Investigation of the Influence of Polyamines on Mature Embryo Culture and DNA Methylation of Wheat (Triticum aestivum L.) Using the Machine Learning Algorithm Method

Abstract

Numerous factors can impact the efficiency of callus formation and in vitro regeneration in wheat cultures through the introduction of exogenous polyamines (PAs). The present study aimed to investigate in vitro plant regeneration and DNA methylation patterns utilizing the inter-primer binding site (iPBS) retrotransposon and coupled restriction enzyme digestion-iPBS (CRED-iPBS) methods in wheat. This investigation involved the application of distinct types of PAs (Put: putrescine, Spd: spermidine, and Spm: spermine) at varying concentrations (0, 0.5, 1, and 1.5 mM). The subsequent outcomes were subjected to predictive modeling using diverse machine learning (ML) algorithms. Based on the specific polyamine type and concentration utilized, the results indicated that 1 mM Put and Spd were the most favorable PAs for supporting endosperm-associated mature embryos. Employing an epigenetic approach, Put at concentrations of 0.5 and 1.5 mM exhibited the highest levels of genomic template stability (GTS) (73.9%). Elevated Spd levels correlated with DNA hypermethylation while reduced Spm levels were linked to DNA hypomethylation. The in vitro and epigenetic characteristics were predicted using ML techniques such as the support vector machine (SVM), extreme gradient boosting (XGBoost), and random forest (RF) models. These models were employed to establish relationships between input variables (PAs, concentration, GTS rates, Msp I polymorphism, and Hpa II polymorphism) and output parameters (in vitro measurements). This comparative analysis aimed to evaluate the performance of the models and interpret the generated data. The outcomes demonstrated that the XGBoost method exhibited the highest performance scores for callus induction (CI%), regeneration efficiency (RE), and the number of plantlets (NP), with R² scores explaining 38.3%, 73.8%, and 85.3% of the variances, respectively. Additionally, the RF algorithm explained 41.5% of the total variance and showcased superior efficacy in terms of embryogenic callus induction (ECI%). Furthermore, the SVM model, which provided the most robust statistics for responding embryogenic calluses (RECs%), yielded an R² value of 84.1%, signifying its ability to account for a substantial portion of the total variance present in the data. In summary, this study exemplifies the application of diverse ML models to the cultivation of mature wheat embryos in the presence of various exogenous PAs and concentrations. Additionally, it explores the impact of polymorphic variations in the CRED-iPBS profile and DNA methylation on epigenetic changes, thereby contributing to a comprehensive understanding of these regulatory mechanisms.

Keywords: DNA methylation; genomic template stability; iPBS; machine learning.

Figure 1

iPBS profiles for various PA experimental groups with 2384 primers. M: 100–1000 bp DNA ladder; 1: control; 2: 0.5 mM Put; 3: 1 mM Put; 4: 1.5 mM Put; 5: 0.5 mM Spd; 6: 1 mM Spd;7: 0.5 mM Spd; 8: 0.5 mM Spm; 9: 1 µM Spm; 10: 1.5 mM Spm.

Figure 2

DNA methylation changes in the wheat exposed to PAs: (A) total band; (B) polymorphism; (C) GTS value as estimated using different MSH experimental groups.

Figure 3

CRED–iPBS profiles for various PA experimental groups with iPBS 2384 primers; M, 100–1000 bp DNA ladder; 1: control Hpa II; 2: control Msp I; 3: 0.5 mM Put Hpa II; 4: 0.5 mM Put Msp I; 5: 1 mM Put Hpa II; 6: 1 mM Put Msp I; 7: 1.5 mM Put Hpa II; 8: 1.5 mM Put Msp I; 9: 0.5 mM Spd Hpa II; 10: 0.5 mM Spd Msp I; 11: 1 mM Spd Hpa II; 12: 1 mM Spd Msp I; 13: 1.5 mM Spd Hpa II; 14: 1.5 mM Spd Msp I; 15: 0.5 mM Spm Hpa II; 16: 0.5 mM Spm Msp I; 17: 1 mM Spm Hpa II; 18: 1 mM Spm Msp I; 19: 1.5 mM Spm Hpa II; 20: 1.5 mM Spm Msp I.

Figure 4

The effect of PAs on polymorphism percentages in different experimental groups of wheat in the seedling growth stage.