This file details all steps of the analysis.

Set-up

Set up working directory structure

cd ~/Downloads
mkdir RNAseq
  # this will be the project directory
cd RNAseq
mkdir data
cd data
mkdir memonet_data AIBS_data

Clone repo

cd ~/Downloads/RNAseq
mkdir memonet_github_repo
cd memonet_github_repo
git clone [email protected]:zhanglab/memonet.git

You can find scripts here: ~/Downloads/RNAseq/memonet_github_repo/memonet/scripts

Download Packages

For the full list for all dependencies by script, see memonet/scripts/README

Cell Ranger v4.0.0: https://support.10xgenomics.com/single-cell-gene-expression/software/pipelines/latest/installation

make note of download location

Seurat v4: https://satijalab.org/seurat/articles/install.html

DESC v2.1.1: https://eleozzr.github.io/desc/installation.html

DESeq2 v1.36.0: https://bioconductor.org/packages/release/bioc/html/DESeq2.html

clusterProfiler v4.2.2: https://bioconductor.org/packages/release/bioc/html/clusterProfiler.html

Download datasets

cd ~/Downloads/RNAseq/data/memonet_data
mkdir data_download

MEMONET dataset:

Location of data: https://docs.google.com/spreadsheets/d/1mU7l8Oj-Fr4FYE6IlmTcX_s2YdCtZXx0GgJI3jBE3J4/edit#gid=0
Script for downloading: download.sh
Download to: ~/Downloads/RNAseq/data/memonet_data/data_download

AIBS dataset:

Location of data: /data/zhanglab/AWS-bkup/MEMONET/data_BICCN/10X-v3_sn_AIBS/
Download to: ~/Downloads/RNAseq/data/AIBS_data
Rename the files to: aibs_barcodes.tsv, aibs_genes.tsv, aibs_matrix.mtx

Pre-processing of memonet data

Aggregate the data of all mice using Cell Ranger commands

Use Cell Ranger's aggr command to combine the data of all mice:

Wd: ~/Downloads/RNAseq/data/memonet_data
Generate a libraries.csv file with the locations of each mouse's data, in the following format:
- the order of rows determines the suffix attached to the barcodes of each mouse (in order to differentiate which mouse is which in the combined file), ie mouse 262 will have -1, 263 -2, etc
- trained mice: 263,277,279 ie -2,-4,-6
- control mice: 262,276,278 ie -1,-3,-5
Run aggr command:
- First run this line on the terminal so that Cell Ranger package can be found, replacing the path with the location of your Cell Ranger download location:
```
export PATH=<path to cellranger download>/cellranger-4.0.0:$PATH
```
- Then run aggr. --none turns off depth normalization, due to the requirement of the DESC clustering package needing unnormalized counts as input. --id is the name of the output file
```
nohup srun -o jobs%j.out -c 20 \cellranger aggr --csv=/data/memonet_data/libraries.csv --none --id=combined_cellranger_no-normalization &
```
- The output directory will be found here: ~/Downloads/RNAseq/data/memonet_data/combined_cellranger_no-normalization/outs/filtered_feature_bc_matrix

Determine QC thresholds

cd ~/Downloads/RNAseq
mkdir QC

Script: QCthresholds.r

Thresholds were chosen based on violin plots of the data before QC

Wd: ~/Downloads/RNAseq/QC

Input:

Data directory for each mouse, ie ~/Downloads/RNAseq/data/memonet_data/data_download/slPsiwmg_JB_262_1_2_3/filtered_feature_bc_matrix/

Output:

ctrl_without_cutoff.png: violin plots of the 3 control mice before QC
QC_ctrl.png: violin plots of the 3 control mice after QC
train_without_cutoff.png
QC_train.png

The number of cells removed with each filter step are given with filterSteps.r

Input:

MEMONET dataset: ~/Downloads/RNAseq/data/memonet_data/combined_cellranger_no-normalization/outs/filtered_feature_bc_matrix/

Output:

Numbers are output on R console

Map MEMONET data onto AIBS dataset ("OA mapping")

cd ~/Downloads/RNAseq
mkdir AIBSmapping
cd AIBSmapping
mkdir OA test
  # 'OA' refers to Our (MEMONET) data mapped to AIBS data
cd OA
mkdir barcode_files count_matrices

Script: mapping_OA.r

Wd: ~/Downloads/RNAseq/AIBSmapping/OA

Query: MEMONET data

Reference: AIBS data

Input:

10X directory of all mice combined: ~/Downloads/RNAseq/data/memonet_data/combined_cellranger_no-normalization/outs/filtered_feature_bc_matrix/
AIBS dataset. The three files here: ~/Downloads/RNAseq/data/AIBS_data

Output:

prediction_scores.csv: lists each cell, the predicted cell type label, and prediction scores for each cell type
umap_AIBS_subclassLabelLegend.png: umap of the reference (AIBS data) with cell type subclass labels
umap_MEMONET_subclassLabelLegend.png: umap of the query (MEMONET data) projected onto AIBS space, labeled with the predicted labels

Test the accuracy of mapping on AIBS data

1. Generate a test dataset (downsample to 25% of each cell type; remove sample cells from rest of reference) and perform label transfer from the remaining 75% of data. Do this 100 times.

Query: 25% of AIBS data

Reference: remaining 75% of AIBS data

Script: mapping_test1.r

Wd: ~/Downloads/RNAseq/AIBSmapping/test

Input:

AIBS dataset. The three files here: RNAseq/data/AIBS_data

Output:

100 prediction_scores_*.csv files

2. Combine the 100 prediction score files into one file

Script: mapping_test2.r

Wd: ~/Downloads/RNAseq/AIBSmapping/test

Input:

100 prediction_scores_*.csv files

Output:

prediction_cutoff.csv

3. Summarize results: calculate false classification percentage, generate confusion matrix, compare mean and median scores of the test and OA mapping

Script: mapping_test3.r

Wd: ~/Downloads/RNAseq/AIBSmapping/test

Input:

~/Downloads/RNAseq/AIBSmapping/test/prediction_cutoff.csv
~/Downloads/RNAseq/AIBSmapping/OA/prediction_scores.csv

Output:

AIBStest_confusionMtx.png: heatmap of original cell type labels vs predicted labels for AIBS testing
maxPredictionScores-AIBStest_and_OA.csv: table of mean and median prediction.score.max for AIBS testing and OA mapping; shows calculation for all celltypes and subset for L2/3 cells

Determine a L2/3 mapping reliability cutoff

Each MEMONET cell gets a prediction score for each cell type in the AIBS dataset; all scores add up to 1. The cell type with the largest score becomes that cell's predicted cell type. To ensure we analyze only L2/3 neurons, we want to remove cells with low reliability for being labeled as L2/3. Reliability is calculated as the sum of prediction scores for L2/3 IT_1, L2/3 IT_2, and L2/3 IT_3.

Generate lists of barcodes remaining at each cutoff value

Script: prediction_score_cutoff_barcodes.r

This script generates barcode lists for various prediction score cutoff values (0 to 0.5). Wd: ~/Downloads/RNAseq/AIBSmapping/OA/barcode_files/

Input:

~/Downloads/RNAseq/AIBSmapping/OA/prediction_scores.csv

Output:

One file for each cutoff, listing the cell barcodes remaining: L23barcodes-fromAIBS_*.csv

Calculate normalized counts for the L2/3 cells

Script: DESCnormalization.r

Wd: ~/Downloads/RNAseq/AIBSmapping/OA/count_matrices

Set the 'cutoff' variable to '0' so that all cells predicted to be L2/3 are kept for normalization.

Input:

Barcode file for all predicted L2/3 cells: ~/Downloads/memonet/prediction_score_cutoff_barcodes/L23barcodes-fromAIBS_0.csv

Output:

DESCnormalized_counts_L23_0.csv

Determine proper cutoff based on marker gene expression

Script: celltype_contaminants.r

This script calculates the mean expression of L2/3 marker genes (Cux2, Otof, Rtn4rl1, Slc30a3, Cacna2d3) and glial marker (Mertk) for the cells that are removed at each cutoff. We want to choose a cutoff that removes cells with low values of L2/3 markers but high values of Mertk, to ensure that mostly glial cells are removed.

Wd: ~/Downloads/RNAseq/AIBSmapping/OA/barcode_files/

Input:

Barcode files for each cutoff: ~/Downloads/RNAseq/AIBSmapping/OA/barcode_files/L23barcodes-fromAIBS_*.csv
Normalized expression: ~/Downloads/RNAseq/AIBSmapping/OA/count_matrices/DESCnormalized_counts_L23_0.csv

Output:

markerGeneExp_predictionCutoffs.csv: table listing the average expression of each marker gene for each cutoff.

We chose a cutoff value of 0.3, as it has the highest mean expression of Mertk while having low expression of L2/3 markers. Any cell that was predicted to be L2/3 and has a sum of prediction scores for L2/3 IT_1, L2/3 IT_2, L2/3 IT_3 <= 0.3 will not be included in downstream analysis.

Investigate cell type proportions

Script: dataset_descriptions.r, Part A

Functions:

calculate cell subclass proportion (neurons and glia)
calculate inhibitory neuron percentage
calculate neuron subclass proportion (neurons only)
plot neuron subclass proportion, comparing MEMONET dataset and AIBS

Input:

AIBS metadata: ~/Downloads/RNAseq/data/AIBS_data/aibs_barcodes.tsv
cell type predictions for MEMONET data: ~/Downloads/RNAseq/AIBSmapping/OA/prediction_scores.csv

Output:

plot of neuronal subclass proportions: ~/Downloads/RNAseq/AIBSmapping/OA/CelltypePropPerDataset.png

L2/3 subtypes: Is there train/control enrichment?

Script: statistics.r

Wd: ~/Downloads/RNAseq/AIBSmapping/OA/

Input:

Choose an input file at the beginning of script. For this part calculating L2/3 proportions, use ~/Downloads/RNAseq/AIBSmapping/OA/prediction_scores.csv
L2/3 barcode list: ~/Downloads/RNAseq/AIBSmapping/OA/barcode_files/L23barcodes-fromAIBS_0.3.csv

Output:

L23subclass_tr_ctrl_prop_hist.png
'summary' variable lists p-values

Cluster L2/3 neurons

1. Normalize counts

Script: DESCnormalization.r

Set 'cutoff' variable to '0.3'

Input:

MEMONET data: ~/Downloads/RNAseq/data/memonet_data/combined_cellranger_no-normalization/outs/filtered_feature_bc_matrix

Output:

~/Downloads/RNAseq/QC/cells_after_QC.csv: list of cells remaining after QC steps on whole dataset. Needed for DE analysis of clusters
~/Downloads/RNAseq/QC/genes_after_QC.csv: list of genes remaining after QC steps. Needed for the background gene list for GO analyses
~/Downloads/RNAseq/AIBSmapping/OA/count_matrices/unnormalized_counts_L23_0.3.csv: unnormalized counts of L2/3 cells, 0.3 cutoff
~/Downloads/RNAseq/AIBSmapping/OA/count_matrices/DESCnormalized_counts_L23_0.3.csv: normalized counts of L2/3 cells, 0.3 cutoff. This will be used as input for classifier training
~/Downloads/RNAseq/AIBSmapping/OA/count_matrices/sampleIDs_0.3.csv: list of cells and mouse ID, for reference during classifier training
~/Downloads/RNAseq/AIBSmapping/OA/count_matrices/sampleIDs_0.3_transpose.csv: same as above but the transposed version

2. Train logistic classifier

Scripts related to the logistic classifier (courtesy of Dr. Nathan Hedrick):

PlotLogisticFunction.m

Plot the logistic function

ShuffledLogisticClassifier.m

Test the classifier accuracy when cell labels ('train' or 'control') have been shuffled 100x

TopGeneOptimization.m

Makes the prediction accuracy optimization curve by looping over different numbers of top genes (sorted by their logistic regression indices)
Input: memonet/downloads/LogisticRegressionIndices.xlsx
- This file lists the logistic regression weights for each gene, where indices match the order of genes in ~/Downloads/RNAseq/AIBSmapping/OA/count_matrices/DESCnormalized_counts_L23_0.3.csv

Classifier_byAnimal.m

Calculates the prediction accuracy of the logistic classifier for each animal

3. Subset to EDGs

cd ~/Downloads/RNAseq/
mkdir cluster_by_genes
cd cluster_by_genes
mkdir 0.3cutoff

First, download the EDG list

Location on github: memonet/downloads/PredictionGenesDescending0.3.csv

This file contains all genes in the dataset, ranked by their weight in the linear classifier. We will subset to the top 3000 as the EDGs

Download to: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff

Then, subset the normalized counts to the EDGs

Script: DESC_inputFormat.r

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff

Input:

Ranked gene list: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/PredictionGenesDescending0.3.csv
L2/3 normalized counts: ~/Downloads/RNAseq/AIBSmapping/OA/count_matrices/DESCnormalized_counts_L23_0.3.csv

Output: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/L23_0.3_EDGmtx.csv

4. Determine parameter settings for running DESC

We varied the n_neighbors (n) and louvain_resolution (L) parameters and calculated silhouette scores of the resulting clusters in order to find the best parameters for clustering our data.

Cluster with various parameter settings and calculate silhouette scores

cd ~/Downloads/RNAseq/cluster_by_genes/
mkdir DESC_parameter_test

Script: DESC_EDGclustering_loop.py

This script loops through various parameter combinations

Wd: ~/Downloads/RNAseq/cluster_by_genes/DESC_parameter_test/

Input:

Normalized L2/3 counts: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/L23_0.3_EDGmtx.csv

Output: n represents the value set for n_neighbors; L represents the value set for louvain_resolution

cluster_n*.L*.csv
umap_n*.L*.csv
sil_scores_n*.L*.csv

Summarize the silhouette scores of each run

Calculate average silhouette score for each clustering result, and visualize scores with respect to the n and L parameters.

Wd: ~/Downloads/RNAseq/cluster_by_genes/DESC_parameter_test/

Script: DESCparameterSummary.r

Input:

Silhouette scores for each run: ~/Downloads/RNAseq/cluster_by_genes/DESC_parameter_test/sil_scores_n*.L*.csv

Output:

parameter_silhouette_score_avgs.csv: summary of the silhouette score and number of clusters produced with each parameter setting combination
heatmap_silScore.png: a heatmap showing the average silhouette score at each combination of n and L parameters

Most scores are very good. A score of -1 indicates wrong clustering, +1 indicates correct clustering, and 0 indicates overlapping clusters.

Run a smoothing algorithm to find the peak

Script: sil_score_smooth.py

Courtesy of Dr. EunJung Hwang

Wd: ~/Downloads/RNAseq/cluster_by_genes/DESC_parameter_test/

Input:

Silhouette score summary: ~/Downloads/RNAseq/cluster_by_genes/DESC_parameter_test/parameter_silhouette_score_avgs.csv

Output:

Smooth_Silhouette.png: left plot is a heatmap of silhouette scores by n and L parameters (similar to heatmap_silScore.png); right plot is a smoothed version, with red x indicating the peak

Running a smoothing algorithm on the heatmap of silhouette scores by n and L parameters reveals n=25 and L=0.65 to be at the peak of scores. Therefore, we chose n=25 and L=0.65.

5. Run DESC clustering

cd ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff
mkdir DESC

Script: DESC_EDGclustering.py

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC

Parameters are set to:

n_neighbors = n = 25
louvain_resolution = L = 0.65

Input:

Normalized expression matrix, subset to EDGs: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/L23_0.3_EDGmtx.csv

Output:

clusters_n25.L0.65.csv: lists each cell barcode and the cluster it is assigned
umap_n25.L0.65.csv: umap x and y coordinates; row indices correspond to the barcode order in clusters_n25.L0.65.csv
tsne_n25.L0.65.csv: tsne x and y coordinates
desc_result.n25.L0.65.h5ad: AnnData object containing the information produced by clustering, including: cluster assignments, umap and tsne coordinates, etc. This will be used for trajectory analysis later
result_DESC.n25.L0.65/: directory for encoder weights and model info
figures/
- umap0.65desc.n25.L0.65.png: visual of cluster umap projection

6. Visualize cluster train/control proportion

Script: classifier_umap_plot.r

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/figures

Input:

Cluster file: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/clusters_n25.L0.65.csv
Umap coordinate file: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/umap_n25.L0.65.csv

Output: classifier_umap_tr_ctrl.svg

Is there train/control enrichment?

Script: statistics.r

Plot train/control proportions across cluster and calculate p-values for significance

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/figures

Input:

Choose an input file at the beginning of script. For this part calculating cluster train/control proportions, use ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/clusters_n25.L0.65.csv

Output:

cluster_tr_ctrl_prop_hist.svg: bar plot of train/control proportion across cluster
ClusterMouseProp_mouseDots.csv: file containing data for above plot, plus proportion of each mouse across clusters
'summary' variable lists p-values

Script: ClusterMouseProp.py

Plot train/control proportions across cluster and have individual mouse proportions overlaid

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/figures

Input:

ClusterMouseProp_mouseDots.csv

Output:

ClusterMouseProp_mouseDots.svg: bar plot of train/control proportion across cluster, with dots for each mouse overlaid

How many cells are in each cluster?

Script: dataset_descriptions.r, Part D

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/figures

Input:

Clusters file: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/clusters_n25.L0.65.csv

Output:

cluster_percentage_tbl.csv

GO analysis of the 3000 EDGs

What biological processes are supported by the 3000 EDGs?

cd ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff
mkdir GO
cd GO
mkdir EDG clusters

Script: GOvisualization.r, part A; see GoTermAnalysis.m (courtesy of Dr. Nathan Hedrick) for generating manuscript GO term figures

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/GO/EDG

Input:

Ranked gene list (containing the EDGs): ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/PredictionGenesDescending0.3.csv
Background gene list: ~/Downloads/RNAseq/QC/genes_after_QC.csv. The background gene list represents all genes in the dataset.

Output:

EDG_GO.csv: GO results for the 3000 EDGs
EDG_GO.png: treeplot showing GO results for the 3000 EDGs

GO analysis of the unique EDGs

What biological processes are supported by the unique EDGs (those that do not appear in the DE results of L2/3 train vs control)?

Script: GOvisualization.r, part A-2

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/GO/EDG

Input:

Ranked gene list (containing the EDGs): ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/PredictionGenesDescending0.3.csv
Background gene list: ~/Downloads/RNAseq/QC/genes_after_QC.csv. The background gene list represents all genes in the dataset.
DEGs from L2/3 tr vs ctrl: ~/Downloads/RNAseq/AIBSmapping/OA/DESeq2/L23_0.3_tr_vs_ctrl/train_vs_control_sig_genes.csv

Output:

L2/3 tr vs ctrl DEG results with an extra column specifying whether the DEG is also an EDG: ~/Downloads/RNAseq/AIBSmapping/OA/DESeq2/L23_0.3_tr_vs_ctrl/train_vs_control_sig_genes_EDGcolumn.csv
GO results for the unique EDGs: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/GO/EDG/uniqueEDG_GO.csv

DE analysis for neuron subtypes

1. Generate barcode files for various subtypes

Script: dataset_descriptions.r, Part B

Input: data_neurons variable from Part A

Output:

~/Downloads/RNAseq/AIBSmapping/OA/barcode_files/glut_barcodes.csv
~/Downloads/RNAseq/AIBSmapping/OA/barcode_files/GABA_barcodes.csv
~/Downloads/RNAseq/AIBSmapping/OA/barcode_files/L5_barcodes.csv
~/Downloads/RNAseq/AIBSmapping/OA/barcode_files/pvalb_barcodes.csv
~/Downloads/RNAseq/AIBSmapping/OA/barcode_files/sst_barcodes.csv
~/Downloads/RNAseq/AIBSmapping/OA/barcode_files/vip_barcodes.csv

2. Run DESeq2, train vs control

cd ~/Downloads/RNAseq/AIBSmapping/OA
mkdir DESeq2
cd DESeq2
mkdir glut_tr_vs_ctrl GABA_tr_vs_ctrl L5_tr_vs_ctrl pvalb_tr_vs_ctrl sst_tr_vs_ctrl vip_tr_vs_ctrl

Script: DESeq2_tr_vs_ctrl.r

DE analysis: Run the script in each subtype folder

Wd: ~/Downloads/RNAseq/AIBSmapping/OA/DESeq2/_tr_vs_ctrl

i.e. L5_tr_vs_ctrl

Input: ~/Downloads/RNAseq/AIBSmapping/OA/barcode_files/_barcodes.csv

i.e. L5_barcodes.csv

Output:

unnormalized_counts_from_dds.csv: unnormalized gene expression
normalized_sizeFactors_calculateSumFactors.csv: size factors that generate the normalized data
normalized_counts_from_dds.csv: normalized gene expression
train_vs_control_all_genes.csv: DESeq2 results for all genes
train_vs_control_sig_genes.csv: DESeq2 results for significant genes (padj <0.05)

Rename the train_vs_control_sig_genes.csv file with the neuron subtype appended to the end like so: train_vs_control_sig_genes_pvalb.csv, train_vs_control_sig_genes_sst.csv, etc

DE analysis for L2/3 neurons

cd ~/Downloads/RNAseq/AIBSmapping/OA/DESeq2
mkdir L23_0.3_tr_vs_ctrl

1. Run DESeq2: train vs control

Script: DESeq2_tr_vs_ctrl.r

Wd: ~/Downloads/RNAseq/AIBSmapping/OA/DESeq2/L23_0.3_tr_vs_ctrl

Input: L2/3 barcode list: ~/Downloads/RNAseq/AIBSmapping/OA/barcode_files/L23barcodes-fromAIBS_0.3.csv

Output:

unnormalized_counts_from_dds.csv: unnormalized gene expression
normalized_sizeFactors_calculateSumFactors.csv: size factors that generate the normalized data
normalized_counts_from_dds.csv: normalized gene expression
train_vs_control_all_genes.csv: DESeq2 results for all genes
train_vs_control_sig_genes.csv: DESeq2 results for significant genes (padj <0.05)

2. Summarize DE results and compare across neuron subtypes: what IEGs are significant?

Script: IEGcomparison.r

Input:

DE results for each neuron type -L2/3: ~/Downloads/RNAseq/AIBSmapping/OA/DESeq2/L23_0.3_tr_vs_ctrl/train_vs_control_sig_genes.csv -L5: ~/Downloads/RNAseq/AIBSmapping/OA/DESeq2/L5_tr_vs_ctrl/train_vs_control_sig_genes_L5.csv -Pvalb: ~/Downloads/RNAseq/AIBSmapping/OA/DESeq2/pvalb_tr_vs_ctrl/train_vs_control_sig_genes_pvalb.csv -Sst: ~/Downloads/RNAseq/AIBSmapping/OA/DESeq2/sst_tr_vs_ctrl/train_vs_control_sig_genes_sst.csv -Vip: ~/Downloads/RNAseq/AIBSmapping/OA/DESeq2/vip_tr_vs_ctrl/train_vs_control_sig_genes_vip.csv

Output:

Table listing the IEGs significantly up- or down-regulated in each neuron type: ~/Downloads/RNAseq/AIBSmapping/OA/DESeq2/IEGcomparison_neuronTypes2.csv

3. How many DEGs overlap with the 3000 experience-dependent genes (EDGs) used for clustering?

Script: dataset_descriptions.r, part C

Input:

Ranked gene list: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/PredictionGenesDescending0.3.csv
L2/3 DEG list: ~/Downloads/RNAseq/AIBSmapping/OA/DESeq2/L23_0.3_tr_vs_ctrl/train_vs_control_sig_genes.csv

Output: The length of DEGoverlap$gene will tell you how many genes are shared between the two gene sets

DE analysis of clusters

1. Run DESeq2, one cluster vs the others

cd ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC
mkdir DESeq2
cd DESeq2
mkdir n25.L0.65

Script: DESeq2_clusterX_vs_others.r

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65

Input:

Cluster file: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/clusters_n25.L0.65.csv
unnormalized counts file: ~/Downloads/RNAseq/AIBSmapping/OA/count_matrices/unnormalized_counts_L23_0.3.csv

Output directory: all_cells/

unnormalized_counts_from_dds.csv: unnormalized gene expression
normalized_sizeFactors_calculateSumFactors.csv: size factors that generate the normalized data
normalized_counts_from_dds.csv: normalized gene expression
*_vs_others_all_genes.csv: DESeq2 results for all genes
*_vs_others_sig_genes.csv: DESeq2 results for significant genes (padj <0.05)

Combine cluster result files into one file:

Script: DESeq2-table.r

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65/all_cells

Input: files ending in *_all_genes.csv

Output:

DEGstats_allGenes.csv: all genes
DEGstats_padj0.05.csv: significant genes (padj <0.05)

2. Summarize DE results: what IEGs are significant?

DEGvisuals.r, Part A

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65/all_cells/

Input:

DE results: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65/all_cells/DEGstats_allGenes.csv
DESeq2 normalized counts: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65/all_cells/normalized_counts_from_dds.csv
Cluster file: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/clusters_n25.L0.65.csv

Output:

IEG_lineplot_zscore.png: line plot indicating the average z-scored expression value of IEGs across clusters. DE significance is indicated by open or closed dot.

Map AIBS L2/3 cells onto MEMONET clusters and annotate by cluster

cd ~/Downloads/RNAseq/AIBSmapping
mkdir AO
  # 'AO' stands for AIBS onto Our (MEMONET) cells
cd AO
mkdir n25.L0.65

Script: mapping_AO.r

Wd: ~/Downloads/RNAseq/AIBSmapping/AO/n25.L0.65

Input:

MEMONET data: ~/Downloads/RNAseq/data/memonet_data/combined_cellranger_no-normalization/outs/filtered_feature_bc_matrix/
AIBS dataset. The three files here: ~/Downloads/RNAseq/data/AIBS_data
Cluster file: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/clusters_n25.L0.65.csv
Umap coordinate file: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/umap_n25.L0.65.csv

Output:

prediction_scores.csv: prediction score for the AIBS cells onto the clusters
Images:
- umap_MEMONETL23.png: MEMONET cells visualized in the DESC-generated umap space
- umap_AIBSL23.png: AIBS cells placed in classifier space, colored by their predicted cluster

What proportion of AIBS cells map to each cluster? Script: dataset_description.r, Part E

Wd: ~/Downloads/RNAseq/AIBSmapping/AO/n25.L0.65

Input: ~/Downloads/RNAseq/AIBSmapping/AO/n25.L0.65/prediction_scores.csv

Output:

AIBSpie.svg: pie chart of proportions
AIBSmap_table: summary table
AIBS_L23_mappedToClusters.csv: counts of AIBS L2/3 subtypes mapped to C0-5 (Table 6)

DE analysis of clusters

1. Run DESeq2, one cluster vs baseline

Script: DESeq2_clusterX_vs_clusterY.r

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65/

Input:

Cluster file: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/clusters_n25.L0.65.csv

Set 'ref' variable to '0' to indicate cluster 0 as the baseline comparison

Output directory: x_vs_0/all_cells/

unnormalized_counts_from_dds.csv: unnormalized gene expression
normalized_sizeFactors_calculateSumFactors.csv: size factors that generate the normalized data
normalized_counts_from_dds.csv: normalized gene expression
x_vs_0_all_genes.csv: DESeq2 results for all genes
x_vs_0_sig_genes.csv: DESeq2 results for significant genes (padj <0.05)

Combine cluster result files into one file: Script: DESeq2-table.r

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65/x_vs_0/all_cells/

Input: files ending in *_all_genes.csv

Output:

DEGstats_allGenes.csv: all genes
DEGstats_padj0.05.csv: significant genes (padj <0.05)

2. Summarize DE results: what IEGs are significant?

DEGvisuals.r, Part B

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65/x_vs_0/all_cells/

Input:

DE results: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65/x_vs_0/all_cells/DEGstats_allGenes.csv
DESeq2 normalized counts: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65/x_vs_0/all_cells/normalized_counts_from_dds.csv
Cluster file: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/clusters_n25.L0.65.csv

Output:

IEG_lineplot_refC0.png: line plot indicating the LFC value of IEGs across clusters. DE significance is indicated by open or closed dot. LFC is a metric used by DESeq2 to assess how much a gene’s expression has changed between the two comparisons. A positive value indicates upregulation in comparison to C0 while negative indicates downregulation

Gene Ontology analysis of clusters

Script: GOvisualization.r, Part B; see GoTermAnalysis.m for generating manuscript GO term figures

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/GO/clusters/n25.L0.65/x_vs_0

Set 'regulation' variable to 'up' in order to analyze the upregulated genes only

Input:

Background gene list: ~/Downloads/RNAseq/QC/genes_after_QC.csv
DE results from x vs 0 scenario: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65/x_vs_0/all_cells/DEGstats_padj0.05.csv

Output:

GOresults_n25.L0.65_Bio-upregDEG.csv: csv file of results for all clusters
- up_treeBio_n25.L0.65.png: tree plot showing the top 30 most significant GO terms; one tree plot for each cluster

Investigate C4 contamination

C4's tree plot shows terms characterizing other cell types. Perhaps this cluster contains contaminants that were not filtered out by the L2/3 prediction cutoff threshold.

Script: celltype_contaminants.r, Part B

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/

Input:

Normalized counts: ~/Downloads/RNAseq/AIBSmapping/OA/count_matrices/DESCnormalized_counts_L23_0.3.csv
Clusters: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/clusters_n25.L0.65.csv

Output:

L23_glial_marker_exp.csv

Visualize select GO terms in the clusters

Script: GOvisualization.r, Part C.1; see GoTermAnalysis.m for generating manuscript GO term figures

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/GO/clusters/n25.L0.65/x_vs_0

Input:

GO results: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/GO/clusters/n25.L0.65/x_vs_0/GOresults_n25.L0.65_Bio-upregDEG.csv
DESeq2-normalized counts: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65/all_cells/normalized_counts_from_dds.csv
Clusters: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/clusters_n25.L0.65.csv

Output:

GO_line.png: line plot showing the average z-scored gene expressions in each cluster per GO term
GO_heatmap.png: heatmap for each GO term showing the z-scored expression value of each gene in the GO term per cluster

Visualize processes unique to C5

Script: GOvisualization.r, Part C.2; see GoTermAnalysis.m for generating manuscript GO term figures

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/GO/clusters/n25.L0.65/x_vs_0

Input:

GO results: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/GO/clusters/n25.L0.65/x_vs_0/GOresults_n25.L0.65_Bio-upregDEG.csv
DESeq2-normalized counts: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65/all_cells/normalized_counts_from_dds.csv
Clusters: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/clusters_n25.L0.65.csv

Output:

C5_GOtermLinePlot_zscore.png: line plot showing the average z-scored gene expressions in each cluster per GO term
C5_GOtermHeatmap.png: heatmap for each GO term showing the z-scored expression value of each gene in the GO term per cluster

Note: Glycolysis and NE (norepinephrine) receptors are not GO terms, we just visualize in the same way as GO terms to show the trends.

Reactivation Score

First download the reference reactivation gene list

Location on github: memonet/downloads/DG_React_NotReact.csv

This file is from Jaeger et al. (2018) Supplementary table 3, tab ‘DG_React_NotReact’

Download to: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65/all_cells/

Calculate reactivation score

Script: reactivation_score.r

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65/all_cells/

Input:

DE results of each cluster vs the others: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65/all_cells/DEGstats_padj0.05.csv
Reactivation gene list: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/DESeq2/n25.L0.65/all_cells/DG_React_NotReact.csv

Output:

ReactivationScores.csv
reactivation_line.png: line plot indicating the reactivation score of each cluster
reactivation_matches.png: bar plot indicating the number of DEGs in each cluster that either match or don't match the sign of the reference list

Trajectory Analysis

Script: Trajectory_Analysis.ipynb

Courtesy of Dr. Rudiyanto Gunawan

Use this Jupyter Notebook script to run trajectory analysis and visualize plots.

Wd: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/

Input:

AnnData object containing clustering results: ~/Downloads/RNAseq/cluster_by_genes/0.3cutoff/DESC/desc_result.n25.L0.65.h5ad

Output:

Plots are visualized within the Jupyter Notebook

zhanglab / memonet Goto Github PK

memonet's Introduction

Set-up

Set up working directory structure

Clone repo

Download Packages

Download datasets

Pre-processing of memonet data

Aggregate the data of all mice using Cell Ranger commands

Determine QC thresholds

Map MEMONET data onto AIBS dataset ("OA mapping")

Test the accuracy of mapping on AIBS data

1. Generate a test dataset (downsample to 25% of each cell type; remove sample cells from rest of reference) and perform label transfer from the remaining 75% of data. Do this 100 times.

2. Combine the 100 prediction score files into one file

3. Summarize results: calculate false classification percentage, generate confusion matrix, compare mean and median scores of the test and OA mapping

Determine a L2/3 mapping reliability cutoff

Generate lists of barcodes remaining at each cutoff value

Calculate normalized counts for the L2/3 cells

Determine proper cutoff based on marker gene expression

Investigate cell type proportions

L2/3 subtypes: Is there train/control enrichment?

Cluster L2/3 neurons

1. Normalize counts

2. Train logistic classifier

Scripts related to the logistic classifier (courtesy of Dr. Nathan Hedrick):

3. Subset to EDGs

First, download the EDG list

Then, subset the normalized counts to the EDGs

4. Determine parameter settings for running DESC

Cluster with various parameter settings and calculate silhouette scores

Summarize the silhouette scores of each run

Run a smoothing algorithm to find the peak

5. Run DESC clustering

6. Visualize cluster train/control proportion

Is there train/control enrichment?

How many cells are in each cluster?

GO analysis of the 3000 EDGs

GO analysis of the unique EDGs

DE analysis for neuron subtypes

1. Generate barcode files for various subtypes

2. Run DESeq2, train vs control

DE analysis for L2/3 neurons

1. Run DESeq2: train vs control

2. Summarize DE results and compare across neuron subtypes: what IEGs are significant?

3. How many DEGs overlap with the 3000 experience-dependent genes (EDGs) used for clustering?

DE analysis of clusters

1. Run DESeq2, one cluster vs the others

2. Summarize DE results: what IEGs are significant?

Map AIBS L2/3 cells onto MEMONET clusters and annotate by cluster

DE analysis of clusters

1. Run DESeq2, one cluster vs baseline

2. Summarize DE results: what IEGs are significant?

Gene Ontology analysis of clusters

Investigate C4 contamination

Visualize select GO terms in the clusters

Visualize processes unique to C5

Reactivation Score

First download the reference reactivation gene list

Calculate reactivation score

Trajectory Analysis

memonet's People

Contributors

Watchers

Recommend Projects

Recommend Topics

Recommend Org