Genome-wide cis-decoding for expression design in tomato using cistrome data and explainable deep learning.

Area of Science:

• Plant Molecular Biology
• Genomics
• Evolutionary Biology

Background:

• Variation in cis-regulatory elements (CREs) drives plant trait evolution and gene expression diversification.
• Predicting gene expression from CREs is challenging due to biological complexity.

Purpose of the Study:

• To develop a predictive model for genome-wide expression patterns in tomato fruit ripening using DNA sequences.
• To identify critical nucleotide residues within CREs that influence gene expression.

Main Methods:

• Utilized cistrome datasets and explainable convolutional neural network (CNN) frameworks.
• Employed single-cell type spatiotemporal transcriptome data to control for trans-acting factors.
• Performed experimental validation of identified critical nucleotide residues.

Main Results:

• Developed a CNN-based prediction model for crucial expression patterns during tomato fruit ripening initiation.
• Identified specific nucleotide residues critical for objective expression patterns through CNN feature visualization.
• Experimentally validated the functional impact of these identified nucleotide residues.

Conclusions:

• The cis-decoding framework advances understanding of CREs and transcription factor interactions in regulatory networks.
• This approach offers a method for designing alleles to optimize gene expression in plants.