Leveraging multi-modal feature learning for predictions of antibody viscosity

mAbs, Год журнала: 2025, Номер 17(1)

Опубликована: Апрель 11, 2025

The shift toward subcutaneous administration for biologic therapeutics has gained momentum due to its patient-friendly nature, convenience, reduced healthcare burden, and improved compliance compared traditional intravenous infusions. However, a significant challenge associated with this transition is managing the viscosity of administered solutions. High poses substantial development manufacturability challenges, directly affecting patients by increasing injection time causing pain at site. Furthermore, high formulations can prolong residence site, absorption kinetics potentially altering intended pharmacological profile therapeutic efficacy candidate. Here, we report application multimodal feature learning workflow predicting antibodies in discovery. It integrates multiple data sources including sequence, structural, physicochemical properties, as well embeddings from language model. This approach enables model learn various underlying rules, such rules molecular simulations protein evolutionary patterns captured large, pre-trained deep models. By comparing effectiveness other selected published prediction methods, study provides insights into their intrinsic predictive potential usability early-stage antibody pipelines.

Язык: Английский

Computational pipeline for sustainable enzyme discovery through (re)use of metagenomic data

Journal of Environmental Management, Год журнала: 2025, Номер 382, С. 125381 - 125381

Опубликована: Апрель 18, 2025

Enzymes derived from extremophilic organisms, also known as extremozymes, offer sustainable and efficient solutions for industrial applications. Valued their resilience low environmental impact, extremozymes have found use catalysts in various processes, ranging dairy production to pharmaceutical manufacturing. However, discovery of novel is often hindered by challenges such culturing difficulties, underrepresentation extreme environments reference databases, limitations traditional sequence-based screening methods. In this work, we present a computational pipeline designed discover enzymes metagenomic data environments. This represents versatile approach that promotes reuse recycling existing datasets minimises the need additional sampling. its core, algorithm integrates both bioinformatic techniques recent advances structural prediction, enabling rapid accurate identification enzymes. due design, relies heavily on which can limit effectiveness situations where scarce or when encountering protein families. As proof-of-concept, applied deep-sea hydrothermal vents, with focus β-galactosidases. The identified 11 potential candidate proteins, out 10 showed vitro activity. One selected enzymes, βGal_UW07, strong enzyme exhibited optimal activity at 70 °C was exceptionally resistant high pH presence metal ions reducing agents. Overall, our results indicate highly play key role bioprospecting, leveraging minimising situ interventions pristine regions.

Язык: Английский