Created a model S133202 of a protein on the 5th of April. This model has a C-Score of -3.61, which corresponds to a TM-Score range of 0.21 to 0.43
Five weeks later, I recreate the same model using the same inputs. In the meantime there have been weekly updates to the databases used by the service and two other improvements listed on news. I would expect a slight improvement in the C-Score for the newer model.
The new model S138189, generated on the 16th of May, has a C-Score of -4.11, corresponding to a TM-Score range of 0.19 to 0.37.
A few questions:
This is a sizable decrease - what caused this?
Shouldn't the C-Score be expected to improve over time?
Is the newer model less accurate?
Why is the server generating worse models over time?
Thanks