??AI for Potato: Late Blight Detection ??

Essential Information on Late Blight Impacting Potato Crops

• Caused by the oomycete Phytophthora infestans.
• Favours cool, wet conditions with high humidity (>90%) and moderate temperatures (10-25°C).
• Develops rapidly when leaf surfaces are wet for prolonged periods.
• Can spread long distances through wind-dispersed spores, making it highly contagious.
• Can cause complete crop loss within days if not managed promptly.
• Responsible for historical potato famines, making it a major threat to potato production worldwide
• Late Blight is more dangerous than Early Blight due to its rapid spread, potential for complete crop destruction, and complexity of management.

Evolution of Phytophthora infestans on its Potato Host Since the Irish Potato Famine

Research Institution:

?? North Carolina State University, USA ????

Published: 5 August 2024

The study aims to explore the evolutionary changes in Phytophthora infestans and its potato host since the Irish Potato Famine, using targeted genomic sequencing to analyze pathogen and host gene evolution.

The main question the researchers wanted to answer is: How have the pathogen’s virulence genes and the potato host’s resistance genes co-evolved since the Irish Potato Famine, and what implications do these changes have for current disease resistance strategies?        

The researchers used targeted enrichment sequencing on 29 historic herbarium specimens (collected between 1845-1954) containing both Phytophthora infestans and potato host DNA.

This method involved extracting DNA and using "bait libraries" designed to capture specific resistance (R) genes and effector genes, allowing for a detailed genetic analysis of evolutionary changes.

The study focused on the RXLR effector genes, known for their role in the pathogen's virulence, and various potato R genes, such as R1 and R3a. This approach enabled the researchers to track genetic shifts in both the host and pathogen over nearly a century. One niche insight was using historical herbarium samples, which provided a unique view of pathogen-host evolution over time, compared to modern genomes.

AI played a crucial role in this research by utilizing machine learning algorithms for variant calling and genetic alignment.

This enabled accurate identification of single nucleotide polymorphisms (SNPs), insertions, deletions, and other mutations within both the Phytophthora and Solanum genomes. AI algorithms were also employed to automate the detection of recombination events and to build phylogenetic trees, providing a high-resolution view of gene flow and evolutionary patterns. Furthermore, AI-assisted analysis helped manage the degraded nature of historic DNA, overcoming challenges associated with fragmented sequences.

In the figure above: A Amino acid alignment of a 100 amino acid excerpt of Avr1 from P. infestans. The first line is the Avr1 avirulent allele which is recognized by R1. The second line is the AL virulent, resistance-breaking allele which escapes detection by R1. The subsequent amino acid sequences are from high coverage samples analyzed in this study. All historic samples show a premature stop codon (indicated by x) leading to the virulent allele and are thus able to overcome R1. B Alignment of a 100 amino acid region excerpted from late blight resistance gene R1 cloned from Solanum demissum and R1 homologs in our samples and the reference genome. The functional S. demissum allele is shown on the top line, followed chronologically by the samples analyzed in this study. The final line is the R1 homolog in the reference SolTub3.0 genome. Several samples have premature stop codons (indicated by x) signifying a truncation of the R1 protein as compared to S. demissum.

Key findings

1. There has been a significant increase in RXLR effector gene diversity in Phytophthora infestans since the Irish Potato Famine.
2. The researchers identified 521,699 unique SNPs and observed a 13.7% increase in RXLR genes in post-Famine pathogen samples compared to Famine-era samples.
3. Furthermore, triploid lineages of P. infestans, which emerged around the 1940s, demonstrated higher resistance-breaking capabilities against traditional R genes, such as R1 and R2.
4. These findings highlight the pathogen’s rapid adaptation, driven by both natural selection and breeding efforts in the host plants, complicating the development of durable resistance.

Agricultural researchers, breeders, and crop protection specialists can apply these results to improve breeding programs and develop new strategies for durable disease resistance.

Technologies used

• Enrichment sequencing with bait libraries
• DNA extraction kits (QIAGEN DNeasy Plant Mini Kit)
• Sequencing platforms (NovaSeq with 150bp paired-end reads)
• Variant calling and alignment software (GATK, BWA-MEM)
• Machine learning algorithms for phylogenetic and evolutionary analysis
• Python scripts for data visualization and coverage analysis
• SPAdes de novo assembler for unmapped read characterization
• BLAST for gene alignment

Revolutionizing Potato Late Blight Surveillance: UAV-Driven Object Detection Innovations

Research Institution:

?? Mohammed First University, Morocco ????

Published: 15 April 2024

The study focuses on leveraging UAV (Unmanned Aerial Vehicle) technology and advanced deep learning models, such as YOLO and Faster-RCNN, for real-time detection and monitoring of Potato Late Blight, aiming to provide an efficient, scalable solution for large-scale agricultural surveillance.

The primary question the researchers aim to answer is: Which deep learning object detection model—YOLO or Faster-RCNN—is more effective for UAV-based monitoring of Potato Late Blight in terms of precision, processing speed, and real-time applicability?        

1. To address this, the research team utilized a comprehensive methodology involving high-resolution image data collection using Mavic Air drones fitted with advanced cameras.
2. The study compared two distinct deep learning approaches: single-stage models (YOLOv6, YOLOv7, and YOLOv8) and double-stage models (Faster-RCNN with ResNet50, VGG16, and VGG19 backbones).
3. The gathered images were pre-processed through intensity normalization, data augmentation, and distortion correction to create a robust dataset for training.
4. During model training, various configurations were tested to evaluate the performance across key metrics such as mean Average Precision (mAP), F1 score, and inference time. One unique insight was the integration of field-acquired drone images with diverse internet-sourced data to build a comprehensive dataset, enhancing model robustness.

The key findings revealed that the Faster-RCNN model with ResNet50 backbones achieved the highest precision (93.92%), F1 score (93.97%), and [email protected] of 95.32%, making it ideal for scenarios requiring high accuracy. However, YOLOv8, with a lower [email protected] of 91.45%, demonstrated superior speed (1.43 ms per image), making it the best option for real-time applications.

This dual comparison highlights that for large-scale surveillance, a balance between accuracy and processing speed is crucial, with YOLO models being more suitable for quick field assessments and Faster-RCNN for detailed, high-precision analyses.

Technologies used

- Hardware: Mavic Air Drone, Dell PowerEdge R740 server, Intel Xeon Silver 4210 processor, NVIDIA RTX A5000 GPUs.

- Software: YOLOv6, YOLOv7, YOLOv8, Faster-RCNN (with ResNet50, VGG16, VGG19 backbones), TensorFlow, Python.

- Tools: Open-Source Data Labeling Software, SPAdes de novo assembler, BLAST.

References

What's on

If you'd like to receive the regular 'AI in Agriculture' newsletter in your inbox, simply add your email to my mailing list.

Join over 8,742 readers who enjoy weekly updates on AI advancements in agriculture!

To get more information about onion phenotyping capabilities with mobile - ask Petiole Pro