Skip to main content
. 2021 Nov 16;6(6):e00299-21. doi: 10.1128/mSystems.00299-21

TABLE 1.

Performance of protein-encoding methods with deep learninga

Encoding method Accuracy Precision Sensitivity Specificity F1 score MCC
Test data (706 sequences)
PC6 0.8895 0.9205 0.8527 0.9263 0.8853 0.7812
PC7 0.8782 0.9083 0.8414 0.9150 0.8735 0.7584
AC6 0.7040 0.7130 0.6827 0.7252 0.6975 0.4083
AC7 0.7507 0.7580 0.7365 0.7649 0.7471 0.5016
Embedding layer 0.8952 0.9091 0.8782 0.9122 0.8934 0.7908
Word2Vec 0.9065 0.9415 0.8669 0.9462 0.9027 0.8156
External testing data (1,130 sequences)
PC6 0.8850 0.9035 0.8620 0.9080 0.8822 0.7707
PC7 0.8610 0.8849 0.8301 0.8920 0.8566 0.7235
AC6 0.7818 0.8534 0.7009 0.8695 0.7697 0.5760
AC7 0.7311 0.8152 0.6248 0.8464 0.7074 0.4808
Embedding layer 0.8690 0.9096 0.8195 0.9186 0.8622 0.7417
Word2Vec 0.8460 0.8574 0.8301 0.8619 0.8435 0.6924
a

Top three ranked methods for each index are presented using text format: first in boldface, second with underline, third in normal text format, and all the rest in italic.