If you prefer to use pip to install DeepDETAILS, you may have to install some non-Python dependencies yourself.

For bedGraphToBigwig (and other standalone programs made by UCSC Genome Browser team): you can directly download the binary version from the UCSC FTP site

1. Download the tools from: UCSC Genome Browser Utilities
2. Choose the version appropriate for your operating system (Linux, Mac, or Windows). For Linux, you can download it from the directory: http://hgdownload.soe.ucsc.edu/admin/exe/linux.x86_64/
3. Once downloaded, you can add the location of the binary (e.g., bedGraphToBigwig) to your system's PATH, or you can simply move it to a directory that's already in your PATH (e.g., /usr/local/bin/).
4. After that, you should be able to run the command like this: bedGraphToBigwig input.bedGraph genome.size output.bigWig

For bedtools, it's available from many package managers. If you are an Ubuntu user, you can use sudo apt-get install bedtools.

If you encounter with errors like the following

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 160.00 MiB (GPU 0; 23.69 GiB total capacity; 22.24 GiB already allocated; 114.06 MiB free; 23.19 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

It means you don't have enough GPU memory to support the deconvolution of your sample with the current DeepDETAILS configuration. Here are several things you can try:

• Use the fused version of DeepDETAILS instead of the sequence only model (don't add --seq in your command)
• Use a smaller batch size, the default batch size is 32, smaller values such as 16 or 8 can be helpful. Example: --batch-size 16
• Aggregate similar cell types to reduce the total number of deconvolution target

While our compendium includes many libraries from the same tissue, users should prioritize deconvolution results based on high-quality bulk libraries and biologically well-matched bulk–reference pairs that are as close as possible to the context they want to study. These factors may affect prediction reliability. Results may be less reliable when bulk data are low quality or when the reference and bulk samples differ substantially in tissue context, developmental stage, age, sex, disease state, or other biological characteristics. We therefore recommend reviewing the quality metrics and sample metadata provided in the portal and, when possible, confirming that relevant marker genes or known features are enriched in the selected deconvolved tracks.