Skip to main content

Table 3 Overall descriptor set ranking

From: Benchmarking of protein descriptor sets in proteochemometric modeling (part 2): modeling performance of 13 amino acid descriptor sets

Type Final rank (MAD) Rank ACE Rank GPCR Rank NNRTI Rank PI
ProtFP (Feature) and Z-Scales (3) 4 (±2.0) 8.5 4 7.5 2
Z-Scales (3) and Z-Scales (Avg) 4 (±2.5) 11 1 8.5 4.5
Z-Scales (3) 4.5 (±1.5) 6 5 8.5 2.5
MS-WHIM 5.5 (±2.5) 4 8.5 7 4
Z-Scales (5) 6.5 (±3.5) 3 12.5 11.5 4.5
ProtFP (PCA3) 7 (±2.0) 10 9 5 7.5
Z-Scales (Binned) 7.5 (±2.5) 1.5 6.5 10 7.5
ProtFP (PCA5) 9 (±2.0) 15 5.5 5 9.5
FASGAI 9.5 (±2.5) 6.5 9.5 11 9.5
ProtFP (PCA3) and Z-Scales (Binned) 10 (±3.0) 1.5 8 11 12.5
T_Scales 10 (±2.0) 8.5 13 9 9
VHSE 10 (4.0) 5.5 9 10.5 10.5
BLOSUM 11.5 (±4.5) 13.5 8 10 7.5
ProtFP (PCA8) 12 (±2.0) 13.5 8 10.5 13.5
ST_Scales 13 (±1.0) 12 13.5 13 14.5
ProtFP (Feature) 16 (±0.0) 16 16 1 16
  1. The descriptor sets are sorted based on their median (final) rank calculated over all 14 benchmarks and the MAD associated with this value. Also shown is the rank each descriptor set receives in each individual data set (calculated over the four benchmarks per data set). The best performance is achieved by combining the Z-scales (3) with complementary descriptor types (ProtFP (Feature) or Z-scales (Avg)), closely followed by Z-Scales (3), MS-WHIM, and Z-Scales (5).