Tpgen: a language model for stable protein design with a specific topology structure

BMC Bioinformatics

Table 1 Evaluation of various treatment approaches and their respective scores for fine-tuning sequence stability prediction models

	Pretrain ^a	\(1.2-0.6\)^b	\(1.0-0.0\)^c	OriScale ^d	NoPredictor^e
Proportion of HHH	91.35%	93.82%	90.82%	91.42%	91.54%
Proportion of stable proteins ^f	63.63%	69.90%	64.35%	62.41%	62.51%
Average stability score of HHH	1.05	1.12	1.06	1.04	1.04

^a Model without fine-tuning
^b Fine-tuning with \(\lambda 1\)=1.2 and \(\lambda 2\)=0.6
^c Fine-tuning with \(\lambda 1\)=1.0 and \(\lambda 2\)=0.0
^e Fine-tuning achieved by directly multiplying the stability scores with the discriminative model scores
^dFine-tuning with only the discriminative model
^f Proportion of stable proteins in HHH

ISSN: 1471-2105