A simple implementation of Protein Discovery with Discrete Walk-Jump Sampling
NOTE:
- The validation losses for both EBM and denoiser modules are 0.01 (uses MSE loss) and 1.3 (uses cross-entropy loss) respectively.
- Previous code commmit history is found here
- See a bit of theory study here
TODO:
- Needs to revise the inference code
- Adapts the code for BEND benchmark
- Study and implement the effects of post-translational modifications PTM of RNA-binding proteins RBPs
Credit: o3, Claude, Gemini AI bots