Computational optimization of cell culture media for cultured meat production

• المؤلف الرئيسي: Reji, Rinta Pullamkottu
• مؤلفون آخرون: -
• التنسيق: Final Year Project
• اللغة: English
• منشور في: Nanyang Technological University 2023
• الموضوعات: Science::Biological sciences
• الوصول للمادة أونلاين: https://hdl.handle.net/10356/166433

One of the biggest challenges in making cultured meat accessible is reducing the cost of culture media. This project outlines two methods of reducing cost of media: 1) optimisation of concentration of components and 2) discovery of alternative components. Optimisation of culture media components was modelled as a blackbox optimisation problem with test functions to replace in vitro optimisation. A comparison of algorithmic methods for media optimisation was done at low, medium and high dimension and found that the best performing method was surrogate-based optimisation with truncated differential evolution as the acquisition method. This method is currently under-utilised and thus could improve media optimisation. Protein hydrolysates are viable substitutes for serum in culture media. To classify potentially useful hydrolysates, a structure-based prediction graph convolutional network (GCN) was trained on predicted structural information from AlphaFold2. ESM-2 was trained on amino acid sequences for masked residue prediction in order to extract residue-level features to be embedded into the graphical input. Even though the GCN (without residue-level features) did not perform the classification task well, ESM-2 was able to achieve a phenomenal performance with ROC-AUC of 0.94. ESM-2 is deep enough to capture protein function and can be used to classify serum-equivalent hydrolysates.