Can large language models understand molecules?

BMC Bioinformatics

Table 6 Results on link prediction tasks

Models	AUROC	AUPR	AUROC	AUPR
Morgan FP	0.871 ± 0.00	0.847 ± 0.00	0.876 ± 0.00	0.855 ± 0.00
BERT	0.621 ± 0.02	0.563 ± 0.08	0.660 ± 0.02	0.639 ± 0.01
ChemBERTa	0.527 ± 0.02	0.547 ± 0.08	0.519 ± 0.02	0.457 ± 0.01
MolFormer-XL	0.550 ± 0.02	0.701 ± 0.08	0.611 ± 0.02	0.644 ± 0.01
GPT	0.856 ± 0.06	0.812 ± 0.08	0.836 ± 0.05	0.748 ± 0.09
LLaMA	0.921 ± 0.00	0.884 ± 0.02	0.927 ± 0.00	0.872 ± 0.01
LLaMA2	0.941 ± 0.00	0.902 ± 0.02	0.961 ± 0.00	0.933 ± 0.01

The reported performance metrics are the mean and standard deviation of the AUROC and AUPR, calculated across the 10 runs. The Best Performance is Highlighted in Bold

ISSN: 1471-2105