Table 2 Independent evaluation of ProteoCast on standard benchmarks compared to established methods

From: Proteome-wide prediction of the functional impact of missense variants with ProteoCast

ProteoCast functionality

Evaluation benchmark

Comparison methods

Supervised (Y/N)

Used the benchmark (Y/N)

Results summary

Variant classification

ClinVar82 (63k)

EVE8

N

Y

higher coverage for ProteoCast, similar AUC-ROC for both methods, slightly higher recall for EVE on a variant subset (Table S7, Fig. S15)

Functional site identification in unstructured regions

curated set from the yeast proteome55 (526)

phylo-HMM55

N

Y

substantially higher detection rate for ProteoCast (Table S8, Fig. S16)

 

CAID3 Binding-IDR56 (64)

top-10 CAID3 predictors

Y

Na

ProteoCast on par with the other predictors (Tales S9 and S10, Fig. S18)

  1. For each dataset, we indicate the ProteoCast functionality evaluated, the number of data points, the comparison methods used, whether these methods are supervised, and whether the evaluation benchmark was used in the original method publications (indicating potential use during development). Main results are summarised with references to corresponding tables and figures.
  2. aThe methods were trained on other CAID versions or other binding datasets.
Back to article page