Application of Supervised Machine Learning Models with Inter Simple Sequence Repeat Markers for Prediction of Grass Species

Recently, machine learning is used to make accurate predictions using experimental data in biology, and the outcomes of these predictions could be effective for managing future research. In this study, supervised machine learning models in combination with ISSR (inter-simple sequence repeat) markers were used to categorize and predict the grass species and genera. The collected plant materials comprise 46 genotypes representing nine distinct grass species sourced from various regions across Iran. To characterize these accessions, a set of 16 ISSR primers was employed for profiling. The clusters were created by the UPGMA (unweighted pair group method with arithmetic average) clustering algorithm. In order to dimensionality reduction and define the most indicative predictor alleles, weighting algorithms were applied on processed data set. Tree Induction and Naive Bayes were also employed to predict and classify grass species. Performance evaluation of constructed models was performed by cross-validation with 10 times repeats. Based on the results, interspecies genetic diversity was higher than intra-species genetic diversity. In terms of the correlation between grouping patterns and the regions of collection, disparate findings were observed across different genera. The consequences suggested that Naive Bayes and Tree Induction models exhibited effectiveness in predicting grass species. In the Tree Induction models, the highest accuracy of 90.10% was achieved when Decision Tree and Random Forest were run with the information gain ratio criterion on Cdb. However, Bayesian models demonstrated even greater efficiency in predicting grass genera and species, achieving a 100% accuracy rate. This research has successfully developed effective, rapid and inexpensive analytical tools that can be helpful in future grass germplasm management and breeding programs.