ProtoCell4P: an explainable prototype-based neural network for patient classification using single-cell RNA-seq.

Area of Science:

• Computational Biology
• Genomics
• Machine Learning in Medicine

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution gene expression data for biological studies.
• Classifying patient phenotypes using scRNA-seq faces challenges like variable cell counts per sample, limited sample size, and high dimensionality.
• Existing deep learning models lack interpretability, hindering the extraction of biological knowledge.

Purpose of the Study:

• To develop a deep learning model for accurate patient phenotype classification from scRNA-seq data.
• To address challenges of variable cell numbers and sample scarcity in scRNA-seq datasets.
• To provide an interpretable model for identifying key cells and cell types driving disease phenotypes.

Main Methods:

• Proposed a prototype-based and cell-informed deep learning model named ProtoCell4P.
• Leveraged cell prototypes to represent cell knowledge and handle diverse cell numbers.
• Incorporated adaptive information from different cells for precise patient classification.

Main Results:

• ProtoCell4P effectively classifies patient phenotypes using scRNA-seq data, outperforming existing methods.
• The model demonstrates explainability at the single-cell resolution, identifying key cells for classification.
• The approach uncovers associations between cell types and biological classes from a data-driven perspective.

Conclusions:

• ProtoCell4P offers a robust and interpretable solution for patient phenotype classification using scRNA-seq data.
• The model's interpretability facilitates the discovery of disease-driving cells and biological insights.
• This approach advances the application of deep learning in single-cell genomics for clinical research.