A list of Ligands. Maximum 20 ligands allowed.

Optional constraints for the prediction

This parameter controls the number of times the models output is fed back into the network for further refinement. Increasing the number of recycling steps can lead to a more accurate and refined final structure prediction as the model has more opportunities to converge on an optimal solution. However, each recycling step increases the overall computation time. Recommended for: Difficult or large protein complexes where initial predictions may need iterative improvement. Tip: For faster, preliminary assessments, a lower number of recycling steps can be used. For final, high-quality predictions, consider increasing this value.

This setting determines the number of discrete steps the diffusion model takes to generate the 3D structure from an initial noisy state. In each step, the model removes a certain amount of noise to progressively build a coherent and accurate molecular structure. Higher values: Generally result in a more detailed and higher-quality structure, but at the cost of longer processing times. Lower values: Lead to faster generation but may produce a less refined or lower-quality output. A sufficient number of sampling steps is crucial for the diffusion process to effectively denoise and generate a valid structure.

This parameter specifies the total number of independent structures the model will generate. Each sample is created from a different random initial noise distribution, leading to a potential diversity of final predictions. Generating multiple samples is useful for exploring different possible conformations of a structure and assessing the model's confidence and consistency.

This parameter adjusts the magnitude of change applied at each sampling step during the diffusion process. It influences the aggressiveness of the denoising at each iteration. Larger scale: The model takes larger steps in refining the structure. This can speed up convergence but risks overshooting optimal atomic placements, potentially leading to a less accurate or even distorted final structure. Smaller scale: The model takes more cautious, smaller steps. This can lead to a more precise and stable refinement process, but may require more sampling steps to reach a high-quality result.This parameter essentially modulates the trade-off between the speed of structure generation and the fine-grained accuracy of the final atomic coordinates.

Return the results without potentials.

The output format of the returned structure.

Concatenate Multiple Sequence Alignments for a polymer into one alignment.

The number of sampling steps to use for affinity prediction. Higher values may improve accuracy but increase runtime.

The number of diffusion samples to use for affinity prediction. Higher values may improve reliability but increase runtime.

Whether to add the Molecular Weight correction to the affinity value head.