Instead of relying on gene co-expression data, a technique using cell permeabilization and proximity biotinylation can identify the specific cellular machinery physically interacting with and supporting a biologic's production. This reveals critical chaperones and support proteins needed for high titers, moving beyond correlation to causation.
When expressing the human secretome in CHO cells, host cell gene expression correlated more strongly with productivity than the proteins' own structural features. This suggests the production cell line's inherent machinery and metabolic state can be a more dominant factor for success than the design of the biologic itself.
Contrary to the belief that glycosylation is a template-less process, research shows a protein's sequence directly influences its glycan structures. By introducing specific point mutations, developers can predictably tune critical features like fucosylation or sialylation, shifting this complex control problem into early-stage drug design.
The development of powerful foundation models to optimize bioprocessing is hampered less by technical challenges and more by the industry's reluctance to share data. The critical challenge is overcoming the cultural and legal hurdles within companies to create the large, diverse datasets necessary for transformative AI.