Papers › Automatic Annotation Augmentation Boosts Translation between Molecules and Natural Language

Automatic Annotation Augmentation Boosts Translation between Molecules and Natural Language

10 Feb 2025arXiv:2502.06634archive 2025-07-28

Zhiqiang Zhong, Simon Sataa-Yu Larsen, Haoyu Guo, Tao Tang, Kuangyu Zhou, Davide Mottin

Recent advancements in AI for biological research focus on integrating molecular data with natural language to accelerate drug discovery. However, the scarcity of high-quality annotations limits progress in this area. This paper introduces LA³, a Language-based Automatic Annotation Augmentation framework that leverages large language models to augment existing datasets, thereby improving AI training. We demonstrate the effectiveness of LA³ by creating an enhanced dataset, LaChEBI-20, where we systematically rewrite the annotations of molecules from an established dataset. These rewritten annotations preserve essential molecular information while providing more varied sentence structures and vocabulary. Using LaChEBI-20, we train LaMolT5 based on a benchmark architecture to learn the mapping between molecular representations and augmented annotations. Experimental results on text-based *de novo* molecule generation and molecule captioning demonstrate that LaMolT5 outperforms state-of-the-art models. Notably, incorporating LA³ leads to improvements of up to 301% over the benchmark architecture. Furthermore, we validate the effectiveness of LA³ notable applications in *image*, *text* and *graph* tasks, affirming its versatility and utility.

PaperPDFCode

Code

zhiqiangzhongddu/la3 mentioned in paper report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Drug DiscoveryMolecule CaptioningSentenceText-based de novo Molecule Generation

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Molecule Captioning ChEBI-20 LaMolT5-Large BLEU-2 60.2 #12 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Large BLEU-4 52.1 #12 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Large METEOR 63.4 #12 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Large ROUGE-1 65.5 #12 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Large ROUGE-2 51.2 #12 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Large ROUGE-L 59.8 #12 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Large Text2Mol 59.7 #12 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Base BLEU-2 57.4 #20 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Base BLEU-4 48.5 #20 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Base METEOR 59.6 #20 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Base ROUGE-1 63.4 #20 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Base ROUGE-2 47.8 #20 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Base ROUGE-L 56.4 #20 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Base Text2Mol 59.9 #20 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Small BLEU-2 53.9 #29 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Small BLEU-4 44.6 #29 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Small METEOR 56.6 #29 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Small ROUGE-1 62.0 #29 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Small ROUGE-2 46.9 #29 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Small ROUGE-L 56.3 #29 of 33 Archive leaderboard report
Molecule Captioning ChEBI-20 LaMolT5-Small Text2Mol 58.8 #29 of 33 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Focus

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections