Boolean matrix logic programming for active learning of gene functions in genome-scale metabolic network models
This repository contains the official experiment data of Boolean matrix logic programming for active learning of gene functions in genome-scale metabolic network models.
Our new system
We compared with flux-balance analysis (FBA) results of iML1515. Our logic-based simulation approach achieves phenotypic prediction accuracy 88.7% (FBA 93.4%).
Simulated phenotypic classifications are included in experiments/iML1515/sim_acc.
We have demonstrated that the general-purpose Prolog interpreter SWI-Prolog can be augmented with our BMLP approach for faster phenotypic simulations given logically encoded metabolic network models. There is a 170 times improvement in simulation time via our BMLP approach compared to base SWI-Prolog.
Runtime experiment results: /src/experiments/iML1515/runtime
We consider the experimental resource cost and the number of experiments as main factors to optimise. These two costs are related since the number of experiments generally is the main driver of experimental cost.
Active learning experiment results: /src/experiments/iML1515/abduction/output/gene_function_learning
Active learning experiment results: /src/experiments/iML1515/abduction/output/isoenz_learning
@misc{ai2024boolean,
title={Boolean matrix logic programming for active learning of gene functions in genome-scale metabolic network models},
author={Lun Ai and Stephen H. Muggleton and Shi-Shun Liang and Geoff S. Baldwin},
year={2024},
eprint={2405.06724},
archivePrefix={arXiv},
primaryClass={q-bio.MN}
}
Distributed under the MIT License. See LICENSE.txt for more information.
Dr. Lun Ai (corresponding author)
Department of Computing
Imperial College London
Address: 180 Queen’s Gate, SW7 2BZ London, UK
Email: [email protected]