Single-cell pseudotrack inference method, system, and device fusing graph structure and reinforcement learning

By fusing graph structures and reinforcement learning, the computational complexity and robustness issues in single-cell trajectory inference are addressed, enabling high-resolution reconstruction of cell differentiation paths and pseudo-temporal prediction, thereby improving the biological continuity and global robustness of trajectory inference.

CN122245410APending Publication Date: 2026-06-19WUHAN INST OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
WUHAN INST OF TECH
Filing Date
2026-02-10
Publication Date
2026-06-19

AI Technical Summary

Technical Problem

Existing single-cell trajectory inference methods are computationally complex when dealing with large-scale, high-noise data, and have limited ability to resolve complex branch trajectories. The global robustness and biological continuity of trajectory inference need to be improved.

Method used

This paper proposes a method that integrates graph structure and reinforcement learning. It constructs an adaptive connected graph through nonlinear dimensionality reduction, performs reinforcement learning by combining graph topology metrics and composite reward functions, solves the global optimal path using Dijkstra's algorithm, and performs pseudo-time prediction by combining a random forest regression model.

Benefits of technology

It achieves high-resolution reconstruction of cell differentiation pathways in complex nonlinear manifolds, ensuring the biological continuity and computational stability of trajectory inference, and can adapt to various complex topological structures, improving the accuracy of lineage structure recognition and the consistency of pseudo-temporal ordering.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122245410A_ABST
    Figure CN122245410A_ABST
Patent Text Reader

Abstract

This invention belongs to the field of single-cell data analysis technology. Addressing the shortcomings of existing trajectory inference methods in terms of global robustness and biological continuity, this invention proposes a single-cell pseudo-trajectory inference method and system that integrates graph structure and reinforcement learning. It involves nonlinear manifold dimensionality reduction of multidimensional single-cell sequencing data, constructing an adaptive connected graph representing cell states using the obtained low-dimensional embedding representation, locating developmental initiation and termination nodes based on graph topological metric features, defining a reward function based on graph structure distance to guide the reinforcement learning agent in identifying key developmental trajectories and bifurcation structures, and inferring continuous cell fate trajectories in multi-branch lineages based on their topological features through global optimal path backtracking and global mapping strategies. This invention achieves accurate prediction and coverage of pseudo-time for all cells in the global domain, ensuring the integrity and high resolution of trajectory reconstruction.
Need to check novelty before this filing date? Find Prior Art