Unable to reproduce L1000 correlation
Hello, and thank you for this excellent work! I came across the blog post (https://www.sandboxaq.com/post/aqaffinity-open-source-structure-to-affinity-built-on-openfold3-by-sandboxaq) and was very impressed by the AQAffinity model.
I have been trying to reproduce the L1000 assay results shown in Figure 2 of the blog post. Specifically, I ran inference using the example inputs provided in this repository, which produced predicted activity values in the *_binding_head.txt output files. I then retrieved the corresponding experimental values from the CASP16 prediction center (https://predictioncenter.org/download_area/CASP16/targets/pharma_ligands/) and computed the correlation between the predicted and experimental values. Unfortunately, the result was essentially uncorrelated.
I was wondering whether the uploaded model_weights_only.safetensors file contains the fully trained parameters for the affinity prediction head, or if it might be an untrained or partially trained checkpoint. Any guidance on what might be going wrong would be greatly appreciated.
Thank you in advance for your time.