Text Based De Novo Molecule Generation On

評価指標

BLEU

Exact Match

Frechet ChemNet Distance (FCD)

Levenshtein

MACCS FTS

Morgan FTS

Parameter Count

RDK FTS

Text2Mol

Validity

評価結果

このベンチマークにおける各モデルのパフォーマンス結果

モデル名	BLEU	Exact Match	Frechet ChemNet Distance (FCD)	Levenshtein	MACCS FTS	Morgan FTS	Parameter Count	RDK FTS	Text2Mol	Validity	Paper Title	Repository
MolT5-Large	85.4	30.2	1.20	16.07	83.4	68.4	770000000	74.6	55.4	90.5	Translation between Molecules and Natural Language
GIT-Mol-caption	75.6	5.1	-	26.315	73.8	51.9	-	58.2	-	92.8	GIT-Mol: A Multi-modal Large Language Model for Molecular Science with Graph, Image, and Text
Text+Chem T5 base	75	21.2	0.061	27.39	87.4	69.7	220000000	76.7	-	79.2	Unifying Molecular and Textual Representations via Multi-task Language Modelling
MolReGPT (GPT-4-0413)	85.7	28.0	0.41	17.14	90.3	73.9	None	80.5	59.3	89.9	Empowering Molecule Discovery for Molecule-Caption Translation with Large Language Models: A ChatGPT Perspective	-
BioT5+	87.2	52.2	0.353	12.776	90.7	77.9	252000000	83.5	57.9	100	BioT5+: Towards Generalized Biological Understanding with IUPAC Integration and Multi-task Tuning
LDMol	92.6	53.3	0.20	6.750	97.3	93.1	-	95.0	-	94.1	LDMol: Text-to-Molecule Diffusion Model with Structurally Informative Latent Space
MolT5-small	75.5	7.9	2.49	25.988	70.3	51.7	60000000	56.8	48.2	72.1	Translation between Molecules and Natural Language
MolReFlect	90.3	51.0	-	11.84	92.9	81.3	-	86.0	-	97.7	MolReFlect: Towards Fine-grained In-Context Alignment between Molecules and Texts	-
BioT5	86.7	41.3	.43	15.097	88.6	73.4	252000000	80.1	57.6	100	BioT5: Enriching Cross-modal Integration in Biology with Chemical Knowledge and Natural Language Associations
Text+Chem T5 small	73.9	15.7	0.066	28.54	85.9	66	60000000	73.6	-	77.6	Unifying Molecular and Textual Representations via Multi-task Language Modelling
MolFM-Small	80.3	16.9	-	20.868	83.4	72.1	13620000	66.2	57.3	85.9	MolFM: A Multimodal Molecular Foundation Model
MolReGPT (GPT-3.5-turbo)	79.0	13.9	0.57	24.91	84.7	62.4	-	70.8	57.1	88.7	Empowering Molecule Discovery for Molecule-Caption Translation with Large Language Models: A ChatGPT Perspective	-
Text+Chem T5-augm small	81.5	19.1	0.06	21.78	86.4	67.2	60000000	74.4	-	95.1	Unifying Molecular and Textual Representations via Multi-task Language Modelling
TGM-DLM	82.6	24.2	0.77	17.003	85.4	68.8	180000000	73.9	58.1	87.1	Text-Guided Molecule Generation with Diffusion Language Model
MolXPT	-	21.5	0.45	-	85.9	66.7	350000000	75.7	57.8	98.3	MolXPT: Wrapping Molecules with Text for Generative Pre-training	-
MolT5-Large-HV	81.0	31.4	0.44	16.758	87.2	72.2	770000000	78.6	59.0	99.6	Translation between Molecules and Natural Language
MolT5-base	76.9	8.1	2.18	24.458	72.1	52.9	220000000	58.8	49.6	77.2	Translation between Molecules and Natural Language
MolFM-Base	82.2	21.0	-	19.445	85.4	75.8	296200000	69.7	58.3	89.2	MolFM: A Multimodal Molecular Foundation Model
Text+Chem T5-augm base	85.3	32.2	.05	16.87	90.1	75.7	220000000	81.6	-	94.3	Unifying Molecular and Textual Representations via Multi-task Language Modelling
TGM-DLM w/o corr	82.8	24.2	0.89	16.897	87.4	72.2	180000000	77.1	58.9	78.9	Text-Guided Molecule Generation with Diffusion Language Model

0 of 20 row(s) selected.