Sensitive detection of rare disease-associated cell subsets via representation learning

Area of Science:

• Biotechnology
• Computational Biology
• Immunology

Background:

• Rare cell populations are crucial in disease initiation and progression, particularly in cancer.
• Identifying these rare subpopulations from high-dimensional single-cell data is challenging.

Purpose of the Study:

• To develop and validate CellCnn, a representation learning approach for detecting rare cell subsets linked to diseases.
• To demonstrate the capability of CellCnn in identifying disease-associated rare cells in various clinical contexts.

Main Methods:

• Employed CellCnn, a novel representation learning framework.
• Applied the method to high-dimensional single-cell measurements from peripheral blood and minimal residual disease samples.

Main Results:

• Successfully identified rare cell subsets associated with paracrine signaling, AIDS onset, and CMV infection in peripheral blood.
• Detected extremely rare leukaemic blast populations (as low as 0.01%) in minimal residual disease-like scenarios.

Conclusions:

• CellCnn is an effective tool for discovering rare cell populations associated with various diseases.
• This approach has significant potential for improving disease diagnosis and monitoring through the analysis of single-cell data.