Table 1 Computation resource required for different methodsa

From: Leveraging functional genomic annotations and genome coverage to improve polygenic prediction of complex traits within and between ancestries

Method (no. SNPs)

Runtime (h)

Memory (GB)

Required storage (GB)

SBayesRC (7 M)

8.5

73.3

72

LDpred-funct (7 M)

6.0

120.6

40–50 per trait

PolyPred-S (7 M)

19.8b

71.7

2,800

MegaPRS (7 M)

7.2

247.7

277

LDpred2 (1 M)

5.5

53.4

43

SBayesRC (1 M)

0.8

4.8

5.6

SBayesR (1 M)

0.5

27.0

22

PRS-CSx (1 M)

14.2c

4.7

5.6

MegaPRS (1 M)

0.2

11.6

7.4

  1. aResults were average values across traits using 4 CPU cores when multi-thread was supported for the method (n = 28 traits). Benchmarked with CPU AMD EPYC 7643 on a computing cluster.
  2. bIn PolyPred-S, fine-mapping is the most time-consuming step and is suggested to run by blocks in parallel. Here we used a single core and divided the runtime by 4 for comparison to others.
  3. cPer dataset runtime: total runtime/number of training datasets (=2).
Back to article page