Table 2.
Accuracy computed for the CATH annotation transfer analysis as in Heinzinger et al. (2022).a
ProstT5 |
ProtT5 |
ESM-1b |
Not pLMs-based |
|||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
EBA | EBAplain | AD | EBA | EBAplain | ProtTucker | TM-vec | AD | EBA | EBAplain | AD | Foldseek | HMMER | MMseqs2 | |
C | 91 | 79 | 86 | 87 | 82 | 88 | 89 | 84 | 89 | 75 | 79 | 77 | 70 | 53 |
A | 84 | 66 | 75 | 77 | 68 | 77 | 80 | 67 | 78 | 59 | 61 | 73 | 60 | 33 |
T | 78 | 55 | 63 | 74 | 63 | 68 | 71 | 57 | 70 | 54 | 50 | 59 | 59 | 21 |
H | 88 | 61 | 67 | 85 | 74 | 79 | 81 | 64 | 77 | 61 | 57 | 77 | 77 | 25 |
The reported EBA and EBAplain values are normalized according to the length of the longer sequence in each comparison: EBAmin. Best performance for each CATH category in bold.