Computational Enzyme Engineering for Sustainable Biotechnology

Sammanfattning

Recent developments in computational biology have introduced novel strategies for protein engineering, notably in enhancing the functionality of enzymes like Rubisco—a key target in the quest for climate-resilient solutions. This study presents a computational pipeline for assessing protein features and refining DL-designed (deep learning designed) Rubisco variants. The workflow combines bioinformatics tools such as sequence alignment and structural modeling to evaluate candidates based on physicochemical traits (e.g., hydrophobicity, isoelectric point), structural properties (e.g., TM-score, intrinsic disorder), and functional attributes (e.g., CO₂ binding affinity). Although the datasets of DL-generated variants exhibited greater variability in features such as sequence length and molecular weight relative to natural Rubisco, several still showed similar stability and CO₂ affinity. Unsupervised machine learning (ML) methods like principal component analysis (PCA) and clustering uncovered distinctive patterns between synthetic and natural sequences, underlining the strengths and current limitations of VAE-based enzyme design. The pipeline provides a scalable approach for protein variant evaluation in sustainable biotechnology contexts.