Computing on Phenotypic Descriptions for Candidate Gene Discovery and Crop Improvement.

Area of Science:

• Plant biology
• Bioinformatics
• Computational biology

Background:

• Phenotypic descriptions are crucial for biological research but traditionally rely on manual curation into controlled vocabularies.
• Human curation is time-consuming and does not scale to the vast amount of scientific literature.
• Automated methods for analyzing natural language are emerging as a promising alternative.

Purpose of the Study:

• To explore the application of NLP and ML for automated analysis of plant phenotypic descriptions.
• To develop tools for the plant phenomics community to leverage these automated methods.
• To assess the performance of automated methods compared to human curation.

Main Methods:

• Utilizing natural language processing (NLP) and machine learning (ML) algorithms.
• Applying these techniques to unstructured phenotypic descriptions from scientific literature.
• Comparing the performance of automated data structures against human-curated data.

Main Results:

• Automated methods can create data structures that perform as well as or better than human-curated data.
• These methods show promise for tasks like predicting gene function and biochemical pathway membership.
• Potential for in-field data collection using speech-to-text tools is highlighted.

Conclusions:

• NLP and ML offer scalable and efficient solutions for analyzing plant phenomics data.
• Automated analysis can enhance data aggregation, standardization, and predictive capabilities.
• Future applications include in-field data collection for association genetics and breeding.