A mini-review on perturbation modelling across single-cell omic modalities

Area of Science:

• Cellular and Molecular Biology
• Computational Biology
• Genomics

Background:

• Single-cell omics technologies have advanced cellular characterization.
• Perturbation modeling aims to understand external influences on cellular physiology.
• Machine and deep learning are crucial for analyzing complex perturbational data.

Purpose of the Study:

• To review single-cell perturbation technologies and computational methods.
• To highlight challenges and future perspectives in the field.
• To provide a guide for researchers navigating this expanding area.

Main Methods:

• Review of genetic manipulation techniques (e.g., CRISPR) and compound-based perturbations.
• Survey of computational approaches including statistical inference, machine learning (shallow models, autoencoders), and gene regulatory networks.
• Discussion of foundational models and ensemble learning strategies.

Main Results:

• Summary of diverse single-cell perturbation technologies across omic modalities.
• Overview of a wide range of computational methods, from classical to deep learning and biologically informed approaches.
• Identification of emerging trends like large foundational models.

Conclusions:

• Single-cell perturbation modeling is a rapidly evolving field with significant potential.
• Future directions include perturbation atlases, multi-omics, spatial data, and causal machine learning.
• Addressing challenges in interpretability, interoperability, and benchmarking is crucial for future progress.