Arvind, Siddhanta (2026) Modelling heterologous expression in Artificial Metabolic Networks: Introduction of gene knockins to hybrid metabolic networks. Master's Research Project 2, Biomolecular Sciences.
|
Text
mBioMol2026SiddhantaArvind.pdf Download (2MB) | Preview |
Abstract
With the increased availability of high-throughput data, improving constraint based modelling methods to incorporate machine learning is the next step in metabolic modelling. Artificial Metabolic Networks (AMN) embed a differentiable FBA style surrogate solver within a neural network and supports reaction knockouts, but industrial applications require introducing gene knockins. This research builds on AMN models to incorporate gene knockins to AMN and tests the performance on two E. coli genome scale models. Three reactions are added to the GSM models and tested on AMN model by passing the knockin reaction as a secondary objective alongside the biomass reaction. On the validation dataset, the model is able to accurately predict knockin fluxes with a Q2 of approximately 0.95, with biomass flux predictions at a Q2 of 0.84 on hyper-parameter tuned AMN model and 0.66 on the other. When knockouts and knockins are performed together, knockin performance is retained (Q2 of 0.94) but biomass prediction performance drops of 0.66. This fall in performance is observed also in the second AMN model, on which no hyper-parameter tuning was performed. The AMN models are therefore able to predict reliably across both metabolic models and AMN model designs, but the accuracy of biomass prediction is limited by the difference in magnitude between the objectives.
| Item Type: | Thesis (Master's Research Project 2) |
|---|---|
| Supervisor name: | Jong, A. de |
| Degree programme: | Biomolecular Sciences |
| Thesis type: | Master's Research Project 2 |
| Language: | English |
| Date Deposited: | 25 Aug 2026 13:52 |
| Last Modified: | 25 Aug 2026 13:52 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/37512 |
Actions (login required)
 |
View Item |