Regenie v4.0

• Time-to-event analysis is now available using options --t2e and --eventColList to specify time-to-event analysis and the event phenotype name, respectively
  • Cox proportional hazard regression is used in both step 1 and 2
  • Firth regression is available for time-to-event phenotypes with low event rates
  • see documentation for further details
• Fix algorithm used to fit logistic Firth model when using --write-null-firth to match closer to the approach used in step 2

Regenie v3.6

• Bug fix for the approximate Firth test when ultra-rare variants (MAC<50) are being tested:
  • Fix to the calculation of the null deviance used in the likelihood ratio test.
  • BETA estimates remain unchanged, but p-value will change (SE will also change is using --firth-se.
  • This fix also addresses convergence failures with Firth.
• Improvements to address convergence failures & speed-up exact Firth by using warm starts based on null model with just covariates.

Regenie v3.5

• Added CHR/POS columns to snplist output file when using --write-mask-snplist to make it tabixable
• Genotype counts are now reported in the sumstats file when using --no-split (note these are based on all of the analyzed samples)
  • The added columns names are N_RR|N_RA|N_AA for homRef/het/homALT genotypes, respectively.
• Improved efficiency of LOOCV scheme in ridge level 0
• Detect carriage return in fam/psam/bim/pvar/sample files
• Bug fix for BURDEN-ACAT test when only a single mask is in the GENE_P group
• Address convergence failures when applying Firth correction to single variants in gene-based tests

Regenie v3.4.1

• Reduction in memory usage for LD computation when writing to text files
  • Expected memory usage is (3NB+ M^2)*8 bytes where N is sample size, B is block size and M is number of variants in LD matrix
• Fix bug rejecting valid PVAR files (#492)

Regenie v3.4

• Reduction in memory usage for LD computation when dosages are present;
  • compute LD matrix block-wise rather than all at once
  • expected memory usage is (3NB+ M^2)*8 bytes where N is sample size, B is block size and M is number of variants in LD matrix
  • we recommend using blocks of sizes 1000 as choosing too small block size will increase the number of block pairs evaluated
• Minor bug fixes for LD computation;
• Bug fix for carriage return in optional files
  • in keep/remove/extract/exclude/mask-definition/annotation files

Regenie v3.3

• New strategy for approximate Firth LRT with ultra-rare variants (MAC below 50) by updating log-likelihood only based on variant carriers (can disable using --skip-fast-firth)
• Faster implementation of approximate Firth LRT using customized code for single variant testing as well as better starting values for rare variants (based on Firth without covariates instead of starting at 0)
• Relaxed convergence criterion of Firth LRT from 1E-4 to 2.5E-4

Regenie v3.2.9

• Switch to robust version of ACAT to handle very small p-values
• Bug fix for Step1 when sex chromosome was included in the analysis (issue #413) - thanks to juliandwillett for helping pinpoint to the cause of the issue
• Allow for up to 64 domains per gene when using the 4-column annotation file format with gene domains
• Reduced memory usage for conditional analysis

Regenie v3.2.8

• New option --bgi to specify custom index bgi file accompagnying BGEN file;
• Relax matching criteria between BGEN and index bgi files to use CPRA instead of variant ID

Regenie v3.2.7

• New option --force-mac-filter to apply different MAC filter to subset of SNPs;
• Extend maximum number of domains to 32 for 4-column anno-file;
• Update to external PGEN library

Regenie v3.2.6

• Relax tolerance parameter for null unpenalized logistic regression from 1e-8 to 1e-6
• Bug fix for SKATO when one of the p-values at fixed rho was numerically 0
• More verbose debugging messages