Source: ARKANSAS STATE UNIVERSITY submitted to NRP
DSFAS: MACHINE LEARNING INTEGRATION OF MULTITEMPORAL IMAGERY AND GENOMICS TO ACCELERATE DEVELOPMENT OF CLIMATE-SMART RICE
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
ACTIVE
Funding Source
Reporting Frequency
Annual
Accession No.
1031244
Grant No.
2023-67022-40859
Cumulative Award Amt.
$297,976.00
Proposal No.
2022-11562
Multistate No.
(N/A)
Project Start Date
Jul 1, 2023
Project End Date
Jun 30, 2026
Grant Year
2023
Program Code
[A1541]- Food and Agriculture Cyberinformatics and Tools
Recipient Organization
ARKANSAS STATE UNIVERSITY
(N/A)
STATE UNIVERSITY,AR 72467
Performing Department
Chemistry & Physics
Non Technical Summary
Finding novel genetic targets to improve crop yields and maintain quality is increasingly important as plant breeders race to keep up increasingly unpredictable climates, rising costs of agricultural inputs, and the need to conserve natural resources. This project focuses on two themes towards accelerating plant breeding efforts: 1) development of new approaches for characterizing plant traits from high resolution drone imagery and 2) leveraging machine learning and artificial intelligence to identify genes important for improved crop performance under low input environments. Our proposed research will determine whether genome associations with multiple images, instead of just a single image, could help reveal novel targets for genetic improvement of rice, a critically important staple food. Rice production can have outsized environmental impacts due to water-intensive cultivation systems and greenhouse gas emissions, and so development of varieties that perform well under low-input conditions is a top priority. Additionally, we will determine whether genetic variants associated with improved nitrogen and water use efficiency could be determined from existing ecological and genomic datasets. Our model public-private partnership contributes to the National Artificial Intelligence Initiative through its focus on increasing crop yields while conserving natural resources, and contributes to the design and validation of novel methods for analysis and integration of big agricultural data, one of the top priorities for the Data Science for Food and Agricultural Systems Program Area.Development of rice varieties with improved performance under low water input management conditions will help to conserve freshwater resources in the face of increasing water scarcity. Additionally, reducing fertilizer inputs through development of nitrogen-efficient rice varieties and precision management tools will reduce nutrient pollution and decrease greenhouse gas emissions in water-limited rice production systems. Reduced input costs for irrigation and fertilizer will translate to improved economic viability for rice production, as costs for nitrogen and water inputs continue to rise. Beyond rice, our approach to identifying genes associated with reduced inputs in modern agricultural systems could be broadly applicable to other crops and systems.
Animal Health Component
20%
Research Effort Categories
Basic
80%
Applied
20%
Developmental
(N/A)
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
20215302080100%
Knowledge Area
202 - Plant Genetic Resources;

Subject Of Investigation
1530 - Rice;

Field Of Science
2080 - Mathematics and computer sciences;
Goals / Objectives
The major goal of the proposed research is to accelerate development and deployment of improved crop varieties. To enable more rapid and accurate identification of genetic variants underlying agronomic traits of interest, we will develop improved computational frameworks for 1) characterizing high-dimensional plant phenotypes and 2) leveraging existing germplasm, genomic, and environmental data resources for crop landraces. As an example system, this project focuses specifically on nitrogen- and water-use efficiency in rice, a globally important crop that has outsized environmental impacts due to water-intensive cultivation systems and greenhouse gas emissions.First, we will develop a new analytical framework for characterizing dynamic phenotypes from multispectral, multitemporal imagery (Aim 1). Genome-wide association studies (GWAS) in breeding programs may consider individual trait measurements from images at specific timepoints (e.g. leaf area, plant height), excluding potentially useful information reflecting connections across several images over time. The objectives of Aim 1 include:Testing the use of manifold learning techniques for capturing phenotype trajectories, using drone images from two years of field experiments with rice,Comparing models trained on derived features vs. models trained directly on image embeddings,Evaluating phenotype trajectory offsets for nitrogen responsive vs. unresponsive rice cultivars as a way to capture variation in environmental tolerance, andComparing accuracy of trajectory-based models vs. models based on individual timepoints for predicting agronomic characteristics including yield, days to 50% heading, etc.Second, we will evaluate use of improved genotype-environment association (GEA) methods to identify genomic variants associated with rice landrace adaptation to water-limited environments, and determine whether these are useful targets for modern plant breeding efforts. Objectives for Aim 2 include:Developing a novel GEA method for identifying genetic variants associated with rice landrace adaptation to upland and rainfed lowland irrigation systems, both expected to impose strong selection for increased water use efficiency,Training a suite of genomic prediction models using data from published GWAS. We will train 'baseline' models representing the current state-of-the-art for genomic prediction and compare them with models trained using subsets of markers identified in (1),Identifying a subset of rice accessions from the USDA Mini-Core Collection with the largest differences in predicted phenotypes according to models in (2),Experimental testing of rice accessions identified in (3) to determine whether environment-associated genetic variants improve trait prediction under low input conditions.
Project Methods
General scientific methods for the project include data analysis/modeling techniques standard in the data science and machine learning communities as well as conducting greenhouse experiments with rice and collecting high-throughput plant phenotype data.Efforts to be used include scientific research conducted by the project team, hands-on experiential learning opportunities for the students and postdoctoral scientist, presentations to the scientific community and the public through formal and informal educational programs, and presentations and networking with breeders and growers at local and national conferences and events.For project evaluation, key milestones and indicators of success include 1) career development and success of trainees involved in the project, 2) submission/success of subsequent grant proposals based on the work supported by this Seed Grant by the project leaders, and 3) publication of two peer-reviewed manuscripts describing the outcomes of the supported research.

Progress 07/01/23 to 06/30/24

Outputs
Target Audience:Our target audience are the plant scientific community in general, rice breeders, rice farmers, funding agencies and other relevant stakeholders . We seek to disseminate our findings via peer review publications, oral presentations at relevant venues. Changes/Problems:There have been two main challenges in the Arkansas State University component of the project: 1) The approval of the work visa for the post-doc selected to work on this project took significantly longer than expected. 2) PD Lorence had a health issue in the Fall of 2024. Steps were taken to mininize the impact of this unexpected issue in the progress of the research. What opportunities for training and professional development has the project provided?We have trained the following students and professionals: At Arkansas State University (A-State) Dr. Reinier Gesto-Borroto, Post-doctoral Research Associate, Jan 24 to date Bishnu Prasad Joshi, PhD student, August 24 - date Carlos Efrain Cruz Bahena, undergrad, Jan 24 - August 24 Alexander Flores, undergrad, Jan 24 to date Tatyana Herrien, undergrad, May 24 to date Merone Kebede, undergrad, August 24 to date Ali Abdel-Karim, undergrad, Jan 25 to date Ravi Chaudhary, undergrad, Jan 25 to date At Avalo Fared Farag, MSc student; Internship at Avalo headquarters June - August, 2024. He was hired in a full-time position at Avalo in Fall 2024. How have the results been disseminated to communities of interest?Results from Aim 1 were published in a peer-reviewed manuscript. We have also presented our progress at 7 invited talks and seminars and 3 poster presentations at local, regional, national and intrenational meetings: Evolution 2023 (national), US Davis Plant Biology Seminar (national), NC State International Plant Breeding Seminar (international), NC INTRINSyC Seminar (international), International Meeting Natural Products Research (international) ,International Symposium in Ruce Functional Genomics (international), Plant and Animal Genome Conferene (international), ABI Fall Symposium (regional) and the International Plant Phenotyping Symposium (international). What do you plan to do during the next reporting period to accomplish the goals?Complete the ongoing large phenotyping experiment where we are screening candidate rice accessions looking for water limitation tolerance (data acquisition). Analyze all data Write a manuscript with the main findings Write and submit a full proposal to test candidate genes for drought tolerance in rice.

Impacts
What was accomplished under these goals? Aim 1 was fuly completed.. That resulted in a peer-review publication:Farag F, Huggins T, Edwards J, McClung A, Hashem A, Causey J, and E Bellis. Manifold and spatiotemporal learning on multispectral unoccupied aerial system imagery for phenotype prediction. The Plant Phenome Journal, 2024;7:e70006. https://doi.org/10.1002/ppj2.70006. The progress on Aim 2 is as follows: Drs. Bellis and Alvarez (Avalo) developed genotype-environment association models to identify candidate genetic variants associated with adaptation to rainfed lowland environment in rice landraces. 8 potential targets for follow-up analysis were identified. The Arkansas State team is working on validating 1 of these targets via gene editing approaches. After validation this information may be used to develop more drought tolerance rice varieties. Drs. Bellis and Alvarez (Avalo) developed genomic prediction models and identified 20 accessions for further characterization of phenomic response to water limitation. These are being further studied in experiments at Arkansas State University using high throughput multitemporal phenotyping. PD Lorence requested seeds of 40 accessions identified by the Avalo team from the USDA Dale Bumpers National Rice Research Center. These seeds are the drought tolerant candidates we are studying in detail. The Lorence team acquired images from 50 seeds from each accession and did image analysis to extract key phenotypic characteristics including seed length, width, area, perimeter, and eccentricity. This data will be later compared to the one obtained from seeds of plants exposed to water limitation conditions. The Arkansas State team led by PD Lorence installed a new high throughput phenotyping system at the Arkansas Biosciences Institute greenhouse. The system uses Raspberry-pi computers and RGB cameras. The system was tested after growing rice plants var. Kitaake under normal and water limiting (drought) conditions. The acquired RGB images were analyzed using algorithms developed by Dr. Suxing Li (University of Arizona, collaborator) The Arkansas State team led by PD Lorence is conducting a large experiment involving 360 plants from the rice accessions identified by Drs. Bellis and Alvarez. Seeds were clean and planted in plant culture media. Healthy seedlings are currently growing in environmental controlled chambers. After adaptation to soil, the seedlings will be divided into 2 groups. Half of the plants will grow under normal watering conditions (85% soil capacity), while the other half will receive a drought treatment (50% soil capacity). The drought treatment will be stopped when plants reached the flowering stage. The effects of drought will be assessed from RGB images acquired weekly and from non-destructive photosynthesis efficiency measurements. At the end of the plant life cycle, seed yield will be quantified looking for those drought tolerant accessions. This experiment is testing the accuracy of the models developed by Dr. Bellis. This data may be used use to help developed rice varieties capable of maintaining yields under water limitation conditions.

Publications

  • Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: Bellis E, Abernathy E, Lorence A, Alvarez M. Accelerating development of water- and nutrient-efficient rice with evolutionary genomics and machine learning. Evolution 2023, Albuquerque, NM, June 21-25, 2023.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: Bellis E. Understanding crop landrace evolution to accelerate breeding for current and future environments. UC Davis Plant Biology Seminar, Davis, CA, October 27, 2023.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2024 Citation: Gesto-Borroto R, Abernathy E, Farag F, Cruz-Bahena CE, Alvarez M, Bellis E, Lorence A. Machine learning integration of multitemporal imagery and genomics to accelerate development of climate-smart rice. 8th International Plant Phenotyping Symposium, Lincoln, NE, October 7-11, 2024.
  • Type: Other Journal Articles Status: Published Year Published: 2024 Citation: Farag F, Huggins T, Edwards J, McClung A, Hashem A, Causey J, Bellis E. Manifold and spatiotemporal learning on multispectral unoccupied aerial system imagery for phenotype prediction. The Plant Phenome Journal, 2024;7:e70006. https://doi.org/10.1002/ppj2.70006
  • Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: Bellis E. Understanding crop landrace evolution to accelerate breeding for current and future environments. North Carolina State University International Plant Breeding Seminar, Raleigh, NC, October 12, 2023.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2024 Citation: Bellis E. Accelerating crop adaptation to climate with interpretable machine learning and global ecogeographic models. NC State INTRINSyC Seminar, Raleigh, NC, January 19, 2024.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2024 Citation: Lorence A. Engineering climate resilient crops with elevated ascorbate content. 19a Reuni�n Internacional de Investigaci�n en Productos Naturales, Asociaci�n Mexicana de Investigaci�n en Productos Naturales, Cuernavaca, Mexico, May 21-24, 2024.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2024 Citation: Lorence A. Rice field phenotyping. International Symposium on Rice Functional Genomics: Celebrating 20 Years of the Rice Genome, Little Rock, AR, September 9-11, 2024.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2025 Citation: Bellis ES, Alvarez MF. Genomic selection for latent environments to develop climate-adapted crops. Plant and Animal Genome Conference. San Diego, CA, January 13, 2025.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2024 Citation: Gesto-Borroto R, Abernathy E, Farag F, Cruz-Bahena CE, Alvarez M, Bellis E, Lorence A. DFAS: Machine learning integration of multitemporal imagery and genomics to accelerate development of climate-smart rice. Agricultural and Food Research Initiative 2024 Project Directors Meeting, Manhattan, KS, July 25-26, 2024.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2024 Citation: Gesto-Borroto R, Lorence A. Digital Phenotyping at the A-State Plant Phenomics Facility. ABI Fall Research Symposium, Jonesboro, AR, September 19, 2024.