Genome-wide association studies (GWAS) are widely used to infer the genetic basis of traits in organisms; however, selecting appropriate thresholds for analysis remains a significant challenge. In this study, we introduce the Sequential SNP Prioritization Algorithm (SSPA) to investigate the genetic underpinnings of two key phenotypes in Sorghum bicolor: maximum canopy height and maximum growth rate. Using a subset of the Sorghum Bioenergy Association Panel cultivated at the Maricopa Agricultural Center in Arizona, we performed GWAS with specific permissive-filtered thresholds to identify genetic markers associated with these traits, enabling the identification of a broader range of explanatory candidate genes. Building on this, our proposed method employed a feature engineering approach leveraging statistical correlation coefficients to unravel patterns between phenotypic similarity and genetic proximity across 274 accessions. This approach helps prioritize Single Nucleotide Polymorphisms (SNPs) that are likely to be associated with the studied phenotype. Additionally, we conducted a complementary analysis to evaluate the impact of SSPA by including all variants (SNPs) as inputs, without applying GWAS. Empirical evidence, including ontology-based gene function, spatial and temporal expression, and similarity to known homologs demonstrates that SSPA effectively prioritizes SNPs and genes influencing the phenotype of interest, providing valuable insights for functional genetics research.
Accurate phenology models are important to predict how global climate change will continue to alter the timing of plant phenological events, such as spring greenup in deciduous broadleaf forests. While there is merit in long-term predictions, investigating how models can predict near-term (1– 35 days) canopy greenness throughout the spring allows us to validate performance and understanding now. The Ecological Forecasting Initiative’s NEON Forecasting Challenge, is an open challenge to the community to predict daily greenness values, measured through digital images collected by the PhenoCam Network at National Ecological Observatory Network (NEON) sites. For the first round of the challenge, which is presented here, teams were tasked to forecast phenology at eight deciduous broadleaf sites. A total of 192,536 predictions were submitted, representing eighteen models, including a persistence and a day of year mean null models. We found that forecast skill was highest when forecasting earlier in the greenup curve compared to the end, for shorter lead times, for sites that greened up earlier, and when submitting forecasts during times other than near budburst. The models based on day of year historical mean had the highest predictive skill across the challenge period. In this first round of the challenge, by synthesizing across forecasts, we started to elucidate what factors affect the predictive skill of near-term phenology forecasts, which future rounds will continue to investigate.
One of the major challenges in machine learning is maintaining the accuracy of the deployed model (e.g., a classifier) in a non-stationary environment. The non-stationary environment results in distribution shifts and, consequently, a degradation in accuracy. Continuous learning of the deployed model with new data could be one remedy. However, the question arises as to how we should update the model with new training data so that it retains its accuracy on the old data while adapting to the new data. In this work, we propose a task-conditioned ensemble of models to maintain the performance of the existing model. The method involves an ensemble of expert models based on task membership information. The in-domain models-based on the local outlier concept (different from the expert models) provide task membership information dynamically at run-time to each probe sample. To evaluate the proposed method, we experiment with three setups: the first represents distribution shift between tasks (LivDet-Iris-2017), the second represents distribution shift both between and within tasks (LivDet-Iris-2020), and the third represents disjoint distribution between tasks (Split MNIST). The experiments highlight the benefits of the proposed method. The source code is available at https://github.com/iPRoBe-lab/Continuous_Learning_FE_DM.
Plant phenology and phenotype prediction using remote sensing data is increasingly gaining attention in order to enhance agricultural productivity. This work aims to generate synthetic forestry images that satisfy a specific phenotypic attribute, viz. canopy greenness. We harness a Generative Adversarial Network (GAN) to generate biologically plausible and phenotypically stable forestry images conditioned on the greenness of vegetation over a specific region of interest (describing a particular vegetation type in a mixed forest). The training data is based on the automated digital camera imagery captured by the National Ecological Observatory Network (NEON) and processed by the PhenoCam Network. This method helps render the appearance of forest sites specific to a greenness value. Further, synthetic images are utilized to predict another phenotypic attribute, viz., redness of plants. The Structural SIMilarity (SSIM) index is used to assess the quality of the synthetic images and their greenness and redness indices are compared against that of the original images using Root Mean Squared Error (RMSE) to evaluate the accuracy and integrity. The generalizability and scalability of our proposed GAN model is determined by effectively transforming it to generate synthetic images for other forest sites and vegetation types. From a broader perspective, this technique could be used to visualize forestry based on different phenotypic attributes in the context of various environmental parameters. This work provides a useful step in leveraging generative AI principles from pattern recognition and computer vision for plant phenological research.