Table 4 Comparison of SPIRED-Stab with other ΔTm predictors on the S557 dataset

From: An end-to-end framework for the prediction of protein structure and fitness from single sequence

Method

ρ

τ

r

Top-5

Top-10

Structure-based

     

AUTO-MUTE

0.23  ± 0.38

0.19  ± 0.32

0.27  ± 0.37

0.33  ± 0.29

0.61  ± 0.32

HoTMuSiC

0.32  ± 0.31

0.24  ± 0.24

0.30  ± 0.29

0.36  ± 0.18

0.64  ± 0.22

GeoDTm-3D (old)

0.40  ± 0.35

0.29  ± 0.27

0.39  ± 0.35

0.40  ± 0.22

0.69  ± 0.23

Sequence-based

     

GeoDTm-Seq (old)

0.45  ± 0.27

0.33  ± 0.23

0.41  ± 0.30

0.46  ± 0.27

0.66  ± 0.27

Mutate everything (ESM-2)

0.40  ± 0.31

0.31  ± 0.26

0.38  ± 0.33

0.39  ± 0.29

0.60  ± 0.30

SPIRED-Fitness (zero-shot)

0.46  ± 0.29

0.35  ± 0.25

0.47  ± 0.29

0.46  ± 0.34

0.62  ± 0.27

GeoStab v2

0.49  ± 0.34

0.38  ± 0.29

0.44  ± 0.43

0.44  ± 0.30

0.69  ± 0.27

SPIRED-Stab

0.50  ± 0.36

0.39  ± 0.30

0.43  ± 0.39

0.47  ± 0.30

0.69  ± 0.28

  1. All metrics are calculated within each protein and then reported as mean  ± standard-deviation over all tested proteins. All prediction methods are classified into a number of categories (shown in italic). For each metric, the winner among all methods is chosen based on the mean value and is highlighted in bold. Here, ρ, τ and r denote Spearman, Kendall and Pearson correlation coefficients, respectively, while Top-5 and Top-10 refer to the Top 5 precision and Top 10 precision, respectively, as defined in “Methods” section.
Back to article page