Unsupervised and semi-supervised learning: the next frontier in machine learning for plant systems biology

Area of Science:

• Plant biology
• Bioinformatics
• Computational biology

Background:

• High-throughput omics technologies generate big data in plant biology.
• Machine learning (ML) is vital for analyzing this big data in plant systems biology.
• Supervised ML requires extensive labeled data, which is often unavailable or costly.

Purpose of the Study:

• To review unsupervised learning (UL) and semi-supervised learning (SSL) in plant systems biology.
• To introduce basic ML concepts and representative UL/SSL algorithms.
• To discuss the applications, limitations, and future challenges of UL/SSL in plant research.

Main Methods:

• Introduction to machine learning concepts.
• Explanation of unsupervised learning algorithms (clustering, dimensionality reduction, self-supervised learning).
• Explanation of semi-supervised learning algorithms (positive-unlabeled learning, transfer learning).

Main Results:

• UL and SSL are indispensable when labeled training data is limited in plant biology.
• These methods have diverse applications in plant systems biology and phenotyping.
• Recent advances demonstrate the growing utility of UL/SSL paradigms.

Conclusions:

• UL and SSL strategies are significant for overcoming data limitations in plant systems biology.
• Further research is needed to address the limitations and challenges of these approaches.
• These methods are essential for advancing plant big data analysis and research.