Binder Desing via Editing Config File

[WAI272]

Hi,

I’ve been working on setting up a Binder desing by editing the config files manually (base.yaml) but it always returns a error, perhaps I should go with codes shown in readme?

[roccomoretti]

Starting with the provided examples and then tweaking them for your particular use case is generally a good strategy.

I will say that the typical use case for RFdiffusion is to provide the relevant system-specific configuration parameters on the command line, rather than modifying yaml configuration files directly. Unless you're doing something advanced, you shouldn't need to edit the yaml files. (Even if you do want to use a custom yaml file, my recommendation would be to make a copy and edit that, rather than modifying the main one directly.)

If you're still getting an error message, it would be helpful to copy-and-paste it into your help requests, so we can have a better sense of what specifically is going wrong in your case.

[WAI272]

Thank you for your reply!
Here`s the error:
[2025-07-31 20:46:26,773][rfdiffusion.inference.model_runners][INFO] - Loading checkpoint.
[2025-07-31 20:46:30,242][rfdiffusion.diffusion][INFO] - Using cached IGSO3.
Successful diffuser init
[2025-07-31 20:46:30,260][main][INFO] - Making design samples/design/
[2025-07-31 20:46:30,266][rfdiffusion.inference.model_runners][INFO] - Using contig: ['A103-112', '70-100']
Error executing job with overrides: []
Traceback (most recent call last):
File "/home/wall/Programs/RFdiffusion/./scripts/run_inference.py", line 84, in main
x_init, seq_init = sampler.sample_init()
File "/home/wall/Programs/RFdiffusion/rfdiffusion/inference/model_runners.py", line 341, in sample_init
assert xyz_27.shape[0] == L_mapped, f"there must be a coordinate in the input PDB for
AssertionError: there must be a coordinate in the input PDB for each residue implied by the contig string for partial diffusion. length of input PDB != length of contig string: 117 != 10

Set the environment variable HYDRA_FULL_ERROR=1 for a complete stack trace.

[WAI272]

But when I switch it to configs: ['47-47' ,'A103-112', '70-100'],
It stars to return this error message:
Error executing job with overrides: []
Traceback (most recent call last):
File "/home/wall/Programs/RFdiffusion/./scripts/run_inference.py", line 84, in main
x_init, seq_init = sampler.sample_init()
File "/home/wall/Programs/RFdiffusion/rfdiffusion/inference/model_runners.py", line 309, in sample_init
self.hotspot_0idx=iu.get_idx0_hotspots(self.mappings, self.ppi_conf, self.binderlen)
File "/home/wall/Programs/RFdiffusion/rfdiffusion/inference/utils.py", line 657, in get_idx0_hotspots
for i, res in enumerate(mappings["receptor_con_ref_pdb_idx"]):
KeyError: 'receptor_con_ref_pdb_idx'

[roccomoretti]

contig string for partial diffusion

You have partial diffusion enabled. With partial diffusion, you're perturbing the input PDB slightly. As such, you need to have starting residues to perturb. -- You can't combine partial diffusion with de novo structure generation.

Your contigs, therefore, need to match the length of your input PDB. Having a designation of '70-100' in your contigs means that you're randomly sampling somewhere between 70 to 100 residues, which won't necessarily match the length of your input PDB. If you do want to do partial diffusion, all of your contig segments must be of a single length, and must match up with the input PDB length.

If you did want to do both de novo generation and partial diffusion, you'll need to split it into two steps. First do the de novo generation of backbones (with the fixed context), then in a separate second step take the output of that first step and run partial diffusion on those structures.

[WAI272]

I couldn't thank you, enough. I was bit frustrated because of this. I'd like to ask you one more thing though, is there any way to install RFdiffuision for only for cpu usage? I've read your instruction for "RuntimeError: NVTX functions not installed. " and you suggested:

conda install -c pytorch libpytorch pytorch

It did work for once but for the second time while I was installing it didn't work and pro-mpted something like there was a no module like that.
Thank you for your time.

[roccomoretti]

Putting together a robust installation guide for CPU only installs is something that's in the works. There's a beta documentation installation page, and the instructions for Apple Silicon Architectures might be of help (one of the hurdles with Apple machines is the lack of CUDA GPUs).