Expression-based machine learning models for predicting plant tissue identity.

Area of Science:

• Genomics
• Plant Biology
• Bioinformatics

Background:

• Arabidopsis thaliana is a widely used model organism in plant genomic research.
• Its selection facilitated genome-enabled research, but its broad applicability is questioned.

Purpose of the Study:

• To evaluate the translatability of Arabidopsis gene expression data for predicting tissue identity in other flowering plants.
• To compare the effectiveness of different machine learning algorithms for this predictive task.

Main Methods:

• Developed and tested machine learning models using Arabidopsis gene expression data.
• Assessed model performance in predicting tissue identity within Arabidopsis and across diverse flowering plant species.
• Compared various algorithms, including k-nearest neighbors.

Main Results:

• Models trained on Arabidopsis data achieved high accuracy for within-species predictions.
• Cross-species predictions showed moderate precision (0.69–0.74) and recall (0.54–0.64).
• Belowground tissue prediction was more accurate; k-nearest neighbors performed best, highlighting gene expression signatures over marker genes.

Conclusions:

• Knowledge derived from Arabidopsis is not universally transferable to all flowering plants.
• The study advocates for re-evaluating the emphasis on Arabidopsis and prioritizing plant diversity in genomic research.
• Gene expression signatures are crucial for developing robust plant tissue and cell type prediction models.