Classification of rbcL Gene in Plants Based on Protein K-mer Representation and Random Forest

Abstract

The ribulose-1,5-bisphosphate carboxylase/oxygenase large subunit gene (rbcL) is one of the most widely used molecular markers in plant phylogenetics, DNA barcoding, and taxonomy. Automatic identification of this gene from large-scale protein sequence collections represents a compelling bioinformatics challenge, especially with the rapid growth of publicly available genomic and proteomic data. This study aims to develop a machine-learning–based classification model to distinguish rbcL from other plastid genes such as ndhF, psbA, atpB, and rbcS. Protein sequence data were obtained from the NCBI Protein database and automatically labeled using FASTA metadata. Numerical representation of sequences was performed using protein k-mers (k = 2), yielding 400 features per sequence. A Random Forest model with 300 estimators was trained using an 80% training and 20% testing split. Evaluation using confusion matrix, precision, recall, F1-score, and ROC–AUC demonstrated high classification performance, with accuracy ≥ 0.95 and an AUC value approaching 1.0. These findings indicate that an alignment-free k-mer approach can effectively identify rbcL sequences and has strong potential for application in AI-driven gene annotation pipelines for plant plastid genomes