How to install it?

[harryyiheyang]

Hi Hui and Rahul, I'm truly inspired by the capabilities of HEELS, especially its superior accuracy demonstrated in simulations compared to LDSC. However, I'm currently facing some hurdles when attempting to install it on our HPC infrastructure.

Specifically, the 'git clone' command doesn't seem to be working for downloading from GitHub. This is a minor concern though.

What's more concerning is the potential presence of incompatible packages, which is impeding a successful installation. I'm keen on harnessing the power of HEELS for our research, but these installation issues are causing delays.

Additionally, I've come across a valuable resource from PRScs and PRScsx. They provide population-specific reference panels that can be downloaded from a Dropbox link (https://www.dropbox.com/s/y3hsc15kwjxwjtd/UKBB_ref.txt?dl=0). Is it possible to incorporate it into HEELS, as it is always a significant problem in addressing the complex task of estimating large LD matrices in GWAS?

Thank you very much

[huilisabrina]

Hi Harry,

Thanks for your interest in using our software! Please see my answers to your questions below:

• Installation: I tried fixing the clone command. Could you try rerunning it from our Wiki page? From what I can see in the screenshot, it seems like you might need to wait a bit longer to set up the environment / it was still loading. What's the error you got when the command conda env create --file heels.yml failed or stopped running? Also, which version of Python are you using?
• LD reference panel: Yes it would be possible to incorporate those LD matrices into HEELS, but there are a few pre-processing steps to be done, such as block-diagonalizing the LD matrices (since HEELS estimates local h2 for each block) and converting the matrices saved in HDF5 format (provided by PRC-cs) into a format that is readable by Python. Additionally, if you're interested, you can try out our banded + LR approximations (see tutorial here).

Hope this helps, Hui

[Hugolyu]

Hi Hui Thank you for your excellent work in establishing local heritability. However, I encountered the same problem in fixing the _conda_ environment after several attempts using _conda_ and _mamba_. It appears to be a conflict or uninstallable in some dependency packages.

So can you provide the **singularity** wrapper for HEELS? Thank you very much!!!!

[anbai106]

The same problem here:

Collecting package metadata (repodata.json): done Solving environment: / Found conflicts! Looking for incompatible packages. This can take several minutes. Press CTRL-C to abort. failed / Solving environment: / Found conflicts! Looking for incompatible packages. This can take several minutes. Press CTRL-C to abort. failed

[huilisabrina]

Hi @anbai106 and @Hugolyu, thanks for your feedback. Just confirming that I've received your questions and will look into as soon as I can. Thanks, Hui

[huilisabrina]

Hi, I made two updates that can potentially address the installation issues, but I'm not entirely sure what's causing the error and conflicts. Can you please try re-pulling the repo, and run the following commands?

I'd really hope this package can be useful for you, so please keep me posted if things have been fixed or you need more support.

Thanks, Hui

[harryyiheyang]

Hi Hui, I still suffers from this problem that: "Found conflicts! Looking for incompatible packages. This can take several minutes. Press CTRL-C to abort. failed" On the other hand, I read your nice wiki but find that I may need to estimate a series of LD matrix myself which may takes considerable time. I guess you may consider the suggestion again -- offer the public LD matrices we can use directly. For example, if you can deal with the public LD matrices offered by PRSCSX (estimated from UKBB). Thank you

[huilisabrina]

Hi @harryyiheyang @anbai106 @Hugolyu ,

Thank you again for your patience. I was able to reproduce your problem (thanks to one of my lab mates) and figured out a solution now. Could you please run the following to set up the environment for running HEELS?

# Replace the following line with any codes that enable you to run python in the terminal
module load python/3.10.12-fasrc01 

# Create a new env called heels, adding packages that are not provided in the base
mamba create -n heels python=3.10.12 pip numpy pandas scipy pandas-plink joblib

# Install the scikit-learn package for some SVD functions
source activate heels
mamba install scikit-learn-intelex -c conda-forge

Note: If you do not have mamba, you can replace it with conda in the lines of code above (but it's recommended that you use conda to activate to the env, see below). The codes should still run, but mamba is significantly faster than conda.

To check the installation is successful, you can go to the HEELS package folder and run the following. There should be a list of flags and options availble for use.

conda activate heels
python run_HEELS.py --help

Please let me know if this helps. I will update the manual and wiki pages accordingly soon.

Hui