Biochemical Knowledge Extraction through Similarity Search via Finetuned Open-Source Large Language Models

• Paulo do Carmo InfAI
• Marcos Gôlo USP
• Jonas Gwozdz Leipzig University of Applied Sciences
• Edgard Marx InfAI
• Ricardo Marcacini USP

DOI: https://doi.org/10.5753/eniac.2026.29814

Resumo

Natural products have played a central role in drug development, and organizing all these compounds remains a time-consuming, often manual task, motivating efforts to automatically extract biochemical knowledge. Previous approaches based on knowledge graph embeddings and the auxiliary use of Large Language Models (LLMs) have shown promising results. However, these methods still have limitations in distinguishing new from cited properties within the text and rely on proprietary models and constrained prompting strategies. Thus, we propose combining finetuning of open-source LLMs with similarity search. Finetuning adapts the models to the biochemical domain, while similarity search serves as a retrieval layer that maps generated outputs to biochemical candidates. This combination provides an advantage, since newly emerging compounds can be incorporated into the candidate space without requiring new fine-tuning. Results on the NatUKE benchmark show that the proposed strategy achieves state-of-the-art performance compared to open-source models.

Referências

Abdin, M., Aneja, J., Behl, H., Bubeck, S., Eldan, R., Gunasekar, S., Harrison, M., Hewett, R. J., Javaheripi, M., Kauffmann, P., et al. (2024). Phi-4 technical report.

Alibaba, T. (2024). Qwen2 technical report.

Atanasov, A. G., Zotchev, S. B., Dirsch, V. M., and Supuran, C. T. (2021). Natural products in drug discovery: advances and opportunities. Nature reviews Drug discovery.

Bitelli, B. and Finger, M. (2025). Numerical information extraction in legal texts using open and closed Large Language Models. In Anais do XXII Encontro Nacional de Inteligência Artificial e Computacional, pages 1197–1208, Porto Alegre, RS, Brasil.

Blecher, L., Cucurull, G., Scialom, T., and Stojnic, R. (2023). Nougat: Neural optical understanding for academic documents.

Chand, B. and Tiwari, S. (2025). Sphota: Knowledge graph structure prediction with a hybrid orientation of textual alignment using k-bert. In Proceedings of the Second Biochemical Knowledge Extraction challenge with Text2KG at ESWC.

Dettmers, T., Pagnoni, A., Holtzman, A., and Zettlemoyer, L. (2023). Qlora: Efficient finetuning of quantized llms. In Oh, A., Naumann, T., Globerson, A., Saenko, K., Hardt, M., and Levine, S., editors, Advances in Neural Information Processing Systems, volume 36, pages 10088–10115. Curran Associates, Inc.

Dichte, S. S. and Mócsy, I. J. (2023). Improving natural product automatic extraction with named entity recognition. In Proceedings of the First Biochemical Knowledge Extraction (BiKE) challenge co-located with Text2KG at ESWC, pages 226–234. CEUR-WS Proceedings.

Do Carmo, P. and Marcacini, R. (2021). Embedding propagation over heterogeneous event networks for link prediction. In 2021 IEEE International Conference on Big Data (Big Data), pages 4812–4821. IEEE.

Do Carmo, P. V., Marx, E., Marcacini, R., Valli, M., Silva e Silva, J. V., and Pilon, A. (2023). Natuke: A benchmark for natural product knowledge extraction from academic literature. In 2023 IEEE 17th International Conference on Semantic Computing (ICSC), pages 199–203.

Do Carmo, P. V., Silva Gôlo, M. P., Gwozdz, J., Marx, E., and Marcondes Marcacini, R. (2025). Improving natural product knowledge extraction from academic literature with enhanced pdf text extraction and large language models. In Proceedings of the 40th ACM/SIGAPP Symposium on Applied Computing, pages 980–987.

D’Souza, J., Laubach, Z., Mustafa, T. A., Zarrieß, S., Frühstückl, R., and Illari, P. (2025). Mining for species, locations, habitats, and ecosystems from scientific papers in invasion biology: A large-scale exploratory study with large language models. arXiv preprint arXiv:2501.18287.

Fröhlich, P., Gwozdz, J., and Jooß, M. (2023). Leveraging chatgpt api for enhanced data preprocessing in natuke. In Proceedings of the First Biochemical Knowledge Extraction (BiKE) challenge co-located with Text2KG at ESWC, pages 244–255. CEUR-WS Proceedings.

Gôlo, M. P. S., Medeiros Júnior, J. G. B. d., Vilas Boas, G. S., Lobato, F. M. F., Silva, D. F., and Marcacini, R. M. (2025). Large language models ensemble for biochemical properties detection in scientific articles. In Proceedings of the Second Biochemical Knowledge Extraction (BiKE) challenge co-located with Text2KG at ESWC.

Gozzi, M. and Di Maio, F. (2024). Comparative analysis of prompt strategies for large language models: Single-task vs. multitask prompts. Electronics, 13(23):4712.

Hu, E. J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., Chen, W., et al. (2022). Lora: Low-rank adaptation of large language models. ICLR, 1(2):3.

Marx, E., Valli, M., da Silva e Silva, J., and do Carmo, P. V. (2025). Preface of the second international biochemical knowledge extraction challenge (bike). In LLM-Integrated Knowledge Graph Generation From Text and International BiKE Challenge, volume Vol-4020 of CEUR Workshop Proceedings. CEUR-WS.org.

Marx, E., Valli, M., da Silva e Silva, J., Tiwari, S., and do Carmo, P. (2023). Preface of the first international biochemical knowledge extraction challenge (bike). In Joint Proceedings of TEXT2KG 2023 and BiKE 2023, volume Vol-3447 of CEUR Workshop Proceedings. CEUR-WS.org.

Meta (2024). The llama 3 herd of models.

Newman, D. J. and Cragg, G. M. (2020). Natural products as sources of new drugs over the nearly four decades from 01/1981 to 09/2019. Journal of natural products.

Oliveira, M. S., Pio, P. B., and Garcia, L. P. (2025). Generative slms meet brazilian legal documents: Efficient ner via lora fine-tuning. In Brazilian Conference on Intelligent Systems, pages 586–599. Springer.

Pilon, A. C., Valli, M., Dametto, A. C., Pinto, M. E. F., Freire, R. T., Castro-Gamboa, I., Andricopulo, A. D., and Bolzani, V. S. (2017). Nubbedb: an updated database to uncover chemical and biological information from brazilian biodiversity. Scientific Reports, 7(1):7215.

Réda, C., Kaufmann, E., and Delahaye-Duriez, A. (2020). Machine learning applications in drug development. Computational and structural biotechnology journal, 18.

Zhai, C. (2024). Large language models and future of information retrieval: opportunities and challenges. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 481–490.

Zhang, L., Song, J., Kong, L., Yuan, T., Li, W., Zhang, W., Hou, B., Lu, Y., and Du, G. (2020). The strategies and techniques of drug discovery from natural products. Pharmacology & Therapeutics, 216.

Zope, B., Mishra, S., and Tiwari, S. (2023). Enhancing biochemical extraction with bfs-driven knowledge graph embedding approach. In Proceedings of the First Biochemical Knowledge Extraction (BiKE) challenge co-located with Text2KG at ESWC. CEUR-WS Proceedings.