Efficient large-scale single-step evaluations and indirect genomic prediction of genotyped selection candidates

Table 2 Computational statistics of ssGTABLUP and ssSNPBLUP using the full and reduced datasets

Computational statistic	Original ssGTABLUP		Component-wise ssGTABLUP		ssSNPBLUP
Computational statistic	Full	Reduced	Full	Reduced	Full	Reduced
T_a matrix (or its components)	1004	832	1000	829	–	–
RAM	31.0	23.6	22.3	18.1	–	–
^aTime (h)	31.0	23.6	22.3	18.1	–	–
Neq	372	366	372	366	372	367
Number of PCG iterations	488	595	478	584	828	872
\({\lambda }_{min}\)(10^–5)	4.77	3.13	4.98	3.24	3.42	2.75
\({\lambda }_{max}\)	2.98	2.98	2.98	2.98	3.59	3.34
RAM² (GB)	1015	845	102	95	86	79
Time/iteration (sec)	172.4	143.4	64.9	51.2	65.4	52.7
^bTime (h)	26.4	25.9	9.9	9.5	16.4	13.7
Total time (h)	62.2	52.8	37.0	32.5	21.9	18.8

Neq = number of equations in millions; RAM = software peak random access memory (RAM) defined as the peak resident set size (VmHWM) obtained from the Linux/proc virtual file system; \({\lambda }_{min}\) = smallest effective eigenvalue of the preconditioned coefficient matrix; \({\lambda }_{max}\) = largest effective eigenvalue of the preconditioned coefficient matrix; Time/iteration = Average wall clock time per PCG iteration (expressed in seconds); Total time = wall clock time of the MiXBLUP software expressed in hours
^aTime = wall clock time of the program calc_grm expressed in hours
^bTime = wall clock time of the solver expressed in hours

ISSN: 1297-9686