Supervised learning of high-confidence phenotypic subpopulations from single-cell data
Tao Ren1,2, Canping Chen3,4, Alexey V Danilov5
1Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing, China.
Biorxiv : the Preprint Server for Biology
|March 30, 2023
Summary
PENCIL, a new supervised learning tool, precisely identifies cell subpopulations linked to specific phenotypes in single-cell data. It simultaneously selects genes and predicts cell trajectory changes, advancing biological and clinical research.
Area of Science:
- Single-cell genomics
- Computational biology
- Biomedical data analysis
Background:
- Identifying phenotype-relevant cell subsets is critical for understanding biological and clinical phenotypes.
- Existing methods struggle to simultaneously select informative genes and identify cell subpopulations.
Approach:
- Developed PENCIL, a novel supervised learning framework using a learning with rejection strategy.
- Integrated a feature selection function for simultaneous gene selection and subpopulation identification.
- Incorporated a regression mode for supervised phenotypic trajectory learning.
Key Points:
- PENCIL accurately identifies phenotypic subpopulations missed by other methods.
- Enables simultaneous gene selection and cell subpopulation identification.
- Regression mode facilitates supervised phenotypic trajectory learning from single-cell data.
Conclusions:
- PENCIL offers a scalable and flexible infrastructure for phenotype-associated subpopulation identification.
- Demonstrated utility in identifying T-cell subpopulations in melanoma immunotherapy.
- Revealed drug treatment response trajectories in mantle cell lymphoma using scRNA-seq data.
