Releases: ClementiGroup/pyODEM
Major Update - Protein non-bonded interactions implemented
Non-Bonded Protein Interactions for Larger Proteins
Implemented features to handle larger proteins. This required implementing MPI in order to distribute the protein data and calculation onto multiple cores to deal with scaling to larger proteins. This also required implementing a new coarse-grained model that defines generic non-native interactions to reduce the number of adjustable parameters thereby reducing the search space of protein model parameters.
Updated Features
- Implemented non-bonded pair interactions into a protein model. (requires pysph)
- Implemented parallelization using the mpi4py package to improve scaling to larger proteins and larger data-sets.
- Model loaders now work so that models can be loaded in parallel.
- Implemented helper functions that partition the data onto multiple cores based upon the discrete state assignment of the protein conformation.
- Compute the quality factor in a distributed manner thereby speeding up the computation.
Dependencies
- model_builder (https://github.com/ajkluber/model_builder)
- mdtraj
- numpy
- scipy
- multiprocessing
- mpi4py
- pysph (https://pysph.readthedocs.io/en/latest/)
See publication: Chen, J., Chen, J., Pinamonti, G. & Clementi, C. Learning Effective Molecular Models from Experimental Observables. J. Chem. Theory Comput. 14, 3849–3858 (2018).
Initial Release
Initial Release of pyODEM
Basic features of pyODEM are implemented and tested. The goal is to optimize a coarse-grained protein model using either real or synthetic histogram data, i.e. FRET (Forster Resonance Energy Transfer).
Basic Features
- Load protein models per ajkluber's model_builder package.
- Compute coarse-grained protein potential energy.
- Compute a quality factor and its logarithm based on histogram data, i.e. distance FRET data.
- Determine discrete state probabilities due to changing the coarse-grained protein model's parameters.
- Select optimal coarse-grained protein model parameters based upon quality factor using a non-linear optimizer.
Dependencies
- model_builder (https://github.com/ajkluber/model_builder)
- mdtraj
- numpy
- scipy
- multiprocessing
See publication: Chen, J., Chen, J., Pinamonti, G. & Clementi, C. Learning Effective Molecular Models from Experimental Observables. J. Chem. Theory Comput. 14, 3849–3858 (2018).