DELVE: feature selection for preserving biological trajectories in single-cell data.

Area of Science:

• Single-cell biology
• Computational biology
• Genomics

Background:

• Single-cell technologies enable high-dimensional molecular profiling of individual cells.
• Analyzing cell development trajectories is crucial but challenging due to noise in single-cell data.
• Identifying dynamic molecular features that define cell states and transitions is a key problem.

Purpose of the Study:

• To present DELVE, an unsupervised feature selection method for identifying robust molecular features that recapitulate cellular trajectories.
• To address the challenge of feature selection in noisy single-cell data for trajectory inference.
• To improve the definition of cell types and cell-type transitions.

Main Methods:

• DELVE employs a bottom-up approach to feature selection, mitigating confounding variation.
• It models cell states using dynamic gene or protein modules linked to regulatory complexes.
• The method was validated using simulations, single-cell RNA sequencing, and iterative immunofluorescence imaging data.

Main Results:

• DELVE effectively identifies a representative subset of molecular features that robustly capture cellular trajectories.
• The selected features provide a clearer definition of cell types and transitions compared to existing methods.
• The approach successfully models cell states based on core regulatory complexes.

Conclusions:

• DELVE offers a powerful unsupervised method for feature selection in single-cell analysis.
• It enhances the ability to study dynamic biological processes like cell cycle and differentiation.
• The open-source package facilitates broader application in single-cell research.