2021.02.14
今天收集了三個(gè)與譜系分化有關(guān)的單細(xì)胞高級(jí)算法:MAGIC(2018 Cell),PHATE(2019 NBT),Palantir(2019 NBT)。
它們與傳統(tǒng)的偽時(shí)序分析方法(pseudotime analysis)略有所不同,用到了流形(manifold)學(xué)習(xí)的原理。這里推薦開(kāi)發(fā)前兩個(gè)包的實(shí)驗(yàn)室推出的一個(gè)培訓(xùn),介紹機(jī)器學(xué)習(xí)在單細(xì)胞數(shù)據(jù)處理中的應(yīng)用,適合已經(jīng)熟悉單細(xì)胞基本處理方法并且想要進(jìn)階的選手:https://www.krishnaswamylab.org/workshop;https://github.com/KrishnaswamyLab/SingleCellWorkshop。
另外,今天摸魚(yú)的時(shí)候看到了Broad Institute的一個(gè)單細(xì)胞培訓(xùn),內(nèi)容非常詳細(xì),也一起分享一下:https://broadinstitute.github.io/2019_scWorkshop/index.html。
MAGIC
處理單細(xì)胞dropout
文獻(xiàn)
David van Dijk, et al. Recovering Gene Interactions from Single-Cell Data Using Data Diffusion. 2018. Cell.
https://linkinghub.elsevier.com/retrieve/pii/S0092867418307244
Krishnaswamy Lab
文檔
R版本:https://github.com/KrishnaswamyLab/magic#r
python版本tutorial:https://nbviewer.jupyter.org/github/KrishnaswamyLab/MAGIC/blob/master/python/tutorial_notebooks/bonemarrow_tutorial.ipynb
文獻(xiàn)閱讀筆記
利用流形學(xué)習(xí)還原單細(xì)胞的基因表達(dá)
配合kNN-DREMI:看基因-基因關(guān)系
DREVI plot:基因聚類(lèi)(基于擬時(shí)序)
基于擬時(shí)序預(yù)測(cè)TF的靶基因(changed with EMT and peaked along with or after TF,DREMI > = 1)
選取轉(zhuǎn)錄因子(kNN-DREMI with VIM is >0.5)
推薦的DEG方法:
earth-mover distance (EMD) used in the archetype analysis
We recommend running diffusion map analysis directly on the raw data (otherwise this could lead to over smoothing). On the other hand, MAGIC imputed data are well-suited to visualize trends along the diffusion components.

PHATE
降維展示的方法,適合發(fā)育分化
與MAGIC同一個(gè)實(shí)驗(yàn)室開(kāi)發(fā),適合作為MAGIC的下游使用
文獻(xiàn)
Visualizing structure and transitions in high-dimensional biological data. 2019 NBT
https://www.nature.com/articles/s41587-019-0336-3
Krishnaswamy Lab
相關(guān)資料
推文教程:https://mp.weixin.qq.com/s/JJQfKul1uvO8XGdTE4mPeA
Github:https://github.com/KrishnaswamyLab/PHATE/
文檔:https://phate.readthedocs.io/en/stable/tutorial.html#
python版本tutorial:https://nbviewer.jupyter.org/github/KrishnaswamyLab/PHATE/blob/master/Python/tutorial/EmbryoidBody.ipynb

Palantir
一種新的譜系分化推斷和展示方法
文獻(xiàn)
文獻(xiàn):Characterization of cell fate probabilities in single-cell data with Palantir. 2019 NBT.
https://www.nature.com/articles/s41587-019-0068-4
Dana Pe'er Lab
相關(guān)資料
推文教程:https://mp.weixin.qq.com/s/SfOfw0CRujw2KPvVTxl3Fw (其中調(diào)用到了MAGIC)
Github: https://github.com/dpeerlab/Palantir/
Python版本tutorial:https://nbviewer.jupyter.org/github/dpeerlab/Palantir/blob/master/notebooks/Palantir_sample_notebook.ipynb