In recent years, there have been huge advances in protein structure prediction methods, which have given us access to vast amounts of highly accurate structural data for previously intractable targets. We have found that properties derived from these models can be used to identify antibody designs that were highly produced in cells, as well as highlighted fundamental differences between natural and designed proteins. Finally, we uncovered systematic variations between the average properties of proteins in a range of organisms to such an extent that, when clustered using these data alone, we partially recovered of the tree of life.

Selection and/or design of signal peptide components is complicated by their product-specific functionality. This talk will introduce our signal peptide design platform, which can forward engineer-optimised, synthetic solutions for any new protein of interest. Underpinned by data from a wide range of cellular and molecular contexts, this tool enables precise predictable control of product translocation rates, facilitating significant increases in recombinant protein titers.

Cell-free expression (CFE) platforms with an eukaryotic origin commonly suffer from low protein yield. The performance relies on the function of numerous enzymes involved in the molecular pathways of protein synthesis. We describe the development and optimisation of a Chinese hamster ovary cell (CHO) derived CFE system. The optimised CFE system reaches the yields of commercially available HeLa based lysates and represents a considerable advancement in CHO cell-based lysate systems.

Construct design towards soluble protein fragments for biochemical, biophysical, and structural analyses has been greatly facilitated by algorithms predicting features such as domains, disorder, and secondary elements. The recent advent of AI tools such as AlphaFold2, however, has transformed in silico structural biology. Here we present practical tips for using bioinformatics and AI tools in construct design to help users improve the likelihood of obtaining functional proteins for their needs.

Within the crowded environment of the cell, quality control machinery is vital for correct spatial and temporal protein distribution. I will discuss the optimisation of design and production for a variety of protein constructs that have allowed us to investigate these mechanisms and understand some of their roles in health and disease.

Identification of domain boundaries for optimal expression of proteins is essential for early drug discovery. We have developed and implemented a machine learning model to predict protein expression. The model was coupled to an in silico screening procedure that systematically designs and assesses thousands of constructs in a high-throughput manner. We will share how this is being used within our protein production platforms at AstraZeneca and some of the challenges faced.

We have developed a machine learning model to predict recombinant protein expression, using a combination of in-house experiment results and publicly available data sets from SGC. Heterogeneity of these data sets presented a challenge during model development. Using a tailored model architecture and training algorithm has yielded an improvement in Area Under ROC Curve of X%. The model was experimentally validated on a carefully selected set of proteins.

Machine learning (ML) is an emerging technology with practical applications in improving cell culture in biotechnology such as protein production. Our research delves into the integration of ML techniques to enhance cell culture, demonstrating that ML can efficiently optimise culture media for bacterial or mammalian cells to increase cell growth and production. These success stories serve as compelling evidence of ML's potential to drive innovation in industry and research.