Integrating Biological Knowledge for Robust Microscopy Image Profiling on De Novo Cell Lines.

Area of Science:

• Computational biology
• Biomedical research
• Drug discovery

Background:

• High-throughput screening using microscopy is vital for drug discovery.
• Analyzing new cell lines (de novo) is difficult due to cell line variability.
• Existing models struggle with generalization across diverse cell lines.

Purpose of the Study:

• To develop a novel framework for enhancing microscopy image profiling models.
• To improve the generalization of these models to previously unseen cell lines.
• To address the challenge of cell line heterogeneity in perturbation screening.

Main Methods:

• Integrating external biological knowledge into pretraining strategies.
• Disentangling perturbation-specific and cell line-specific representations.
• Constructing a knowledge graph using protein interaction data (STRING, Hetionet).
• Incorporating transcriptomic features from single-cell foundation models.

Main Results:

• The proposed framework significantly improves microscopy image profiling for de novo cell lines.
• Demonstrated effectiveness through one-shot and few-shot fine-tuning experiments on RxRx datasets.
• Enhanced model generalization capabilities across different cell lines.

Conclusions:

• The novel framework effectively leverages external biological information for better cell line analysis.
• This approach shows promise for real-world phenotype-based drug discovery applications.
• Disentangled representations are key to overcoming cell line heterogeneity challenges.