Table 1 Similar ratios of neutral/effect predicted by four methods for 1KG SAVs*.

From: Common sequence variants affect molecular function more than rare variants?

SAV set SNAP2* CADD PolyPhen-2 SIFT
effect/all common 61% 19% 18% 19%
rare 50% 50% 42% 40%
all 51% 46% 39% 47%
neutral/all common 39% 81% 82% 81%
rare 50% 50% 58% 60%
all 49% 54% 61% 63%
  1. *Data sets: Rare SAVs (LDAF < 0.01): all methods agreed within ten percentage points on the ratio; common SAVs (LDAF ≥ 0.05); the values for “all” also included uncommon SAVs (0.01 ≤ LDAF < 0.05). Methods: SNAP2* implies error-corrected estimates for SNAP2 (below). The four methods compared here have different aims and use different thresholds. Here, we applied defaults to simplify the comparison and interpreted predictions as binary (effect/neutral) according to those thresholds. In particular, we chose the following thresholds. Effect: SNAP2-score > 0, CADDv1.3 raw score > 3, PolyPhen-2 = probably or possibly damaging, SIFT = deleterious; neutral: SNAP2-score ≤ 0, CADDv1.3 raw score ≤ 3, PolyPhen-2 = benign, SIFT = tolerated. SNAP2* error correction: Values for SNAP2 were corrected for false positives and false negatives (e.g. Neff* = Neff(raw) − FPR(Neff(raw)) + FNR(Nneu(raw)). Error correction lowered the estimates for common effect and increased that for rare effect.