This project has established CitrusShield——an intelligent biological system driven by synthetic biology and empowered by digital technology, aimed at revolutionizing the pest and disease management paradigm in citrus cultivation. We designed three core, synergistic "BioBrick" modules: an early warning biosensor based on Bacillus subtilis, a preventative therapy using multigenerational iterative RNA interference (RNAi) delivered via MS2 virus-like particles, and an engineered Metarhizium anisopliae as a targeted therapeutic agent using RNAi with a safety switch. These modules are seamlessly integrated through a digital platform that incorporates environmental sensing, AI analysis, and user interaction, forming a complete closed-loop of "biological sensing-digital decision-making-biological response." CitrusShield represents a fundamental shift from passive response to active prevention and precision treatment, transforming complex synthetic biology technologies into accessible, intelligent, and sustainable agricultural solutions for farmers.

The citrus industry, as a significant pillar of global agricultural economy, is currently facing severe challenges in pest and disease management. Pests such as the brown citrus aphid (Toxoptera citricida), characterized by rapid reproduction and strong concealment, pose difficulties for traditional control methods. To address this situation, we have developed an intelligent citrus protection system—CitrusShield, which aims to provide proactive and scientific integrated pest management solutions through digital technology.

This project is based on the core concept of "digital agriculture"^[1], integrating traditional pest control methods with modern information technology. Hardware sensors are responsible for real-time collection of environmental data, while AI deep learning technology enables intelligent disease recognition. Combined with an intuitive and user-friendly mobile interface, we have constructed a complete protection system integrating "monitoring—identifying—intervening". By establishing an early warning mechanism and providing scientific control guidance, CitrusShield significantly enhances the intelligence level of citrus cultivation, offering reliable decision support for farmers.

In the context of traditional citrus pest management relying on manual observation and broad-spectrum pesticides, CitrusShield constructs an intelligent ecological system that achieves a "perception-judgment-response" closed loop. This system facilitates a paradigm shift from passive responses to active interventions through the synergistic integration of three core modules: An engineered Bacillus subtilis serves as a "biosensor", converting sucrose in aphid honeydew into remotely detectable methyl salicylate signals^[2]. A multi-generational iterative RNAi delivery system forms a "persistent defense system" targeting latent pests. A "rapid intervention system" built on engineered Metarhizium anisopliae achieves dual strikes of infection and gene silencing.

These three modules seamlessly interact through a digital platform, together building a comprehensive biological defense network from early warning to precise intervention. This allows farmers to choose preventive or therapeutic pest control measures precisely based on the intelligent system's assessments of "latent" or "outbreak" phase conditions, realizing the technological advancement of orchard management.

Fig 1. CitrusShield Smart Ecosystem

Within the complete architecture of CitrusShield, digital platform plays the crucial role of a "central nervous system". This intelligent hub, based on WeChat Mini Program and cloud services, not only parses and visualizes in real-time the methyl salicylate (MeSA) biological signals released by the engineered bacteria—providing farmers with an intuitive overview of pest situations—but, more importantly, translates complex biological data into "clear control decisions". It intelligently recommends precise intervention strategies such as "preventive RNAi injections" or "initiating Metarhizium spray". Furthermore, by continuously collecting and analyzing field data, the platform constantly feeds back into and optimizes our AI algorithms and synthetic biology designs, creating a virtuous cycle of evolution where "technology drives application, and application feedback refines technology." This ensures the entire system grows smarter and more precise with each use in practical scenarios.

Fig 2. Data Flow Design of the CitrusShield Smart Ecosystem

To translate the intelligent protection capabilities of CitrusShield into a "user-friendly experience", we developed a fully functional mobile application based on the WeChat Mini Program native framework. A stable and reliable data transmission channel was constructed using the wx.request API and a Promise-based network request scheme, while the integration of ECharts data visualization components enabled the intuitive presentation of environmental monitoring data.

In the functional architecture, we designed four core modules: "The Environmental Monitoring Module" processes and displays key parameters such as temperature, humidity, CO₂, and TVOC in real-time through the processEnvironmentData^[3] method, forming a comprehensive orchard environment profile. "The AI Disease Identification Module" utilizes the camera API and a deep learning model to provide users with a real-time disease diagnosis service via photo capture. The "Knowledge Base Module" systematically integrates prevention and treatment plans and expert experience for 17 common diseases, creating a plant protection guide accessible for consultation at any time. "The Community Exchange Module" establishes a platform for interaction between farmers and experts, as well as among farmers themselves, facilitating the sharing and dissemination of cultivation experience.

Within the CitrusShield system, "The Environmental Monitoring Module" serves as a critical bridge connecting biosensing with precise intervention, providing essential data support for the three core biological modules. This module constructs a comprehensive orchard microenvironment profile through real-time collection and processing of key environmental parameters such as temperature, humidity, CO2, and TVOC. Among these, abnormal fluctuations in CO₂ concentration can serve as an indirect indicator for assessing field anomalies, while continuous TVOC monitoring assists in verifying the methyl salicylate signal released by the engineered Bacillus subtilis. The methyl salicylate signal directly reflects pest presence in the field. These standardized environmental data, cross-validated with the biosensing signals, provide a reliable basis for the system to accurately determine the "Latent Phase" or "Outbreak Phase." This guides farmers to initiate RNAi preventive treatment or targeted eradication using engineered Metarhizium anisopliae at the most opportune time, achieving a precise connection from environmental perception to biological control.

This architecture not only ensures the stable operation of the system but, more importantly, transforms complex biosensing technologies and AI diagnostic capabilities into a simple and operable interface for farmers, allowing cutting-edge agricultural technology to be genuinely integrated into daily cultivation management.

Fig 3. Functional Interfaces of CitrusShield. Environmental signal detection (left), community exchange and pest/disease reporting (middle left and right), and citrus pest and disease knowledge base (right).

During field research, we observed that farmers often appear confused and helpless when confronted with various lesions on leaves—what is the difference between anthracnose and canker? What chemical should be used for scab? These seemingly technical questions directly impact the annual harvest. It was these real-world challenges encountered in the field that motivated us to transform cutting-edge artificial intelligence technology into an accessible production tool. In our CitrusShield system, we have initially developed and deployed an intelligent citrus disease identification function based on deep learning^[4]. By constructing a dataset comprising 15,420 high-quality images covering 17 common disease categories such as black spot and canker, and employing the YOLOv11 algorithm to train an efficient recognition model, this function has been preliminarily integrated into the WeChat Mini Program. Users can quickly obtain disease diagnosis results by taking photos. To ensure stable service capability, we built a cloud architecture based on Tencent Cloud, utilizing TensorFlow Serving to provide inference services. Coupled with Docker containerized deployment to ensure high system availability, and adopting a cloud-edge collaborative architecture where knowledge distillation techniques optimize model performance on the edge side, we aim to provide farmers with accurate disease identification services.

Fig 4. Partial Dataset Training Charts

The value of this function lies not only in its technological innovation, but also in enabling every fruit farmer to carry a "portable plant doctor" ——take out a phone, snap a photo, and obtain accurate disease diagnosis and control recommendations within minutes. This shifts cultivation management, which once relied solely on experience, into a new data-driven stage.

During our field research, we found that many citrus growers face challenges such as information isolation and a lack of effective communication channels – often, they only become aware of pest problems after they have reached a significant scale, and even after discovering issues, they struggle to obtain timely professional guidance and support. This current situation of "difficult detection, difficult feedback, difficult resolution" severely constrains the efficiency of pest and disease control. To address this pain point, CitrusShield has meticulously built an open and collaborative user interaction ecosystem^[5]. By establishing convenient pest reporting channels, farmers can submit field conditions at any time and receive real-time system feedback and professional advice. Simultaneously, the established community exchange platform breaks down information silos, allowing cultivation experiences and control insights to flow freely. Here, every farmer is both a knowledge receiver and an experience sharer, forming a virtuous cycle from individual alerts to collective intelligence. This ecosystem not only enables early detection and rapid response to pest problems but also, by building a network connecting farmers with experts and farmers with each other, transforms traditional one-way control into a multi-party collaborative win-win model. It truly makes knowledge flow and experience sharing possible, injecting new vitality and wisdom into the citrus industry.

Although the CitrusShield system has established a preliminary framework integrating monitoring, identification, and intervention, there are still shortcomings in practical application. The current AI disease identification model, while covering 17 common diseases, has a training dataset that remains limited relative to the complex and variable field environment. The model's identification accuracy and robustness face challenges when dealing with citrus performances across different regions, seasons, and growth stages. Furthermore, due to the lack of large-scale, long-term field data correlating actual pest occurrences with environmental factors, the system's decision support currently cannot build accurate pest prediction models. The current "intelligent decision-making" primarily involves data presentation and basic alerts, and has not yet achieved early warnings based on multi-dimensional data fusion or personalized control recommendations. There is still a gap compared to a truly "intelligent agriculture" decision system. To address these deficiencies, we will leverage the hardware systems deployed in the field to continuously collect multi-dimensional environmental signals —— including Methyl Salicylate (MeSA), CO₂, temperature, and humidity —— and their relationship with aphid populations, to further optimize the decision-making model. Concurrently, by introducing a continuous learning mechanism, the AI model will utilize this continuous stream of field data for incremental training and fine-tuning. This will enable it not only to identify diseases but also to provide early warnings of aphid density changes reported by the engineered bacteria, thereby achieving a leap from "identifying diseases that have already occurred" to "predicting pests that will occur." This will significantly enhance the model's generalization capability and foresight.

In summary, our vision is for CitrusShield to transcend being a simple tool and become a citrus grove guardian ecosystem that integrates digital intelligence and synthetic life, capable of autonomous sensing, intelligent decision-making, and precise execution, ultimately establishing a new paradigm for achieving green and sustainable modern agriculture.