Question about deconvolution analysis in PsychENCODE paper

Q:
I have a question about the deconvolution method used in the flagship PsychENCODE paper Comprehensive functional genomic resource and integrative model for the human brain. I would like to perform a similar analysis on my own bulk samples using the single cell expression profiles used in the paper, however it is unclear how these profiles are formed.

Specifically, supplementary file DER-23 lists the cell type fractions for 24 cell types. These coefficients presumably came from solving the following:

B = C * W

Where B is the marker gene by samples matrix, C is the marker gene by cell type matrix, and W is the appropriate weights matrix. How do I go about obtaining or reproducing the 24 cell type profiles? From what I can tell, these profiles were not released along with the other supplemental data sets.

If you could please answer my question or forward this email on the appropriate author(s), I would appreciate it.

A:
Sorry for the late reply. I think the profiles you want are on resource.psychencode.org

Requesting information about cQTL and fQTL data from PsychENCODE

Q:
I am writing in regards to the datasets posted on PsychENCODE website. I noticed that full summary statistics for QTL maps are posted for eQTLs and isoQTLs, but cQTLS and fQTLs only have top SNP information. Is there a chance you could upload full summary stats for cQTLs and fQTLs as well?

A:
We calculated cQTLs and fQTLs differently from eQTLs and isoQTLs. So we only have top SNP information for cQTLs and fQTLs.

Information about datasets from PsychENCODE

Q1:
I am writing to ask where I might be able to find a list of all of the datasets generated for the PsychENCODE project. Specifically, we would like to know how many single-cell and bulk RNA-seq datasets were generated, and what the sex and age is of the samples used to generate these datasets. I was not able to find this information in the supplementary materials from your 2018 Science paper or on the PsychENCODE website, but perhaps I am missing something. Before we start the application process to access the raw data, it would be very helpful to have this information.

A1:
You should be able to find a list of all datasets used for Wang et al.
(’18) from resource.psychencode.org . Please contact Prashant (copied)

Please refer to http://resource.psychencode.org/Datasets/RawData/RAW-01_PEC_Table_of_Datasets.xlsx

This contains the set of datasets associated with the analysis in the Wang et al paper, focusing on the adult samples. Of course, this is a subset of the total PsychENCODE datasets. I can see if there is a simple resource for you to access to get the information from the superset. I will let you know soon.

I am going to answer your question in two parts. Here is Part 1:

The metadata for the prenatal and adult single-nuclei datasets is available at http://development.psychencode.org/ under the "Processed Data" heading, "Single cell/nucleus RNA-seq". The ages and sexes of the sampled individuals can be found in the .xlsx files with the labels "QC" appended.

Q2:
Can you tell me if additional single-cell RNA-seq datasets will be generated in the next phase of this project and what the timeframe might be?

A2:
There will definitely be a significant expansion of the single-cell/nucleus datasets in the next phase, though it is as yet uncertain as to how long that would take. I am hesitant to take a guess right now, but please check back in a couple of months and we may have a better answer.

psych encode derived data types

Q1:
I am looking at
http://resource.psychencode.org/#Derived
under
Derived Data Types
there are a couple of gene expression matrices. What are the columns (samples), ie which ones are which cases and which controls in the header file:
http://resource.psychencode.org/Datasets/Derived/Header_DER-01_PEC_Gene_expression_matrix_normalized.txt

What is the difference between
DER-01_PEC_Gene_expression_matrix_normalized
and
DER-02_PEC_Gene_expression_matrix_TPM
besides the fact that one has 43,886 lines and the other has 57,821 lines and that one has 1,932 columns and the other 1,867 columns.

A1:
1) "What are the columns (samples), ie which ones are which cases and which controls in the header file"

I am unable to pass on this information, as our DCC mentioned that diagnosis information would only be available upon application to Synapse for approval by the NIMH and investigators. Please contact them for access approval.

2)
"What is the difference between
DER-01_PEC_Gene_expression_matrix_normalized
and
DER-02_PEC_Gene_expression_matrix_TPM"

The difference in the numbers is simply between FPKM and TPM units in expression.
3)
"besides the fact that one has 43,886 lines and the other has 57,821 lines"
There is very likely a difference in the thresholding of gene expression applied to these two datasets. I have reached out to my colleague who processed these matrices and will get back to you with a more definite response soon.

4) "that one has 1,932 columns and the other 1,867 columns."

The column number differences arise from the following: 1931 is the original number of DFC samples considered, which includes both adult and non-adult individuals. Once we filtered out the 65 non-adult samples, we obtained the 1866 individuals in the second matrix. Unfortunately, this was not made clear on the website. I will be updating this soon.

Q2:
But wait a minute, these files are useless without at least knowing who the cases and who the controls are?

A2:
Here is the rationale:
PsychENCODE placed restrictions on the dissemination of metadata. While adhering to those restrictions, we endeavored to put out as many of the processed datasets from our analyses as possible to allow for reproduction or downstream usage. This includes several intermediate files. Some may require protected data obtained with the permission of the consortium to perform downstream analyses, but even then the files on our website are in a format that would aid such analyses, and that are not available elsewhere.

For completeness, here are the answers to your original questions:
1) Method for generating DER-01: Using the original FPKM file, we filtered on >=10 individuals with >0.1 FPKM (though GTEx also applied a filter of requiring raw read counts greater than 6 — we did not have the raw data from GTEx, so we didn’t apply a filter on raw read counts).

2) Quantile normalization was performed to bring the expression profile of each sample onto the same scale.

3) To protect from outliers, inverse quantile normalization was performed for each gene, mapping each set of expression values to a standard normal.
2) Method for generating DER-02: The TPM file was converted directly from the original FPKM file

Query about QTL calling from Wang et al PsychEncode paper

Q:

I have a quick query about the Wang et al paper from the PsychENCODE study.
Were the QTLs identified from all the samples or the control samples only?
I’ve checked the paper, online resources and the supplementary methods but can’t seem to work this out.

A:
The QTLs were identified from both control and disease samples. You could find the sample information in Table S11. Summary of dataset.

Inquiry regarding PsychENCODE eQTL resource

Q1:

Was the eQTLs calculated on 1,886 unique individuals?

A1:
No, the eQTLs were calculated on 1387 filtered adult samples with matching gene expression and genotypes.

Q2:
In Fig S34, it mentions only 1,432 individuals have genotyped. How was the genotype information determined for the remaining 454 individuals?

A2:
We didn’t have genotype information determined for the remaining 454 individuals. So we didn’t include these 454 individuals in any QTL analysis.

Q3:
The # of samples with genotypes enumerated in Table S1 and Table S11 do not appear to match. For example, Table S1 reports 450 GTEx samples (97 DFC), but Table S11 reports 25 GTEx genotypes from the pre-frontal cortex. There might be some subtlety between these two tables I have missed, could you please clarify how to properly interpret these tables?

A3:
The Genotypes column in Table S11 only includes the filtered high genotyping quality samples (for example, genotype imputation accuracy score R2>0.3) which have matched RNA-seq data.