AlphaFold2-RAVE v 1.0 Please cite the following reference if using this protocol with or without the provided code: AlphaFold2-RAVE: From sequence to Boltzmann ensemble Bodhi P. Vani, Akashnathan Aranganathan, Dedi Wang, Pratyush Tiwary bioRxiv 2022.05.25.493365; doi: https://doi.org/10.1101/2022.05.25.493365 Dedi Wang and Pratyush Tiwary , "State predictive information bottleneck", J. Chem. Phys. 154, 134111 (2021) https://doi.org/10.1063/5.0038198 If the provided code is used for the protocol, please also cite: ** add references for: colabfold, OpenMM/conda-plumed, metadynamics This will be updated on peer reviewed acceptance Components This protocol is essentially a methodology that combines two schools of thought: structure prediction, and enhanced sampling to preserve thermodynamics. In this repository, we demonstrate one instance of such a methodology. For structure prediction, we use AlphaFold2, or more specifically ColabFold. We introduce stochasticity to the ColabFold algorithm by decreasing the MSA cluster size and generating structures with multiple random seeds*. For thermodynamics, we perform molecular dynamics simulations with metadynamics. To bias the simulations, we learn collective variables from the now-diverse set of structures using SPIB*. *The changes to the ColabFold algorithm, as well as the additional code required to use the SPIB for biasing have been provided in this github and their use is demonstrated in the colab files ravefuncs.py - functions required to run this instance of our methodology sample_config.ini - basic input configurations required for SPIB heavydemo.ipynb - our full protocol demonstrated to run on the colab GPU server to show improvement in sampling in a short time frame compared to unbiased simulation. This may take 2-4h to run. litedemo.ipynb - a tutorial of our protocol without running simulations, explaining how to build required code and files to adapt the methodology to your system and computational resources CSP_data - files required to run parts of heavydemo.ipynb and litedemo.ipynb on the benchmark system cold-shock protein 1HZB. Please note that the hyperparameters used in heavydemo are chosen for efficiency, not robustness. Please refer to comments and litedemo for parameters chosen for robustness