Gene expression, transcription factor binding and histone modification predict leaf adaxial-abaxial polarity related genes in Arabidopsis thaliana

Background: Arabidopsis thaliana leaf adaxial-abaxial (ad-abaxial) polarity affects leaf morphology and function. Several genes are known to govern ad-abaxial patterning, but the genetic machinery underlying this process is unclear. Results: To uncover critical genes involved in leaf ad-abaxial patterning, we applied a combination of in silico prediction using machine learning (ML) and experimental analysis. Genes known to influence ad-abaxial polarity were utilized as ground truth to train a Random Forest model. This model used gene expression data from various tissues and conditions and promoter regulation data (based on ChIP-seq data) as input. This approach enabled us to identify novel ad-abaxial polarity-related genes. Parallel to this, available and newly obtained transcriptome data enabled us to identify genes differentially expressed across leaf ad-abaxial sides. The integration of the set of differentially expressed genes with the ML model predictions validated the model and led to our final set of predicted leaf polarity influencing genes. Conclusions: Using a combination of in silico prediction using ML and experimental analysis, we obtained a strategy to identify which genes are involved in leaf ad-abaxial specialization. Our final set of 111 genes expands the knowledge of genetic components underlying leaf ad-abaxial polarity specification.