Test-Run Installation

[JuuelHerza]

Hello,
I have two questions regarding Opus-Mut.

1. Is it possible to keep the original orientations of the sidechains outside of the range of the mutated residue?
   Like if I want to mutate the active site but I'd rather keep other regions constant (like the solvent exposed surface).
2. I installed the conda environment with mut.yml , I have an Nvidia 4080 Super and I get the following error after running "run_opus_mut.py":

2024-11-21 14:23:29.026061: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudnn.so.7
2024-11-21 14:40:33.652735: W tensorflow/stream_executor/gpu/asm_compiler.cc:63] Running ptxas --version returned 256
2024-11-21 14:40:33.712515: W tensorflow/stream_executor/gpu/redzone_allocator.cc:314] Internal: ptxas exited with non-zero error code 256, output:
Relying on driver to perform ptx compilation.
Modify $PATH to customize ptxas location.
This message will be only logged once.
2024-11-21 14:40:34.533496: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcublas.so.10
2024-11-21 14:41:25.165263: W tensorflow/core/framework/op_kernel.cc:1763] OP_REQUIRES failed at cwise_op_gpu_base.cc:89 : Internal: Failed to load in-memory CUBIN: CUDA_ERROR_NO_BINARY_FOR_GPU: no kernel image is available for execution on the device

What configuration step am I missing?

Thanks

[thuxugang]

Hi,

1. Sure, modifying the code to achieve the desired functionality is feasible. For example, you can alter line 449 in dlpacker.py "if not refine_only: self._remove_sidechains(self.reconstructed)". Keep the side chain of other residues, and reconstruct the side chains of activte site. However, i am not sure it is a good idea since the mutation will definitely affect the side chains of others. You can reconstruct the side chains of WT and your mutants, and see the differences. BTW, you can try our lastest side-chain modeling method OPUS-Rota5 (https://github.com/thuxugang/opus_rota5).
2. The issue you are experiencing is often attributed to a mismatch between the CUDA version and your GPU. Since my testing was conducted exclusively on a V100 GPU, I recommend upgrading both TensorFlow and CUDA to versions that are compatible with the 4080 GPU.
   Best,

[JuuelHerza]

Hello, thanks for the answer.
Sorry, I haven't been able to install it due to several messages of incompatibilities in the conda environment.

I think that for a multi-domain protein with domain-domain associations it might be risky to alter the experimentally determined sidechains at the interface domain1-domain2 when the purpose is to mutate the catalytic site.

Thanks again!