NEWS

---

SAIGEgds 2.14.0

• SPA root-finding hardening in 'SPATest.cpp': the saddlepoint equation solvers 'Saddle_Prob()' and 'Saddle_Prob_Fast()' now use a bracketed safeguarded Newton (rtsafe) fallback for 'getroot_K1()' / 'getroot_K1_fast()'. Since K1_adj is strictly increasing (dK1/dt = K2 > 0) the root is unique and bracketable, so the bisection safeguard guarantees convergence. The fallback is hybrid: the existing fast Newton runs first and rtsafe is invoked only when it reports non-convergence, so there is no added cost on the common path. This removes the rare non-convergence that silently reverts the variant to the anticonservative normal approximation
• 'seqFitNullGLMM_SPA()' and 'seqAssocGLMM_SPA()' support time-to-event (survival) outcomes via 'trait.type="survival"' and a new 'event.time' argument, fitting a Cox proportional-hazards frailty model through the Breslow Cox-Poisson equivalence and applying a Poisson saddlepoint approximation in the single-variant score test (the GATE method, Dey et al. 2022); the Cox risk-set information correction (W - U) is used in the variance-ratio estimation via a Woodbury identity
• a new vignette "Set-based Rare-Variant Association Analysis in SAIGEgds", covering Burden, SKAT, ACAT-V and ACAT-O tests with full and sparse GRMs
• a new vignette "Time-to-event (survival) analysis in SAIGEgds"

SAIGEgds 2.12.0

• fix a C++ error where 'span' is ambiguous
• new functions 'seqRefitNullGLMM()', 'seqAssocGLMM_GT()' and 'seqAssocGLMM_ACAT_O_GT()'
• SAIGEgds supports optional OpenCL-based GPU acceleration for the GRM cross-product computation during null model fitting
• new 'use.gpu' argument in 'seqFitSparseGRM()' and 'seqFitDenseGRM()' for OpenCL GPU-accelerated GRM construction
• new 'geno.model' argument in 'seqAssocGLMM_SPA()', 'seqAssocGLMM_GT()', 'seqAssocGLMM_Burden()', 'seqAssocGLMM_SKAT()', 'seqAssocGLMM_ACAT_V()', 'seqAssocGLMM_ACAT_O()' and 'seqAssocGLMM_ACAT_O_GT()' for additive, dominant, or recessive genetic models

SAIGEgds 2.10.2

• 'fork.loading' is deprecated in ‘seqFitNullGLMM_SPA()', use ’parallel.loading' instead

SAIGEgds 2.9.2

• improve the workload balancing in 'seqAssocGLMM_SPA()' according to SeqArray_1.49.6
• new option 'verbose.pval' to improve the display in 'seqAssocGLMM_SPA()'

SAIGEgds 2.4.1

• add an option 'load.balancing' to 'seqAssocGLMM_SPA()'

SAIGEgds 2.4.0

• update 'seqFitLDpruning()'
• reduce the memory usage in 'seqAssocGLMM_SPA()' when the genotype file is not split by chromosomes

SAIGEgds 2.2.1

• fix the compiler warning: -Wformat-security
• fix a compiler issue with gcc v7.5

SAIGEgds 2.2.0

• fix the "Matrix-deprecated" warning from the Matrix package when calling 'as(dsCMatrix, "dgCMatrix")'

SAIGEgds 2.0.0

NEW FEATURES

• New features to use a sparse genetic relationship matrix in generalized linear mixed models (GLMMs) according to the SAIGE-GENE paper (Zhou et al., 2020)
• a new argument 'grm.mat' in ‘seqFitNullGLMM_SPA()': the dense or sparse genetic relationship matrix (GRM) can be specified via ’grm.mat'
• MAC categories for multiple variance ratios in 'seqFitNullGLMM_SPA()', 'seqAssocGLMM_SPA()', designed for rare variants
• new function 'seqAssocGLMM_SKAT()' for the SKAT aggregate method
• new feature 'seqAssocGLMM_ACAT_O()' to include the SKAT aggregate method for full ACAT-O tests
• faster 'seqSAIGE_LoadPval()' when merging multiple input files

SAIGEgds 1.12.5

• enable RCPP_PARALLEL_USE_TBB=1 on Windows

SAIGEgds 1.12.4

• fix the compiling issue on ARM64 (see https://github.com/AbbVie-ComputationalGenomics/SAIGEgds/issues/8)

SAIGEgds 1.12.1

• fix the memory issue because of using deprecated tbb::task_scheduler_init in RcppParallel

SAIGEgds 1.6.0

• 'seqFitNullGLMM_SPA()' can use imputed dosages directly without converting the dosages to the best-guess genotypes
• new function 'glmmHeritability()' for approximate heritability estimates

SAIGEgds 1.4.0

• set-based tests: burden, ACAT-V

SAIGEgds 1.2.2

• update the citation
• work around gcc-10

SAIGEgds 1.2.0

• 'seqAssocGLMM_SPA()' can save to a RDS file

SAIGEgds 1.0.2

• fix an issue when there is no covariate in the formula (e.g., y ~ 1)

SAIGEgds 1.0.1

• add a new option for the threshold of missing variant to 'seqAssocGLMM_SPA()'
• change the output column 'AC.alt' to 'mac' in 'seqAssocGLMM_SPA()'
• retry model fitting at least 10 times if the matrix is singular or not positive definite, or large variance is observed

SAIGEgds 1.0.0

• first Bioconductor release

SAIGEgds 0.99.0

• package submitted to Bioconductor

SAIGEgds 0.9.10

• SAIGE algorithm implementation for quantitative outcomes

SAIGEgds 0.9.9

• add a vignette to the package
• the default of random number generator changes in R: "Rounding" was the default in 'RNGkind()' prior to R_3.6.0, but "Rejection" is used in R (>= v3.6.0). For reproduction of the results created by R (< v3.6.0), please use 'RNGkind("Mersenne-Twister", "Inversion", "Rounding")' in R (>= v3.6.0)

SAIGEgds 0.9.7

• seqAssocGLMM_SPA(): load balancing in parallel

SAIGEgds 0.9.0

• first version of SAIGEgds