per iteration or all significant per iteration) to condition upon them and increase power on the weaker signals.

ligera 1.0.4.9000 (2020-07-21)

• Added support for missing values in covariates matrix
• In each column of covariates matrix, missing values are replaced with the within-column mean value.
• All core versions updated (ligera, ligera2, and ligera2_bed).
• This was necessary for multiscan versions (ligera_multi, ligera2_multi, ligera2_bed_multi) to work when the genotype matrix has missing values, as these may be incorporated as covariates in the iteration.
• Multiscan versions stop iterations if and when the number of covariates exceed the number of individuals, to avoid inverting singular matrices (problem becomes overdetermined).
• Fixed class usage now that matrices return a two-element array in R version 4.

ligera 1.0.5.9000 (2020-10-16)

• All 6 ligera versions (ligera, ligera2, ligera2_bed, and *_multi variants)
• Added option m_chunk_max to further reduce memory usage (compared to previous default) without sacrificing speed.
• Added script wrappers (located in scripts/) for main 3 versions of ligera, each with a --multi option to also cover the 3 multiscan versions (actually added 2020-09-29)
• Added sample files to go with scripts (also located in scripts/; also added 2020-09-29)

ligera 1.0.6.9000 (2020-12-08)

• Minor internal improvement (in function conj_grad_scan_bed_wcpp) affecting ligera2_bed and ligera2_bed_multi versions only.

ligera 1.0.6.9000 (2021-01-08)

• Internal improvements: replaced sweep cases with matrix version (redid Tiffany Tu devel branch commits e9629dc and 8493282 with minor edits, thanks Tiffany!!!)
• Forgot to update version.

ligera 1.0.7.9000 (2022-05-26)

• Added functions ligera_f, ligera_f_multi
• This version calculates p-values using an F-test, which gives calibrated statistics under both quantitative and binary traits.
• Compared to the original ligera, which uses the faster Wald test (calibrated for quantitative but not binary traits), the F-test version is quite a bit slower, and is optimized for m >> n, so it is a work in progress.
• Unit tests: fixed issues arising from apearance of fixed loci when individuals with missing traits are excluded
• Reformatted this NEWS.md slightly to improve its automatic parsing.
• Removed "LazyData: true" from DESCRIPTION (to avoid a new "NOTE" on CRAN).

ligera 1.0.8.9000 (2022-05-26)

• Unit tests: reduced simulation size (n = 20, from 100; left m = 1000)
• Reduced its runtime to 3.4s (before it was highly variable but excessive, between 60s-131s).
• Reduction does not appear to increase testing artifacts (such as singular matrix inversions).

ligera 1.0.9.9000 (2022-05-26)

• Function ligera fixed bug that inbr wasn't subset for non-NA individuals when there is missingness in the input trait
• Only occurred if inbr was explicitly provided, otherwise lazy evaluation from kinship (which was subset) ensured that it was subset when it was not provided as explicit parameter.
• Other versions (ligera2, ligera2_bed, ligera_f) do not have the option of providing inbr so were not affected by similar bugs.
• Function ligera_f
• Added option V to specify algorithm (there are four versions, numbered 0-3, roughly falling into two scalability classes). This is a temporary option that will be removed if one version emerges as clearly superior.
• Default algorithm changed to what is now V=0 (default was V=1 before, but internal tests show that it has the lowest numerical stability of all the choices, below machine precision). Previously all versions were commented out except V=1.
• Unit tests
• Replaced trivial true kinship of unstructured simulations (I/2) with popkin estimates in all tests. This was necessary so WG-biased estimates would differ from their unbiased counterparts, and helped identify the numeric stability issues of the previous ligera_f version.
• Added tests that confirm that ligera_f is invariant to WG bias (a type of bias that always results in non-singular kinship matrices).
  • Oddly, ligera is not invariant to WG, and it is unclear at the moment why.

ligera 1.0.10.9000 (2022-09-18)

• Added function ligera2_f, a version of ligera_f that calculates performs all calculations without an explicit kinship matrix estimate (like ligera2 was to ligera).
• Retains four V versions as before, and (except for V=1) all and agree with each other and with ligera_f up to machine precision. Runtime has not been evaluated, might not be good yet for any version (f versions remain experimental).
• Internal function changes:
• popkin_prod added option want_inbr (to skip calculation of inbr, as it is unused in f versions; default is to calculate and return it as before).
• conj_grad_scan added options want_inbr (to skip calculation of inbr), want_b (to return b, default does not as before), and b (allows for providing b as calculated in a previous run). Additionally mean_kinship is now NA by default and it does not have to be provided if b is provided. Lastly, if want_inbr and want_b are both false, then only Z is returned (instead of a list with these objects and Z).
• Corrected minor errors in non-f unit tests (the wrong functions or objects were being tested, though the changes were inconsequential).

ligera 1.0.11.9000 (2022-09-18)

• Functions ligera_f, ligera2_f: removed previous V=1 algorithm, which was numerically unstable. Shifted down previous versions 2:3 to 1:2.

ligera 1.0.12.9000 (2022-09-18)

• Added function ligera2_f_multi.
• Internal function changes:
• conj_grad_scan and cgsolve_mat: handled a case where input Y has a column of zeroes, which should give an output Z with zeroes in the same columns but which the conjugate gradient algorithm gave NaN's for those same columns of Z (because of a 0/0 factor). This special case is now recognized and handled correctly.

ligera 1.0.13.9000 (2022-09-19)

• Added function ligera2_bed_f.

ligera 1.0.14.9000 (2022-09-19)

• Added function ligera2_bed_f_multi.

ligera 1.0.15.9000 (2022-09-20)

• Function ligera2_bed_f_multi added ability to filter discovered loci by LD, using internal function ld_prune.
• Added parameters prune_ld to turn this feature on, and r2_max and pos_window to control how pruning is done.
• Removed parameters m_loci and n_ind now that function just reads BIM and FAM files in addition to BED file that is required.
  • Package genio is now imported to read these files (used to be merely "suggested").
• Function returns a bigger tibble containing the BIM table prepended to the previous association outputs.

ligera 1.0.16.9000 (2022-09-20)

• Function ligera2_bed_multi ported ability to filter discovered loci by LD from ligera2_bed_f_multi, and all other related changes from previous entry.
• Function ligera2_bed_f_multi added parameter V (to pass to ligera2_bed_f).

ligera 1.0.17.9000 (2022-09-22)

• scripts/ligera2_bed.R added --fstat option for running ligera2_bed_f and ligera2_bed_f_multi on a plink1 set of files on the terminal.

ligera 1.0.18.9000 (2023-01-04)

• Preemptively solved an R-devel warning by replacing compound sprintf and stop calls in Rcpp code with direct stop calls.

OchoaLab/ligera documentation built on Jan. 5, 2023, 8:29 p.m.