PLINK 1.9’s implementation of this calculation is designed to compensate for GCTA 1.24’s limitations—it is cross-platform, works in low-memory environments, and uses double-precision arithmetic while remaining within a factor of 2-5 on speed. See Table 3 for timing data. The comparison is with GCTA 1.24 on 64-bit Linux, and v1.02 elsewhere.Table 3 Genomic relationship matrix calculation times (sec) DatasetMachineGCTAPLINK 1.90Ratiosynth1pMac-2222.27.231Mac-12184.75.037Linux32-8248.410.922.8Linux64-5124.49.60.46Win32-2373.139.39.5Win64-2367.26.656synth2pMac-2nomem805.8Mac-1217.0 k138.3123Linux32-8nomem1153.4Linux64-51265.1318.90.20Win32-2nomem2007.2Win64-2nomem450.1chr1snpMac-2nomem87.1Mac-122260.950.944.4Linux32-8nomem94.3Linux64-51258.391.60.64Win32-2nomem317.5Win64-2nomem65.7This involves a variance-normalizing distance function which cannot be efficiently computed with just bit population counts. PLINK 1.9’s lookup table-based algorithm is slower than GCTA 1.24 on heavily multicore machines (see the Linux64-512 results), but has complementary advantages in portability, accuracy, and memory efficiency.
