Skip to content

Methods

1. Backend

1.1 Available backend table

Backend Description Advantage Disadvantage
Python Function The function developed by this project incorporates the latest trajectory inference methods in recent years, making it particularly well-suited to the project's framework. 1. New Methods in recenty years.
1. Different trajectory inference package versions in the same Python environment may conflict.
Conda(recommended) The function run in specific conda environment, related packages don't conflict with other method, which is more flexible.
CFE Docker Docker image for trajectory inference are developed by this project. 1. New Methods in recenty years.
2. The ease of use of Docker.		 1. Docker environment is need.
Dynverse Docker Docker image for trajectory inference refers to dynverse [^dynverse]. 1. The ease of use of Docker 1.Methods on R language not be compatible.
2.Methods are old relatively.
3. Docker environment is need.

1.2 Method Usage

  1. (Recommended) Call the method directly.
cfe.method.paga(fadata, parameters={"cluster_key": "lineage",  "connectivity_cutoff": 0.5})
  1. Create method object with method name, then call it.

method = cfe.method.FateMethod(method_name="paga")
method(fadata, parameters={"cluster_key": "lineage",  "connectivity_cutoff": 0.5})

1.3Reference source

  • Dynverse[^dynverse]: 45 methods filtered from 70 methods before 2019 years. Output results of them can be classfied to 7 wrapper. Paper, github reository, document are available.
  • Github Reporsitory[^sc_pseudotime_github]: A repository keeps track of the latest trajectory inference methods in real-time. Related topics such as upstream opertion(data imputation, dimsional reduction), donstream analysis(GRN inference, trajectory alignment) and reviews are also included.

2.Implementation order

2.1 wrapper and baseline(Completed on 2025.03.11)

Here, baseline methods are easy way to get the specified aimed wrapper input data structure, where MST(Minimum Spanning Tree) are widely used.

  • Dynverse represtive methods for 7 basic wrapper:
  • Direct: PAGA
  • Linear: Component 1(baseline)
  • Cycle: Angle(baseline)
  • Probability: State Component(baseline)
  • Cluster: Cluster MST(baseline)
  • Projection: Projection MST(baseline)
  • Graph: Graph MST(baseline)
  • Velocity wrapper:
  • Strategy from Velocity wrapper to Direct wrapper.
  • represtive method scVelo.
  • CFE Docker:
  • Use Docker to manage environments and version of specific methods.
  • Use Github Action to build and push docker images automatically, (now, the action script is triggerd manually).

2.2 More published methods (Working)

ref: https://github.com/agitter/single-cell-pseudotime

TODO: add paper citation.

  • For other methods, the higher the citation count of the paper, the higher the implementation order(need statistics from google scholar).
Wrapper Type Method Name Finished
Direct PAGA √
Linear Component 1(baseline) √
Palantir √
Cytotrace/Cytotrace2
Cycle Angle(baseline) √
Probability State Component(baseline) √
CellRank
Cluster Cluster MST(baseline) √
Projection Projection MST(baseline) √
Graph Graph MST(baseline) √
Velocity scVelo √
veloVI √
VeloAE
Dynamo

TODO list for velocity methods. Popularity is shown in "Github Stars" and "Paper Citation" column ().

Method Name Finished Time Paper Citation Github Stars
scvelo 2025.09.30 2605 467
velovi 2025.09.30 97 40
veloae 2025.09.30 57 28
dynamo 2025.09.30 385 472
pyrovelocity 2025.10.08 23 47
phylovelo 46 43
latentvelo 30 26
unitvelo 26
regvelo 5 56
deepvelo2 51 46
celldancer 90 70
velovae 35 35
deepvelo 74 14
cell2fate 4 79

Methods to be categorized: WaddingtonOT, TrajectoryNet, pyVIA

TODO

The work integrate trajectory methods from the issue area continuously.

[^dynverse]: Wouter Saelens, Robrecht Cannoodt, Helena Todorov, and Yvan Saeys. A comparison of single-cell trajectory inference methods. Nature biotechnology, 37(5):547–554, 2019. [^sc_pseudotime_github]: Anthony Gitter. Single-cell rna-seq pseudotime estimation algorithms. https://github.com/agitter/single-cell-pseudotime.