Tomatoes are among the most produced and consumed vegetable crops globally, yet a wide range of destructive diseases threatens their production and fruit quality. Automated image-based systems powered by deep learning offer consistent, high-throughput disease assessments that are a scalable alternative to manual scouting. This collaborative project aims to develop an autonomous tomato disease scouting system using Unmanned Aerial Vehicles (UAVs) and Unmanned Ground Vehicles (UGVs). The EnBiSys Lab contributes by generating and leveraging high-quality 2D and 3D image representations of tomato plants to support model development and extract enriched information from controlled sensing modalities not typically available in production environments. These insights are used to improve disease assessments and distill knowledge into practical, deployable tools. The goal is to reduce bottlenecks in deep learning deployability by generating robust multimodal data to close the lab-to-field domain gap and develop models for edge hardware.

This project aims to identify signaling components of the nitrogen deficiency response in Arabidopsis thaliana. Using mutant lines, our collaborators are producing datasets spanning transcriptomics, proteomics, metabolomics, and microbial community profiles. Our group is performing integrative analysis of these data to identify signaling mechanisms which contribute to nitrogen deficiency response.

Biotic stresses, such as SLB, may visually appear on a crop and allow for the quantification of infected regions. The quantification of such stress, which includes the scoring of plants based on the proportion of infected areas to healthy tissue, is often performed manually based on visual symptoms. The process is therefore time intensive, causing crop loss due to a delay between disease identification and management efforts. The EnBiSys Lab is taking two parallel approaches to automate the detection and grading of SLB in corn, which allows for disease detection at the plant and the plot levels. The first approach is focused on using computer vision and machine learning to perform quantification and prediction of biotic stress in corn plants. The second approach focuses on using hyperspectral images of whole fields of corn to detect and grade areas affected by SLB.

Accurate crop yield prediction is challenging due to environmental variability and data limitations. This research investigates how deep learning models can be used to improve crop yield prediction when data is limited and environmental conditions change over time. To address this challenge, the EnBiSys Lab is applying pre-trained deep learning models with transfer learning to enhance the model’s ability to capture complex environmental variables. Specifically, the lab focuses on using feature extraction capabilities from pre-trained models and fine-tuning them on publicly available crop-specific agricultural datasets to improve prediction accuracy. Currently, the study focuses on soybean and canola, two economically significant crops with distinct environmental and genetic characteristics. By validating this approach to these crops, the lab aims to develop a generalized deep-learning framework that can extend to other major crops to improve yield prediction accuracy across diverse agricultural environments.

Regulation of gene expression is of paramount importance in animal development, with improper regulation resulting in developmental defects and diseases. In developing tissues, several genes coding for transcription factors regulate each other in a complex web of interactions known as the genetic regulatory network (GRN). The structure of a GRN is thought to be responsible for the robust and precise cell fate decisions required in a developing tissue. The long-term goal is to deduce the genetic regulatory interactions necessary for robust patterns of gene expression. The overall objective in this proposal is to use the natural variation that occurs in a panel of wild-caught fly lines to characterize the GRN responsible for precise anterior-posterior (AP) patterning in the early Drosophila embryo. This will test the central hypothesis that gene expression patterns in the AP patterning system have undiscovered regulation that can be found by examining the correlation between gene expression and natural variation in the genomes of these flies.

For the project, high throughput screening (HTS) techniques are used to ectopically express transcription factors (TF) in maize protoplasts. RNA-seq data is collected to measure gene expression in cells and to gain insights into the mechanisms of gene regulation. The EnBiSys Lab is leading the development of an improved statistical framework to analyze complex gene expression data. The aim is to overcome technical and biological variations that are typical of HTS protocols. The lab is also taking the lead in deconvolving genetic regulatory networks from the RNA-seq data to uncover cellular programming and produce new understandings of the function of specific TFs. These experiments will offer a greater understanding of TF function and aid in engineering plant systems. This project aims to accelerate plant breeding to create the “crops of tomorrow,” which is increasingly important in a changing climate.

Large language models (LLMs) trained on cross-species genomic sequences now predict gene regulation with unprecedented accuracy, but their internal representations remain opaque, hindering trust. Our study aims to uncover how state-of-the-art plant-DNA LLMs (PlantCaudecus and NTV3) encode species identity, a key determinant of their ability to generalize across taxa and thus to serve as foundational models for plant breeding. Our preliminary findings illuminate the pathway to responsible AI in agriculture. By revealing how models encode biological context, we can assess bias, validate cross-species generalization, and ensure transparency in decision-support tools. If validated, such findings become a bridge between genomic data and actionable breeding strategies. This would enhance efficiency, reduce resource waste, and sustain competitive advantage for producers.

The objective of this project is to develop approaches for Gene Regulatory Network (GRN) inference and multiomics integration to identify drivers of iron deficiency and heat stress response in Sorghum. It is hypothesized that variable iron availability in soils across the world drives changes in how sorghum genotypes allocate carbon. Our work will identify transcription factors and hub genes that may drive the iron deficiency response across Sorghum genotypes and identify connections between these response genes and carbon allocation.