hello，大家好， 今天我們來(lái)分享一個(gè)新的內(nèi)容，利用我們的軌跡分析的結(jié)果來(lái)推斷細(xì)胞之間的通訊交流。文章在Inferring cell-cell interactions from pseudotime ordering of scRNA-Seq data.這個(gè)方法對(duì)于有分化關(guān)系的細(xì)胞類型來(lái)說(shuō)，是一個(gè)新的角度解讀細(xì)胞間的通訊，個(gè)人認(rèn)為更加的重要。我們先來(lái)看看文獻(xiàn)，最后分享示例代碼。

Abstract

1、A major advantage of single cell RNA-Sequencing (scRNA-Seq) data is the ability to reconstruct continuous ordering and trajectories for cells。

2、To date, such ordering was mainly used to group cells and to infer interactions within cells.

3、prior methods that only focus on the average expression levels of genes in clusters or cell types, TraSig fully utilizes the dynamic information to identify significant ligand-receptor pairs with similar trajectories, which in turn are used to score interacting cell clusters（這一句是精髓,細(xì)胞分化接近的細(xì)胞類型通訊才是最重要的）。

Introduction

1、單細(xì)胞軌跡分析mainly focused on the expression similarity between cells in the same cluster or at consecutive time points and on the differences in transcriptional regulation between cell types and over time。

2、單細(xì)胞細(xì)胞通訊通常是識(shí)別ligands in one of the clusters or cell types and corresponding receptors in another cluster and then infer interactions based on the average expression of these ligand-receptor pairs。（cellphoneDB，SingleCellSingleR等軟件）。While successful, most current methods for inferring cell-cell interactions from scRNA-Seq data only use of the average expression levels of ligands and receptors in the two clusters or cell types they test（這個(gè)地方局限性很大）。

3、目前的通訊分析方法（cluster的平均值）While this may be fine for steady state populations，(for example, different cell types in adult tissues)，for studies that focus on development or response modeling, such averages do not take full advantage of the available data in scRNA-Seq studies 。

4、軌跡分析的結(jié)果中，cells on the same branch (or cluster) cannot be assumed to be homogeneous with respect to the expression of key genes. Using average analysis for such clusters may lead to inaccurate predictions about the relationship between ligands and receptors in two different (though parallel in terms of timing) branches.

5、（下圖）While the average expression of a ligand and receptor in two different branches are the same,the first two cases are unlikely to strongly support an interaction between these two cell types while the third and fourth, where both are either increasing or decreasing in their respective ordering, are much more likely to hint at real interactions between the groups。（這是精髓）。

圖片.png

In other words, if two groups of cells are interacting, then we expect to see the genes，encoding signaling molecules in these groups co-express at a similar pace along the pseudotime.（很有道理）。所以在軌跡分析的結(jié)果上進(jìn)行通訊分析，做好的方法就是sliding window approach.（滑動(dòng)窗口法 ）。

6、TraSig利用軌跡分析進(jìn)行細(xì)胞通訊分析的方法，extract expression patterns for ligands and receptors in different edges of the trajectory using a sliding window approach. It then uses these profiles to score temporal interactions between ligand and their known receptors in different edges corresponding to the same time.檢驗(yàn)還是置換檢驗(yàn)。

Result

TraSig workflow. Top Left: For a time series scRNA-seq dataset, we use the reconstructed pseudotime, trajectory and the expression data as inputs. Bottom Left: We next determine expression profiles for genes along each of the edges (clusters) using sliding windows and compute dot product scores for pairs of genes in edges. Right: Finally, we use permutation tests to assign significance levels to the scores we computed.

圖片.png

看看示例結(jié)果，利用CSHMM（隱式馬爾科夫模型）構(gòu)建細(xì)胞類型之間的發(fā)育軌跡，關(guān)于CSHMM，大家可以參考一文搞懂HMM（隱馬爾可夫模型），以及我之前分享的文章10X單細(xì)胞（10X空間轉(zhuǎn)錄組）基礎(chǔ)算法之KL散度。

圖片.png

然后是Inferring cell type interactions for liver development。就是上面我們所說(shuō)的滑動(dòng)窗口法。

圖片.png

Results from comparing TraSig with SingleCellSignalR and CellPhoneDB. Top: Heatmaps for scores assigned by the three different methods for all cluster pairs representing cells sampled at the same time. TraSig and SingleCellSignalR identified more ligand-receptors pairs leading to higher scores. Bottom left: -log10 p-value for enriched GO terms related to endothelial cells and vascular development. Bottom right: Venn diagrams for the overlap in identified ligands and receptors among the three methods. The overlap between TraSig and SingleCellSignalR is high though roughly 50% of the identified proteins by each method are not identified by the other.這里的結(jié)果展示的是方法上的差別。

TraSig identifies ligand-receptor interactions important to vascular development，其實(shí)按照這個(gè)方法更加準(zhǔn)確的得到了細(xì)胞在分化過(guò)程中的準(zhǔn)確通訊。

圖片.png

圖注，Ligand-receptor interaction predictions from TraSig of interest for functional studies. (a) Cartoon of cell signaling interaction between different DesLO cell types (HLC, hepatocyte-like cells; CLC, cholangiocyte-like cells; SLC, stellate-like cells; ELC, endothelial-like cells) (b) Trajectory plot showing cell type assignments with key identifying genes highlighted by different colors (Red = SOX2+ non induced cells, Yellow = SOX9 cholangiocyte-like cells, Blue = Hepatocyte-like cells, Purple = Stellate-like cells, Green = Endothelial-like cells). (c) Sender CXCL12 cells from the Cholangiocyte and Stellate populations in red shown with the receiver CXCR4 expressing endothelial cell population in blue. (d) Sender and receiver signaling populations (red = senders/ligands; blue = receivers/receptors)。

其實(shí)最有價(jià)值的就是計(jì)算通訊分析的方法是滑動(dòng)窗口法。

最后看看示例代碼

import pickle
import sys
import os
import gc
import requests

import numpy as np
import bottleneck as bn
import pandas as pd

# load packages required for analysis 
import statsmodels.api as sm
import statsmodels as sm
import matplotlib
import matplotlib.pyplot as plt
import seaborn as sns
from scipy import stats

Run TraSig on the example data

main.py -i input -o output -d oligodendrocyte-differentiation-clusters_marques -g None -b ti_slingshot -n 1000 -s smallerWindow
usage: main.py [-h] -i INPUT -o OUTPUT -d PROJECT -g PREPROCESS -b MODELNAME
               [-t LISTTYPE] [-l NLAP] [-m METRIC] [-z NAN2ZERO] [-n NUMPERMS]
               [-p MULTIPROCESS] [-c NCORES] [-s STARTINGTREATMENT]

optional arguments:
  -h, --help            show this help message and exit
  -i INPUT, --input INPUT
                        string, folder to find inputs
  -o OUTPUT, --output OUTPUT
                        string, folder to put outputs
  -d PROJECT, --project PROJECT
                        string, project name
  -g PREPROCESS, --preprocess PREPROCESS
                        string, preprocessing steps applied to the data /
                        project, default None
  -b MODELNAME, --modelName MODELNAME
                        string, name of the trajectory model
  -t LISTTYPE, --listType LISTTYPE
                        string, optional, interaction list type, default
                        ligand_receptor
  -l NLAP, --nLap NLAP  integer, optional, sliding window size, default 20
  -m METRIC, --metric METRIC
                        string, optional, scoring metric, default dot
  -z NAN2ZERO, --nan2zero NAN2ZERO
                        boolean, optional, if treat nan as zero, default True
  -n NUMPERMS, --numPerms NUMPERMS
                        integer, optional, number of permutations, default
                        10000
  -p MULTIPROCESS, --multiProcess MULTIPROCESS
                        boolean, optional, if use multi-processing, default
                        True
  -c NCORES, --ncores NCORES
                        integer, optional, number of cores to use for multi-
                        processing, default 4
  -s STARTINGTREATMENT, --startingTreatment STARTINGTREATMENT
                        string, optional, way to treat values at the beginning
                        of an edge with sliding window size smaller than nLap,
                        None/parent/discard/smallerWindow, default
                        smallerWindow, need to provide an extra input
                        'path_info.pickle' for 'parent' option

Prepare inputs for TraSig (from dynverse outputs)

python prepare_inputs.py -i ../trajectory/input -o ../example/input -d oligodendrocyte-differentiation-clusters_marques -t ../trajectory/output/output.h5 -g None -b ti_slingshot -e None
usage: prepare_inputs.py [-h] -i INPUT -o OUTPUT -d PROJECT -t TRAJECTORYFILE
                         -g PREPROCESS -b MODELNAME [-e OTHERIDENTIFIER]

optional arguments:
  -h, --help            show this help message and exit
  -i INPUT, --input INPUT
                        string, folder to find inputs for trajectory inference
  -o OUTPUT, --output OUTPUT
                        string, folder to save inputs for TraSig
  -d PROJECT, --project PROJECT
                        string, project name
  -t TRAJECTORYFILE, --trajectoryFile TRAJECTORYFILE
                        string, trajectory output file from dynverse, default
                        ../trajectory/output/output.h5
  -g PREPROCESS, --preprocess PREPROCESS
                        string, preprocessing steps applied to the data /
                        project, default None
  -b MODELNAME, --modelName MODELNAME
                        string, name of the trajectory model
  -e OTHERIDENTIFIER, --otherIdentifier OTHERIDENTIFIER
                        string, optional, other identifier for the output,
                        default None

Analyze outputs from TraSig

剩下的大家自己看吧，內(nèi)容在TraSig。研究發(fā)育的童鞋，實(shí)時(shí)通訊，才是最好的分析做法。

生活很好，有你更好