Guidance on Replicating Protein Binder Design Results Using RFdiffusion

[Babanpreet05]

I am currently working on replicating the "Design of protein binders to rigid targets" portion of the RFdiffusion paper. For my study, I am focusing on two targets: the IL7 Receptor subtype Alpha (PDB: 3DI3) and the Insulin Receptor (PDB: 4ZXB).
Despite aligning my input parameters, constraints, and settings in pipeline in Google Colab as described in your publication, I am not obtaining results consistent with those reported. I suspect that there may be discrepancies related to software versions, random seed settings, or specific aspects of the energy function configuration that could be influencing the reproducibility of the designs.
I would greatly appreciate any guidance you could provide on the following:

1. Are there key parameters or settings that are especially critical for achieving the published results with these targets?

2. Could there be environmental differences (e.g., Google Colab setup, software/library versions, or random seed initializations) that might affect the outcomes, and if so, what adjustments would you recommend?

3. Do you have any troubleshooting tips or best practices for replicating the binder design for the IL7 Receptor and Insulin Receptor targets?

Parameters I have put in colab notebook
For IL7 Receptor subtype Alpha binder design : PDB: 3DI3, Hotspots- B58, B80, B139
For Insulin Receptor binder design: PDB: 4ZXB, Hotspots- E64, E88, E96
There are inputs that I put into colab notebook

This is the result of rmsd values of my designed protein with RFdiffusion official release binders
insulin_binder.csv

Thank you very much for your time and assistance. I am looking forward to your suggestions and any insights you can offer.