Discussion

In selecting our enzyme target, we were guided by the strong structural foundation established in recent studies. Notably, alginate lyases, as reported by Zhang et al. (2022), have been the subject of extensive structural and mechanistic investigations, providing a well-established theoretical framework. Within this context, PyAly emerges as an especially suitable candidate because it possesses a resolved crystal structure and clearly defined domains. This solid experimental and theoretical basis ensured that our technical approach was both reasonable and feasible. Building on these insights, our design strategy focused on structurally informed residue selection, with the goal of maximizing functional impact and specifically enhancing tetrasaccharide yield.

A central feature of our approach is following the guidance of rational design rather than random screening. Residue selection was informed by structural insights, ensuring that mutagenesis targeted positions with a high likelihood of functional impact and maintained a clear focus on improving tetrasaccharide production. Single-site mutations at these residues served as functional probes, allowing us to validate their local contributions. Building on these insights, we reasoned that rationally designed combinations would have a higher chance of generating synergistic improvements in yield. In this way, our design narrows the mutational search space, increases efficiency, and offers stronger predictability compared with conventional random mutagenesis.

Another key contribution of this study is the application of machine learning to predict high-performing mutation combinations and thereby reduce experimental workload. Multi-site mutations are generally expected to deliver stronger effects than single-site changes; however, the sheer number of possible combinations makes exhaustive experimental testing impractical. To address this challenge, machine learning was applied to prioritize variants with the highest predicted potential, thereby directing experimental resources toward the most informative candidates. Importantly, successful results from one round served as the starting point for the next. This iterative strategy demonstrates how machine learning enables stepwise optimization — progressively building on prior gains, reducing unnecessary trials, and accelerating the discovery of superior enzyme variants.

We developed a framework integrating rational design with machine learning in an iterative optimization cycle. This approach was validated through the construction of single and double mutants, which served the dual purpose of probing key residue functions and generating the foundational dataset for model training. The demonstrated predictive capability confirms our strategy's feasibility.

Future efforts will focus on enhancing this cycle's precision and efficiency. We plan to systematically expand the mutant library, prioritizing F170-containing combinations, while implementing standardized assays to improve data quality. To address current limitations, we will refine our predictive algorithms to better capture higherorder epistatic effects and improve accuracy across diverse mutation combinations. Streamlining experimental workflows will accelerate validation, strengthening the iterative feedback between computation and experimentation. This continuous refinement of both algorithms and experimental processes will establish a robust platform for developing enzyme variants with superior product specificity.

References

1. Bi, D., Yang, X., Yao, L., Hu, Z., Li, H., Xu, X., & Lu, J. (2022). Potential Food and Nutraceutical Applications of Alginate: A Review. Marine Drugs, 20(9), 564. DOI: https://doi.org/10.3390/md20090564
2. Gao, S.-K., Yin, R., Wang, X.-C., Jiang, H.-N., Liu, X.-X., Lv, W., Ma, Y., & Zhou, Y.-X. (2021). Structure Characteristics, Biochemical Properties, and Pharmaceutical Applications of Alginate Lyases. Marine Drugs, 19(11), 628. DOI: https://doi.org/10.3390/md19110628
3. Inoue, A., Mashino, C., Uji, T., Saga, N., Mikami, K., & Ojima, T. (2015). Characterization of an eukaryotic PL-7 Alginate Lyase in the Marine Red Alga Pyropia Yezoensis. Current Biotech- nology, 4(3), 240–258. DOI: https://doi.org/10.2174/2211550104666150915210434
4. Wang, M., Chen, L., & Zhang, Z. (2021). Potential applications of alginate oligosaccharides for biomedicine – A mini review. Carbohydrate Polymers, 271, 118408. DOI: https://doi.org/10.1016/j.carbpol.2021.118408
5. Liu, J., Yang, S., Li, X., Yan, Q., Reaney, M. J. T., & Jiang, Z. (2019). Alginate Oligosaccharides: Production, Biological Activities, and Potential Applications. Comprehensive Reviews in Food Science and Food Safety, 18(6), 1859–1881. DOI: https://doi.org/10.1111/1541-4337.12494
6. Zhang, K., Li, Z., Zhu, Q., Cao, H., He, X., Zhang, X.-H., Liu, W., & Lyu, Q. (2022). Determi- nation of oligosaccharide product distributions of PL7 alginate lyases by their structural elements. Communications Biology, 5(1). DOI: https://doi.org/10.1038/s42003-022-03721-1
7. Fersht, A. R., & Winter, G. (1984). Studying Enzyme-Substrate Interactions by Site-Directed Mutagenesis. In Specificity in Biological Interactions (pp. 123–132). Springer Netherlands. DOI: https://doi.org/10.1007/978-94-009-6457-0_7
8. Tripathi, N., Hérisson, J., & Faulon, J.-L. (2025). Machine learning in predictive biocatalysis: A comparative review of methods and applications. Biotechnology Advances, 84, 108698. DOI: https://doi.org/10.1016/j.biotechadv.2025.108698
9. Cheng D, Jiang C, Xu J, et al. Characteristics and applications of alginate lyases: A review[J]. International Journal of Biological Macromolecules, 2020, 164: 1304-1320. DOI: https://doi.org/10.1016/j.ijbiomac.2020.07.199
10.Jemimah, S., & Gromiha, M. M. (2018). Exploring additivity effects of double mutations on the binding affinity of protein-protein complexes. Proteins, 86(5), 536～547. DOI: https://doi.org/10.1002/prot.25472
11.Otwinowski, J., McCandlish, D. M., & Plotkin, J. B. (2018). Inferring the shape of global epi- stasis. Proceedings of the National Academy of Sciences of the United States of America, 115(32), E7550～E7558. DOI: https://doi.org/10.1073/pnas.1804015115
12.Reddy, G., & Desai, M. M. (2021). Global epistasis emerges from a generic model of a complex trait. eLife, 10, e64740. DOI: https://doi.org/10.7554/eLife.64740
13.Rivas-Fernández, J. P., Vuillemin, M., Pilgaard, B., Klau, L. J., Fredslund, F., Lund-Hanssen, C., Welner, D. H., Meyer, A. S., Morth, J. P., Meilleur, F., Aachmann, F. L., Rovira, C., & Wilkens, C. (2025). Unraveling the molecular mechanism of polysaccharide lyases for efficient alginate degradation. Nature Communications, 16(1), 2670. DOI: https://doi.org/10.1038/s41467-025-56754-5
14.Giguère, S., Marchand, M., Laviolette, F., Drouin, A., & Corbeil, J. (2013). Learning a peptide- protein binding affinity predictor with kernel ridge regression. BMC bioinformatics, 14, 82. DOI: https://doi.org/10.1186/1471-2105-14-82
15.Xu, Y., Verma, D., Sheridan, R. P., Liaw, A., Ma, J., Marshall, N. M., McIntosh, J., Sherer, E. C., Svetnik, V., & Johnston, J. M. (2020). Deep Dive into Machine Learning Models for Protein Engineering. Journal of chemical information and modeling, 60(6), 2773–2790. DOI: https://doi.org/10.1021/acs.jcim.0c00073