Velo-Predictor

Velo-Predictor predicts RNA velocity from single-cell RNA sequencing (scRNA-seq) data to infer dynamic changes in cell states.

Key Features:

• Supervised Learning Classification: Formulates RNA velocity prediction as a supervised classification problem by dividing the cell state space into equal-sized segments and using directions as classes with estimated RNA velocity vectors as ground truth.
• Ensemble Learning with XGBoost: Implements an ensemble learning pipeline and reports best performance when using XGBoost as the base predictor.
• Data Imputation: Imputes RNA velocities for unobserved or sparse cell states in scRNA-seq datasets to extend velocity estimates across the cell state space.
• Parameter Analysis and Visualization: Provides parameter analysis and visualization to construct continuous landscapes of cellular dynamics and support biologically meaningful interpretation.

Scientific Applications:

• RNA velocity estimation: Enables estimation of RNA velocity from gene expression data in single-cell transcriptomics.
• Cellular differentiation and development: Supports analysis of dynamic changes during cellular differentiation and development by predicting state transitions.
• Disease progression: Facilitates mapping of state trajectories to study disease progression at single-cell resolution.
• Continuous landscape construction: Aids construction of continuous cellular-dynamics landscapes for interpreting population-level trends.

Methodology:

Treats prediction as a supervised classification problem by segmenting the cell state space into equal-sized direction classes and using estimated RNA velocity vectors as ground truth; employs an ensemble learning pipeline with XGBoost as the base predictor; performs imputation of RNA velocities for unobserved cell states.