Bt Research 2024, Vol.15, No.5, 240-247 http://microbescipublisher.com/index.php/bt 243 4.2 Early warning systems for resistance management AI can also be instrumental in developing early warning systems for resistance management. By analyzing global resistance monitoring data, researchers have identified early warnings of resistance in various pest populations, which involve genetically based decreases in susceptibility without evidence of reduced field efficacy (Tabashnik et al., 2023). AI algorithms can process large datasets to detect subtle changes in susceptibility, providing timely alerts to implement mitigation strategies. For example, computer vision-based algorithms have been developed to measure maize injury caused by resistant pests more accurately and precisely than human observers, thereby improving the reliability of resistance monitoring (Mendoza-Almanza et al., 2020). 4.3 AI-based models to track Bt resistance in field populations 4.3.1 Data collection and model construction The construction of AI-based models to track Bt resistance in field populations begins with comprehensive data collection. This includes monitoring field-evolved resistance through sentinel plots and collecting larvae from non-Bt host plants for laboratory bioassays (Dively et al., 2020). Spatially explicit models have been developed to assess the impact of spatial heterogeneity on the evolution of resistance, highlighting the importance of considering both spatial and non-spatial conditions in model construction (Huang et al., 2017). 4.3.2 Model validation and result analysis Model validation is a critical step to ensure the accuracy and reliability of AI-based predictions. For instance, genetic markers associated with resistance were validated using WCR populations collected from Cry3Bb1 maize fields, demonstrating the practical applicability of these markers in real-world scenarios. Similarly, the accuracy of computer vision-based algorithms for detecting maize injury was validated against human observer estimates, showing improved precision and reduced false positives (Dorman et al., 2021; Chen et al., 2022). These validation efforts are essential for refining AI models and ensuring their effectiveness in resistance management. 4.3.3 Practical applications and resistance management strategies The practical applications of AI-based models in resistance management are vast. AI can help in designing and implementing effective insect resistance management (IRM) strategies by predicting where future field failures may occur and by characterizing the extent of existing resistance issues. Moreover, AI can assist in developing integrated pest management strategies that incorporate refuges and other tactics to delay resistance evolution (Gassmann and Reisig, 2022; Chen, 2024). By providing accurate and timely predictions, AI-based models can significantly enhance the sustainability of Bt crops and reduce the reliance on conventional insecticides. 5 Case Studies of AI Applications in Bt Research 5.1 Machine learning in predicting insect resistance to Bt Machine learning has been instrumental in predicting insect resistance to Bt toxins. For instance, a computer vision-based algorithm was developed to measure Helicoverpa zea injury in Bt maize fields. This algorithm provided more accurate and precise estimates compared to human observers, reducing the false positive rate and unnecessary insecticide applications. This technological solution standardizes Bt injury metrics, preserves digital data for cross-referencing, and significantly increases sample sizes, thereby improving resistance monitoring efforts (Dorman et al., 2021). Spatially explicit models have been used to predict the evolution of insect resistance in heterogeneous environments. These models highlight the importance of spatial heterogeneity in resistance management strategies, showing that regional resistance can evolve faster in heterogeneous environments due to gene spread from resistance hotspots (Coates and Siegfried, 2015; Shi et al., 2016). 5.2 AI-driven identification of novel Bt toxins AI tools have also been developed to identify novel Bt toxins. CryProcessor, for example, is a robust pipeline designed to mine 3d-Cry toxins from sequencing data. It overcomes the limitations of previous tools by allowing precise mining of Cry toxin sequences directly from assembly graphs, thus enabling the analysis of raw sequencing data and retrieval of specific domain sequences. This tool has shown efficiency in large-scale data processing and can significantly aid in the discovery of new Bt toxins for pest control (Figure 2) (Shikov et al.,
RkJQdWJsaXNoZXIy MjQ4ODYzNA==