Predicting emergent phenotypes from single cell populations using CELLECTION

Area of Science:

• Computational Biology
• Genomics
• Developmental Biology

Background:

• Biological systems display emergent phenotypes from collective component behavior.
• Understanding these emergent properties is key in fields like genetics and developmental biology.

Purpose of the Study:

• To introduce CELLECTION, a deep learning framework for associating instance subgroups with emergent phenotypes.
• To demonstrate CELLECTION's utility in diverse biological research areas.

Main Methods:

• Developed a deep learning framework named CELLECTION.
• Applied CELLECTION to heterogeneous tasks including disease classification and evolutionary analysis.

Main Results:

• CELLECTION enables interpretable predictions for complex biological tasks.
• Successfully applied to identify disease-associated cell subtypes and align developmental stages.
• Used to predict avian relative hand-wing indices, showcasing evolutionary insights.

Conclusions:

• CELLECTION offers a scalable and flexible method for uncovering biological insights.
• Identifies key cellular or genetic signatures underlying complex traits.
• Facilitates research in development, disease, and evolutionary biology.