Previously Funded Seed Grants

Round 1 Funding 7

Application Deadline: March 19, 2021

Empowering High-Throughput Phenotyping using Unoccupied Aerial Vehicles (UAVs)

Max Feldman, Filipe Matias, Jennifer Lachowiec, David LeBauer
Download Narrative

The AG2P Initiative strives to connect genotype to phenotype in many environments. A growing source of phenome data in agricultural research across animals and plants are Unoccupied Aerial Vehicles (UAV). In this proposal we build a foundation to empower more researchers to use UAS. In the objectives, we will complete an international survey of agricultural animal science and plant science researchers of UAV use and build a community-informed web resource to provide instructional information and benchmarking tools to support the growing number of UAV users.

Expected Outcomes and Deliverables: In summary, this project will provide the foundation to accelerate high-throughput phenotyping in agriculture to support the mission of AG2PI. A survey will assess how UAV imagery has been applied to plant and animal sciences in agriculture and identify obstacles common among research groups that will pinpoint potential solutions. Simultaneously, developing standardized best practices will support UAV adoption and reduce barriers to entry. These efforts will be shared through workshops, videos, conferences, and manuscripts.

Ethics, Diversity and Inclusivity in G2P Research

Cassandra Dorius, Shawn Dorius, Kelsey Van Selous, Rachael Voas
Download Narrative

Sustainably supporting worldwide food production is a ‘wicked problem’ of immense scale and complexity. As such, there is an urgent need for novel ideas and technological innovations in the agricultural research. One strategy for rapidly infusing new ideas into existing knowledge networks is by making these person-centric networks more diverse, data more accessible to relevant stakeholders, and by infusing current practices with the kinds of Ethical, Legal, Social, Ecological, and Economic (ELSEE) considerations that will transform agricultural genome and phenome research practices. Bringing underrepresented groups to the table is one way to infuse AG2PI with new ideas. Improving and expediting knowledge transfer (e.g. data sharing), more effectively communicating research findings to the general public, policy makers, and funding agencies, and developing new science practices, can also help AG2PI to achieve sustainable genetic improvements. We propose to advance the aims of the AG2PI by conducting social science research to encourage cross-fertilization of AG2P data and ideas and motivate agriculture focused analysis from an ELSEE perspective. We will also create human-centered

Seeding Public-Private Partnerships for AG2P Training

Addie Thompson, Tammy Long, Jyothi Kumar
Download Narrative

The main goal of this project is to form public-private partnerships to expand meaningful interactions between industry and the public sector. Outcomes will include:

  1. Enhancing graduate student training through the use of current real-life AG2P project scenarios and datasets;
  2. Generating public educational resources including datasets and code for use by other scientific communities for AG2P training;
  3. Serving as a model and test bed for PPP engagement and seeding meaningful ongoing collaborations in interdisciplinary groups.

Optimizing 3D Canopy Architecture for Better Crops

Bedrich Benes, Duke Pauli, Fiona McCarthy, James Schnable

Implementing machine learning approaches to connect genotype-phenotype has been hindered by the lack of available, labeled training datasets. To overcome this limitation, we will use geometric modeling and object reconstruction, using point cloud data, to develop simulated datasets of organismal development. These simulated growth models will generate labeled datasets which can be used by machine learning algorithms as training data to study the complexities of phenotypic diversity. Moreover, we will develop and test a system for illumination estimation of the virtual crops.

Aim 1: We will provide a large, labeled dataset of point cloud data for sorghum plants with varying precision for ML and evaluation by the plant science community. We will develop the generative procedural model of sorghum that will be parameterized by branching angles, plant age, etc.

Aim 2: We will provide a highly parallel path tracer simulating sorghum illumination at a photonic level at varying wavelengths. This simulator will be expandable for further BRDF and BTDF values and will be parameterizable for varying longitude and latitude.

  1. Novel 3D simulations of sorghum grown at high-density with plant parts segmented & labeled,
  2. Plant and organ labeled simulated point cloud data for sorghum at high planting densities,
  3. ML models using simulated segmented plant data tested on greenhouse and field LIDAR data,
  4. Improved understanding of light interception and interaction dynamics with plant canopy,
  5. Engagement with animal researchers to assess how to extend approaches to animal science,
  6. Hosted AG2PI workshop to engage with plant and animal scientists and expand community.

Machine Learning Competitions for G2P and End-of-Season Phenotype Prediction

Abby Stylianou
Download Narrative

In this project, we will organize competitions to engage with the broader machine learning community to produce models that can answer phenomic questions, using curated datasets from the Department of Energy ARPA-E Transportation Energy Resources from Renewable Agriculture Phenotyping Reference Platform (TERRA-REF) program [Burnette et al., 2018]. The TERRA-REF program aimed to transform plant breeding by using remote sensing from a state of the art field scanner gantry system, seen in Figure 1 (top) to increase the speed at which plant traits can be measured. The field scanner includes a number of sensors, including a millimeter resolution laser 3D scanner, high resolution stereo-RGB cameras, multiple hyperspectral sensors, and a thermal camera, among others (example data products are shown at the bottom of Figure 1). Over the course of several seasons, this sensor collected over a petabyte of sensor data for bioenergy sorghum lines, their corresponding genetic data, a large volume of ground truth measurements of plant phenotypes and growing conditions, and a baseline set of algorithmic approaches for extracting phenotypic data.

We propose that structured machine learning contests are an excellent way to share data and ensure that the desired scientific questions are actually the ones that are answered. Contests, where a specific problem is shared, training and testing data are provided, and a specific evaluation protocol is defined, are a frequent and popular means of advancing results to specific questions within the machine learning community. For example, the Fine-Grained Visual Classification community hosts annual contests on difficult visual recognition problems, such as the iNaturalist contest to recognize different species in image data, which had 1,477 submissions last year from 249 different competitors [Van Horn et al., 2018]. These communities are hungry for well organized datasets with specific scientific questions.

Identifying Educational Resources and Gaps in AG2P Data Science Across Plant and Animal Agriculture Genomics

Breno Fragomeni, Cedric Gondro, Margaret Young, Gabriella Dodd, Tasia Taxis
Download Narrative

Recent developments in genomics and the implementation of new technologies in agriculture have enabled a new research horizon in the field (Harper et al., 2018, Morota et al., 2018). The quality and quantity of genomic data, including microbiome, gene expression, high density SNP markers, sequence data, among others, have great potential to improve both plant and animal science enterprises. However, with these new developments a new challenge has arisen: practitioners must be able to manage the large data sets that result from these technologies – a skill that has traditionally is not been taught during the training of agricultural scientists (Eisen, 2008).

The goal of the Seed Grant is to catalog the available resources and resource gaps in data science to support the Agricultural Genome to Phenome (AG2P) initiative and to outline solutions to fill the gaps. We will develop surveys to identify how aware students and researchers are of the available resources. Additionally, we will create an online repository linking to available training materials in both plant and animal agricultural data science. In this repository we aim to provide the community a unified access point to information about workshops, seminars, online and in-person classes, and course curricula for different career stages. We will prepare a white paper describing our findings based on the survey and the catalog, focusing on how to advance data science education in AG2P. We will fund a graduate student to carry out this work. This student will work in the laboratory of PI Fragomeni with additional support from our team of investigators. We will use the results of the seed grant as preliminary data to apply for a large educational project to develop solutions to the needs identified in this project.

Cattle Genome to Herd Phenotyping for Precision Agriculture

Stephanie McKay, Darren Hagen, Robert Schnabel, Brenda Murdoch

The overarching goal of the Cattle Genome to Herd Phenotyping for Precision Ag initiative is to exploit new phenotyping technologies and high throughput genomics to improve cattle productivity and profitability. To accomplish this, we will establish a network of researchers from a variety of disciplines and agencies to facilitate generation and implementation of next generation phenotyping technologies in cattle and ascertain existing resources.

Expected Outcomes and Deliverables: PIs McKay, Hagen, Schnabel and Murdoch expect to form a CG2HP working group consisting of scientists from a variety of disciplines (i.e. nutrition, physiology, engineering and economics). Scientists will represent leaders from universities, government, industry and breed associations. This working group will be expected to attend an initial virtual meeting in September and meet in person in San Diego in January 2022. A CG2HP concept presentation will be given at PAG and a concept paper addressing the needed phenotyping data and technologies necessary to implement CG2HP in cattle will be generated and published. Further directions and future action items will be discussed, including future funding opportunities. Additionally, a presentation will be given at an AG2PI conference or workshop.