Study of Complex Biological Systems through AI-Empowered Metabolomic Analysis and Modeling

Engineering notes

Key topics: autonomous driving, prediction. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

This dissertation establishes an integrated framework that leverages artificial intelligence (AI) and advanced metabolomics to overcome the "complexity bottleneck" in microbial engineering, specifically addressing the global crises of plastic waste and carbon emissions. By framing the research within the complex biosystems for effective bioproduction and bioremediation, this work develops novel strategies to engineer and analyze biological systems for a circular bioeconomy. In the "Build" phase (Chapter 2), the study addresses the instability of synthetic microbial consortia. By utilizing calcium-alginate hydrogel compartmentalization, the research successfully decouples the growth of the phototroph Synechococcus elongatus from its heterotrophic partners (P. putida and Y. lipolytica). This spatial separation protects heterotrophs from oxidative stress (ROS) and enables the stable production of high-value pigments, achieving titers up to 22 times higher than traditional free-cell co-cultures. The hydrogel compartmentalization also effectively simplifies the product recovery and extraction process. Meanwhile, spatially separated co-culturing enables metabolomic analysis for individual species for their biochemical response to different culture conditions. The "Design" and "Learn" phases (Chapter 3) introduce the MAP (Multimodal AI for Photobiorefinery) framework to solve the "cold start" problem for the fast-growing cyanobacterium S. elongatus UTEX 2973. This multimodal approach integrates GPT-4 assisted knowledge mining and Transfer Learning (TL) from well-studied model organisms. The TL models achieved high predictive accuracy (R² > 0.93) for growth and product titer, even under stringent testing conditions. The framework further links these predictions to Techno-Economic Analysis (TEA) to evaluate commercial feasibility, identifying optimal conditions for products like lycopene. Finally, the "Test" and "Analyze" phases (Chapter 4) apply an AI-empowered metabolomics toolkit to deconvolve the complex natural microbiome of the yellow mealworm (Tenebrio molitor) during plastic degradation. To deconvolve the complex metabolic "black box" of the mealworm gut, we developed a high-resolution, in vivo spatial metabolomics workflow. This methodology employed a spatial compartmentalized protocol, either isolating the foregut, midgut, and hindgut, or separately collecting gut content, worm body, and biowaste, combined with flash-frozen quenching to preserve transient metabolic intermediates. The resulting complex datasets were processed using the MSOne AI-software suite, which utilizes Convolutional Neural Network (CNN) for peak classification and a U-Net-based CNN for peak segmentation for automated spectral denoising and peak isolation. By integrating MSOne with SIRIUS 4 for de novo molecular formula annotation, we achieved a ten-fold improvement in compound identification compared to traditional manual curation, identifying over 200 metabolites and recovering more than 4,000 biological signals. Our analysis revealed a striking spatial "division of labor" along the larval gut: amino acid degradation is localized in the foregut; central energy pathways, the TCA cycle, and the glyoxylate shunt are activated in the midgut; and nutrient salvage predominates in the hindgut. This work identifies specific biomarkers of plastic degradation, such as oleic acid, oxidized lipids, and long-chain alkanes, and uncovers a shared microbial response to oxidative stress involving arginine biosynthesis. Validated through ¹³C-carbon tracing and internal standards, this research provides the first comprehensive molecular blueprint of the systemic and spatial response to plastic ingestion, establishing a foundational toolkit for engineering future synthetic plastic-valorization systems. Collectively, this research demonstrates that the seamless integration of predictive AI and high- throughput analytical platforms fundamentally accelerate the bio-engineering lifecycle, providing a scalable blueprint for the industrial translation of sustainable biotechnologies.

Links and sources

• PDF from original source

Comments