SupeRJump: Determining normal and leukemic differentiation fate through semi-supervised jump diffusion modeling

Area of Science:

• Computational Biology
• Genomics
• Hematopoiesis

Background:

• Single cell RNA sequencing (scRNA-seq) offers high resolution into cellular heterogeneity.
• Existing computational models struggle to capture discontinuous differentiation processes, particularly in malignant leukemia.

Purpose of the Study:

• To develop a novel computational framework, SupeRJump, for modeling cell fate decisions.
• To analyze complex differentiation dynamics, including discontinuous transitions in various biological systems.
• To identify cellular biases and transcriptional networks driving differentiation discontinuity.

Main Methods:

• Developed SupeRJump, a supervised cell-fate model based on jump-drift-diffusion processes.
• Implemented a semi-supervised pseudotime strategy and batch correction for lineage fate predictions.
• Utilized absorbing Markov chains for modeling differentiation pathways.

Main Results:

• Applied SupeRJump to human bone marrow, murine aging hematopoiesis, and acute myeloid leukemia models.
• Introduced novel metrics to quantify lineage skewness, progenitor transitions, and discontinuous dynamics.
• Identified cells with differentiation biases and their associated transcriptional networks and gene programs.

Conclusions:

• SupeRJump effectively models complex and discontinuous cell differentiation dynamics.
• The framework provides new insights into cellular heterogeneity and lineage commitment in hematopoiesis and leukemia.
• Identified key gene programs responsible for differentiation discontinuity in malignant cells.