Predicting flowering time using integrated morphophysiological and genomic data with machine learning models
Authors:
- Mehdi BabaeiMehdi Babaei
- Hossein NematiHNHossein Nemati
- Hossein ArouieeHAHossein Arouiee
- Davoud TorkamanehDTDavoud Torkamaneh
Preprint
Abstract
Indigenous Cannabis Sativa populations have adapted to diverse environments, resulting in genetic and phenotypic diversity. Understanding the mechanisms underlying flowering time variation is crucial for optimizing cultivation and breeding. This study employed a novel approach combining temporal phenotypic analysis, genomic data, and machine learning (ML) to identify key features associated with early, medium, and late flowering in cannabis landraces. We collected weekly data on six morphophysiological traits—stem diameter, height, growth rate, node number, internode length, and SPAD chlorophyll index—from 25 cannabis landrace populations 13 weeks for female plants and 11 weeks for male plants. Additionally, 145 accessions were genotyped using high-density genotyping-by-sequencing, resulting in 233,624 high-quality single nucleotide polymorphisms (SNPs). A comprehensive ML framework integrating mutual information (MI), recursive feature elimination (RFE), random forest (RF), and support vector machine (SVM), was used to investigate 234,002 features, encompassing SNPs, morphophysiological traits, and environmental factors. This approach identified 53 key features—22 genetic variants and 31 morphophysiological traits—that effectively distinguish between early, medium, and late flowering types with an accuracy of 96.6%. The identified SNPs were distributed across multiple chromosomes, including chromosomes 08, 09, and X. Notably, key loci like AutoFlower3 ( CsFT3 ) (on chromosome 08) and CircadianFloweringLocus1 ( CsCFL1 ) (on chromosome 09) were identified, with several SNPs located within or near annotated genes. These findings contribute significantly to the understanding of cannabis chronobiology and support the development of “smart crop” strategies by providing valuable markers for early selection and targeted breeding programs aimed at optimizing flowering time under diverse conditions. Key Message: A data-driven machine learning strategy combining genomic and dynamic phenotypic traits enables accurate classification of flowering time in diverse Cannabis landraces.Keywords
This publication has 112 references indexed in Scilit:
- Exploring the multifaceted dynamics of flowering time regulation in field crops: Insight and intervention approachesThe Plant Genome, 2025
- Hormonal influence on maize inflorescence development and reproductionPlant Reproduction, 2024
- Bioinformatics for Molecular Breeding and Enhanced Crop Performance2024
- Exploring ethylene-related genes in Cannabis sativa: implications for sexual plasticityPlant Reproduction, 2024
- Chemical and Morphological Phenotypes in Breeding of Cannabis sativa L.2017
- The Hormonal Regulation of Flower DevelopmentJournal of Plant Growth Regulation, 2010
- Secondary metabolism in cannabisPhytochemistry Reviews, 2008
- Analytical characterization of hemp (Cannabis sativa) seed oil from different agro‐ecological zones of PakistanJournal of the American Oil Chemists' Society, 2006
- Newer Classification and Regression Tree Techniques: Bagging and Random Forests for Ecological PredictionEcosystems, 2006
- The inheritance of chemical phenotype in Cannabis sativa L. (II): Cannabigerol predominant plantsEuphytica, 2005