RNA 速率分析Pseudotime vs latent time

I am doing scRNA-Seq analysis, and I am doing RNA velocity and pseudotime analyses as part of my analysis pipeline. Now, RNA velocity is carried out using scvelo, and this package generates a metric known as latent time. I am having a very had time seeing the difference between latent time and standard pseudotime analysis (usually carried out by slingshot, monocle 3, or PAGA) and whether one of them is redundant or not.


请先 登录 后评论

1 个回答

omicsgene - 生物信息
擅长:重测序,遗传进化,转录组,GWAS


Depending on what mode you run scVelo in (the default stochastic model or the dynamical model) you will get a metric called velocity_pseudotime (stochastic) or latent_time (dynamical) the differences are in how it calculates the splicing model but they're both metrics of cell's position in time like monocle's pseudotime.

The primary conceptual difference in monocle's pseudotime vs velocity's psuedotime/latent time is in the information that is available when computing them. In velocity you incorporate the extra information about introns to compute a model of how cells are processing pre-mRNA into mRNA to determine the cells similarity in their dynamics and by proxy likely transitions, in addition to the changes in gene expression itself. These transitions are then used to fit a model that identify the root cell(s) and endpoint cells.

In standard pseudotime, you only have the "exonic" information, you don't know anything about introns or their splicing dynamics, so you can still determine cell transitions and a cell order from gene expression similarity, but because you don't have a way to model direction of motion using the splicing information it can be difficult to determine which end of the trajectory is the "root" without manually specifying using something like known marker genes.

Assuming both methods work perfectly on perfect data, the results would be ~equivalent, but personally I think the velocity model that incorporates more information is likely to be "better" more often (or at least has the potential to be better more often as the specifics of the computations are refined).



参考:https://www.reddit.com/r/bioinformatics/comments/pug1kv/pseudotime_vs_latent_time/


请先 登录 后评论
  • 1 关注
  • 0 收藏,771 浏览
  • omicsgene 提出于 2024-06-26 07:18

相似问题