Dear Siam Siraji ,

It is hard to say based on your limited info. According to Alpha Fold they state, “Regions with pLDDT between 50 and 70 are low confidence and should be treated with caution.” Since this can be caused by multiple reasons (significant portion unstructured, no related protein in their database, etc.) it is difficult to say what to do next. Suggestions are:

-Use I-Tasser (https://zhanggroup.org/I-TASSER/ ) and compare/combine, see for example https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8592092//

-Use secondary prediction programs to check certain regions (for example is a helical region predicted in the secondary structure prediction present in your predicted 3D structure etc.)

-If your result is caused by the fact that you are looking at possible domain packing, go to the next step “The second metric is PAE (Predicted Aligned Error), which reports AlphaFold’s expected position error at residue x, when the predicted and true structures are aligned on residue y.” (https://www.deepmind.com/publications/enabling-high-accuracy-protein-structure-prediction-at-the-proteome-scale)

But again, it very much depends on the reason(s) why the confidence score is rather low.

Best regards.

Hi Siam Siraji,

It depends very much on what you mean with proceed... proceed with what? You can use the model to get an idea, or a starting model, but with any model (even if the score is better), in order to work with it, it serves to do predictions, but those would need virification before you can rely on them... So using it as a starting model, or trying to relate it to a complex, or to predict sites for mutations, that can all be done based on such a model, but you have to keep in mind that it is only a prediction until you have proven that it is correct. molecular simulations or drug-discovery projects for example, I wouldn't do based on a prediction of a structure, but on a solved structure. Does that help?

Thank You! Linda E Franken Rob Keller