Table 1 Scores of the model for recapitulating the test set.

From: Neural network conditioned to produce thermophilic protein sequences can increase thermal stability

Test Metric

Value

Type

Cross Entropy Loss

0.90

Residue

MSA Cross Entropy Loss

0.94

Natural*, Residue

Transcription Error Rate

47%

Sequence

Sequence Identity

43%

Sequence

Bits per residue, BLOSUM6238

2.4

Sequence

Jenson-Shannon Secondary Structure

0.01

Structure

FATCAT Structural Alignment P-value

0.01

Structure

  1. “Residue” scores are computed on a per-amino acid basis, “Sequence” are computed by comparing full sequences, and “Structure” are derived from comparison of ESMFold structure prediction. *Does not use the NOMELT model, but instead natural variation over thermophilic homologs.
Back to article page