Microbial communities: network reconstruction and control
<p>Microbial communities, prevalent in natural and artificial settings, have garnered increasing attention for their applications in environmental enhancement, bio-engineering, and biomedicine. This thesis embarks on an exploration of microbial networks, focusing on the evaluation of existing...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
2024
|
| Subjects: |
| _version_ | 1824458713947176960 |
|---|---|
| author | Fu, A |
| author2 | Yang, A |
| author_facet | Yang, A Fu, A |
| author_sort | Fu, A |
| collection | OXFORD |
| description | <p>Microbial communities, prevalent in natural and artificial settings, have garnered increasing attention for their applications in environmental enhancement, bio-engineering, and biomedicine. This thesis embarks on an exploration of microbial networks, focusing on the evaluation of existing reconstruction methods, the integration of network inference with causal inference, and the use of reinforcement learning for network control.</p>
<p>A critical component of this research is the comparison of existing microbial network reconstruction methods. These methods are evaluated based on various criteria, such as data availability and network sparsity, highlighting their assumptions, advantages, and limitations. This evaluation not only provides a guide for method selection but also discusses the challenges in microbial interaction studies and suggests future research directions.</p>
<p>The thesis then introduces the Microbial Causal Inference (MCI) framework, designed to infer microbial species interactions, particularly in scenarios with feedback loops. MCI employs structural causal models to accurately represent microbial community dynamics under equilibrium conditions. Its efficacy is demonstrated in both acyclic and cyclic cases, including feedback loops, using algorithms such as Fast Causal Inference (FCI) and Greedy Fast Causal Inference (GFCI). The framework is validated through systematic analysis using simulated data, showcasing its applicability to different community parameters.</p>
<p>Addressing the limitations of observational equilibrium data, the thesis proposes a set of rules and algorithms for designing experiments to obtain interventional data. These experiments are crucial for transforming Partial Ancestral Graph (PAG)-inferred structures into fully identified causal models. This approach not only enhances the understanding of microbial networks but also guides the experimental process in confirming or refining inferred causal relationships.</p>
<p>Further, the thesis explores network control within microbial communities, introducing a novel reinforcement learning-based method for developing control algorithms. This venture into the application of reinforcement learning for network control opens new avenues for research and potential breakthroughs in this domain.</p>
<p>The novelty of this research lies in the development and integration of the MCI framework with existing methods, providing a robust approach to infer microbial interactions under complex conditions such as feedback loops. This framework uniquely bridges network reconstruction and causal inference, validated with innovative algorithms (FCI and GFCI). This thesis extends this novelty by addressing the challenge of inferring complete causal structures from observational equilibrium data in the presence of feedback loops. It proposes adaptive learning methods and experimental design rules to transform PAG-inferred structures into fully identified causal models, thus enhancing our understanding of microbial dynamics and providing a systematic approach for future research in causal inference within complex biological systems. Additionally, the introduction of reinforcement learning for network control is a pioneering step, offering new strategies for manipulating microbial networks.</p>
<p>The thesis concludes with a summary of the research findings, discussing the limitations and potential future directions. It emphasizes the need for ongoing development and refinement of methodologies to better understand and manipulate microbial community networks. The thesis posits that the integration of network reconstruction, causal inference, and reinforcement learning provides a comprehensive approach to studying and controlling microbial communities.</p> |
| first_indexed | 2025-02-19T04:30:16Z |
| format | Thesis |
| id | oxford-uuid:9878f61b-71bf-4675-95a8-7ca383aeb1dd |
| institution | University of Oxford |
| language | English |
| last_indexed | 2025-02-19T04:30:16Z |
| publishDate | 2024 |
| record_format | dspace |
| spelling | oxford-uuid:9878f61b-71bf-4675-95a8-7ca383aeb1dd2024-12-23T07:18:29ZMicrobial communities: network reconstruction and controlThesishttp://purl.org/coar/resource_type/c_db06uuid:9878f61b-71bf-4675-95a8-7ca383aeb1ddComputational biologyMicrobial ecologyExperimental designCausal inferenceEnglishHyrax Deposit2024Fu, AYang, A<p>Microbial communities, prevalent in natural and artificial settings, have garnered increasing attention for their applications in environmental enhancement, bio-engineering, and biomedicine. This thesis embarks on an exploration of microbial networks, focusing on the evaluation of existing reconstruction methods, the integration of network inference with causal inference, and the use of reinforcement learning for network control.</p> <p>A critical component of this research is the comparison of existing microbial network reconstruction methods. These methods are evaluated based on various criteria, such as data availability and network sparsity, highlighting their assumptions, advantages, and limitations. This evaluation not only provides a guide for method selection but also discusses the challenges in microbial interaction studies and suggests future research directions.</p> <p>The thesis then introduces the Microbial Causal Inference (MCI) framework, designed to infer microbial species interactions, particularly in scenarios with feedback loops. MCI employs structural causal models to accurately represent microbial community dynamics under equilibrium conditions. Its efficacy is demonstrated in both acyclic and cyclic cases, including feedback loops, using algorithms such as Fast Causal Inference (FCI) and Greedy Fast Causal Inference (GFCI). The framework is validated through systematic analysis using simulated data, showcasing its applicability to different community parameters.</p> <p>Addressing the limitations of observational equilibrium data, the thesis proposes a set of rules and algorithms for designing experiments to obtain interventional data. These experiments are crucial for transforming Partial Ancestral Graph (PAG)-inferred structures into fully identified causal models. This approach not only enhances the understanding of microbial networks but also guides the experimental process in confirming or refining inferred causal relationships.</p> <p>Further, the thesis explores network control within microbial communities, introducing a novel reinforcement learning-based method for developing control algorithms. This venture into the application of reinforcement learning for network control opens new avenues for research and potential breakthroughs in this domain.</p> <p>The novelty of this research lies in the development and integration of the MCI framework with existing methods, providing a robust approach to infer microbial interactions under complex conditions such as feedback loops. This framework uniquely bridges network reconstruction and causal inference, validated with innovative algorithms (FCI and GFCI). This thesis extends this novelty by addressing the challenge of inferring complete causal structures from observational equilibrium data in the presence of feedback loops. It proposes adaptive learning methods and experimental design rules to transform PAG-inferred structures into fully identified causal models, thus enhancing our understanding of microbial dynamics and providing a systematic approach for future research in causal inference within complex biological systems. Additionally, the introduction of reinforcement learning for network control is a pioneering step, offering new strategies for manipulating microbial networks.</p> <p>The thesis concludes with a summary of the research findings, discussing the limitations and potential future directions. It emphasizes the need for ongoing development and refinement of methodologies to better understand and manipulate microbial community networks. The thesis posits that the integration of network reconstruction, causal inference, and reinforcement learning provides a comprehensive approach to studying and controlling microbial communities.</p> |
| spellingShingle | Computational biology Microbial ecology Experimental design Causal inference Fu, A Microbial communities: network reconstruction and control |
| title | Microbial communities: network reconstruction and control |
| title_full | Microbial communities: network reconstruction and control |
| title_fullStr | Microbial communities: network reconstruction and control |
| title_full_unstemmed | Microbial communities: network reconstruction and control |
| title_short | Microbial communities: network reconstruction and control |
| title_sort | microbial communities network reconstruction and control |
| topic | Computational biology Microbial ecology Experimental design Causal inference |
| work_keys_str_mv | AT fua microbialcommunitiesnetworkreconstructionandcontrol |