The study utilizes a large collection of fluorescently marked Ds-GFP insertional mutations in haploid maize pollen to link gene disruptions with quantitative fitness effects measured as transmission deviations. By integrating genome-derived features (e.g., codon usage) and expression profiling into interpretable machine learning models, they achieve high predictive performance (auROC >90%) for genes influencing pollen fitness, highlighting expression specificity as a key predictor.

The authors used a bottom‑up thermodynamic modelling framework to investigate how plants decode calcium signals, starting from Ca2+ binding to EF‑hand proteins and extending to higher‑order decoding modules. They identified six universal Ca2+-decoding modules that can explain variations in calcium sensitivity among kinases and provide a theoretical basis for interpreting calcium signal amplitude and frequency in plant cells.

The study benchmarked over 20 web‑based gRNA on‑target efficiency prediction tools against an experimental plant CRISPR editing dataset, finding several machine‑learning based tools whose scores strongly correlated with observed InDel frequencies. Additionally, the performance of popular platforms such as CRISPOR and CRISPR‑P was assessed, offering guidance for improved gRNA design in plant genome editing.

The study evaluates the use of single-cell RNA sequencing (scRNA-seq) data to predict plant metabolic pathway genes (MPGs) in Arabidopsis thaliana, comparing five multi-label machine‑learning algorithms against traditional bulk RNA‑seq approaches. scRNA‑seq generated co‑expression networks that, while different, yielded significantly higher MPG classification accuracy, especially when data were split by genetic background or tissue type, and deep learning outperformed classical methods. The authors conclude that scRNA‑seq offers superior predictive power and should be incorporated into future MPG discovery pipelines.

The study integrated weekly morphophysiological measurements with high-density genotyping-by-sequencing data and a machine‑learning pipeline to dissect flowering time variation in diverse Cannabis sativa landraces. By applying mutual information, recursive feature elimination, random forest, and support vector machine classifiers to over 234,000 combined genetic, phenotypic, and environmental features, the authors identified 53 key markers that classify early, medium, and late flowering types with 96.6% accuracy. Notable loci, including CsFT3 and CsCFL1, were highlighted as promising targets for breeding and smart‑crop strategies.

The study used extensive gravimetric load‑cell and ambient sensor data collected over seven years from hundreds of greenhouse-grown crops to train machine‑learning models for predicting daily whole‑plant transpiration. Random Forest and XGBoost achieved the highest accuracy (R² up to 0.89), with ambient temperature identified as the dominant driver. These results highlight the promise of ML‑based tools for precise agricultural water management.

Foliar application of Trichoderma harzianum cell‑free culture filtrates (CF) increased fruit yield, root growth, and photosynthesis in a commercial tomato cultivar under prolonged water deficit in a Mediterranean greenhouse. Integrated physiological, metabolite, and transcriptomic analyses revealed that CF mitigated drought‑induced changes, suppressing about half of water‑stress responsive genes, thereby reducing the plant’s transcriptional sensitivity to water scarcity.

The study used high-resolution chemical imaging to map cell-type specific metabolic changes in plant roots inoculated with a nine-strain endophyte consortium under drought, revealing that endophytes differentially alter root metabolomes across spatial domains. Machine learning identified metabolites and exudates predictive of drought and endophyte treatment, and correlation analyses showed dynamic endophyte–metabolite relationships under stress.

The study introduces ENTRAP-seq, a high‑throughput in‑planta assay that couples protein‑coding libraries with a nuclear magnetic sorting‑based reporter to multiplexively assess transcriptional regulatory activity of thousands of protein variants. Using this platform and machine‑learning analysis, the authors screened 1,495 plant viral proteins, uncovering numerous novel regulatory domains, and applied machine‑guided, semi‑rational design to modify the activity of a plant transcription factor.

The study used phylogeny‑based analyses of 36 legume genomes and a newly created multiparent advanced generation intercross (MAGIC) population of common bean to predict and characterize genome‑wide deleterious mutations. Machine‑learning integration of conservation and protein features identified thousands of potentially deleterious sites, whose variation correlated negatively with flowering time, maturity, and yield, highlighting the impact of genetic load on breeding performance.