Maize Feature Store: A centralized resource to manage and analyze curated maize multi-omics features for machine...

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Big-data analysis of maize genomes is crucial for genetic research and improving agronomic traits.
• Integrating diverse multi-omics datasets and extracting meaningful features are key challenges.
• Machine learning models require high-quality features for successful application in genomics.

Purpose of the Study:

• To present the Maize Feature Store (MFS), an application for hosting and managing maize multi-omics datasets.
• To provide an end-to-end solution for evaluating and linking features to gene annotations.
• To facilitate exploration, modeling, and analysis of complex maize genetic data.

Main Methods:

• Developed the Maize Feature Store (MFS) as a versatile application.
• Integrated diverse genomic, transcriptomic, epigenomic, variomic, and proteomics datasets.
• Populated the MFS with over 14,000 gene-based features for the maize reference genome.

Main Results:

• Created an accurate pan-genome classification model using the MFS.
• Achieved an Area Under the Receiver Operating Characteristic Curve (AUC-ROC) score of 0.87 for the classification model.
• Demonstrated the utility of the MFS in managing and utilizing multi-omics features.

Conclusions:

• The MFS provides a centralized solution for maize multi-omics data and feature management.
• The MFS accelerates machine learning applications in maize genetics research.
• The MFS is publicly accessible, supporting broader scientific exploration and discovery.