Machine learning for perturbational single-cell omics.

Area of Science:

• Cell Biology
• Computational Biology
• Machine Learning

Background:

• Cell biology research faces limitations in collecting comprehensive data on cellular phenotypes and responses to perturbations.
• Advances in big data, deep learning, and computing have enabled solutions in fields like computer vision and speech recognition.
• Machine learning (ML) approaches using single-cell data are emerging for perturbation response prediction.

Purpose of the Study:

• Define objectives for learning perturbation response in single-cell omics.
• Survey existing ML approaches, resources, and datasets for perturbation response studies.
• Explore the potential of a perturbation atlas to enhance deep learning models.

Main Methods:

• Review and define objectives for perturbation response modeling in single-cell omics.
• Survey existing computational approaches, resources, and datasets.
• Discuss the development of a perturbation atlas.
• Examine deep neural network (DNN) applications.

Main Results:

• Identified key objectives and challenges in perturbation response modeling.
• Cataloged existing resources and datasets for single-cell perturbation studies.
• Proposed a framework for a perturbation atlas to inform deep learning.
• Highlighted the potential of DNNs for integrating diverse data and modeling complex systems.

Conclusions:

• Deep learning, powered by single-cell data and a perturbation atlas, can significantly improve the prediction of cellular responses.
• Future research should focus on developing more powerful and explainable deep neural network models for understanding complex biological systems.