Velo-Predictor: an ensemble learning pipeline for RNA velocity prediction

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• RNA velocity infers cell state dynamics from single-cell RNA sequencing (scRNA-seq) data.
• Accurate RNA velocity estimation remains challenging due to data sparsity and unclear kinetic mechanisms.
• Imputation is often needed for cell states not covered in scRNA-seq datasets.

Purpose of the Study:

• To develop a robust method for predicting RNA velocity from scRNA-seq data.
• To address the challenges of data sparsity and improve the accuracy of dynamical cell state inference.
• To provide a simplified and intuitive tool for analyzing cellular dynamics.

Main Methods:

• Formulated RNA velocity prediction as a supervised classification learning problem.
• Developed Velo-Predictor, an ensemble learning pipeline utilizing XGBoost as a base predictor.
• Trained and tested the pipeline on two real-world scRNA-seq datasets.

Main Results:

• Velo-Predictor demonstrated good performance in predicting RNA velocities.
• The ensemble method, particularly with XGBoost, showed robust and biologically meaningful predictions.
• Parameter analysis and visualization confirmed the method's reliability.

Conclusions:

• Velo-Predictor effectively simplifies RNA velocity prediction using gene expression data.
• The method aids in constructing continuous cellular landscapes for intuitive analysis of dynamics.
• This approach offers biologists a clearer picture of cellular trend trajectories.