Figure 2 shows the run time and memory usage of PRSice-2, lassosum, and LDpred. Based on these simulation results, PRSice-2 is the most efficient software in all settings (Fig. 2a), significantly faster than lassosum (P = 1e−58, 1-sided t-test) and LDpred (P = 2e−90, 1-sided t-test). Specifically, PRSice-2 can complete the full PRS analysis on 100,000 samples within 4 minutes (Supplementary Table 1), which is 179 times faster than the 10 hours required by lassosum and 241 times faster than the 13 hours 27 minutes required by LDpred. Likewise, PRSice-2 requires significantly less memory (Fig. 2b) than lassosum (P = 1e−202, 1-sided t-test) and LDpred (P = 9e−112, 1-sided t-test), requiring <500 MB of memory for 100,000 samples, as opposed to 11.2 GB required by lassosum and 45.2 GB required by LDpred (Supplementary Table 2). Likewise, PRSice-2 outperforms PRSice-v1.25, requiring 400 times less time and 8 times less memory for a target sample size of 10,000 (similar memory for small target samples; see Supplementary Fig. 1 and Supplementary Tables 1 and 2 for details). As data size grows, or when more sophisticated PRS analyses are performed at scale [5,23], these gains in computational efficiency could become even more important.
