This user has not added any information to their profile yet.
- Parametric Bayesian Inference Based Optimization
- Method Development
- Machine Learning
- Quantum Computing
- Experimental Data Integration
The Rosetta Energy function is a substantial factor in many Rosetta applications. Yet, it is left untouched by most users. I hypothesize that by applying Bayesian Optimization the weight-configuaration~loss space can be approximated to an extend such that a configuration can be determined that out-performes the standard REF15 energy function on a given objective in reasonable time.
Therefore, i am implementing a generic script that hopefully enables researchers in the future to test whether there is a energy function configuration that is superior to the standard REF15 weights for their given task.
As a proof of concept I am applying this approach to optimize the rosetta fixed backbone design protocol on a 10 protein benchmark. Further, I use the Framework on the Fast Relax application in order to optimize the scaling factors for the repulsive term. The script is based on the scikit-optimize framework. I hope to enable researches to use Rosetta energy function configurations that are tailored to their specific task at hand and thereby achieve even better results in their projects.