I don’t think the EAS PRS performances shown in Figure 5 yet justify that statement, which seems to suggest that the best available EAS PRS models now meet or exceed the best available EUR ones. Since these PRSmix+ EAS PRSs were trained and tuned on only 363k + 80k Taiwanese TPMI samples, it’s still the case that EUR PRSs trained with comparable methods but with much larger GWAS data yield (much) better performance in similarly well-phenotyped EUR test sets. For example, this new PRSmix+, TPMI-trained EAS T2D PRS achieves liability-scale R^2 = 8.5% (Table.S14) in TPMI using data from 89k EAS T2D cases and 325k EAS T2D controls (Table.S1), whereas a Herasight EUR T2D PRS trained using GWAS associations published by Suzuki et al. [1] and derived from 254k EUR T2D cases + 1.49m EUR T2D controls achieves liability-scale R^2 = 20.7% (16.8, 24.8) in the UK Biobank [2]. (Other groups have similarly trained highly predictive EUR T2D PRS [3] exceeding the R^2 = 8.5% threshold set here). But that’s not to say that greatly improved EAS PRSs can’t be trained by combining this new EAS GWAS with other available EAS GWASs! For example, 20% of the effective sample size of the massive Suzuki et al. multi-ancestry T2D GWAS is contributed by non-TPMI EAS sources. New results on this coming soon!