Protein language models can capture protein quaternary state

BMC Bioinformatics

Table 1 Performance of different models for the prediction of quaternary states (qs). Balanced Accuracy and F1 scores

		Annotation transfer, based on		pLM MLP model, based on
		Sequence	pLM	ESM-2 embeddings (QUEEN)	Protbert embeddings
No information about sequence homologs available ¹	BA²	0.15	0.23	0.36	0.19
No information about sequence homologs available ¹	F1³	0.43	0.54	0.52	0.41
Full homology information available	BA	0.6	0.67	–	–
Full homology information available	F1	0.79	0.85

¹ no sequence with > 30% sequence available to transfer from; ²BA: balanced accuracy; ³ F1: F1 score; Precision and Recall values are provided in Additional file 3: Table S3, and Precision-Recall and ROC curves are provided in Additional file 3: Figure S3. See Methods for definitions

ISSN: 1471-2105