Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Symmetry-Driven Discovery of Dynamical Variables in Molecular Simulations
Authors: Jeet Mohapatra, Nima Dehmamy, Csaba Both, Subhro Das, Tommi Jaakkola
ICML 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To evaluate our method s effectiveness, we conducted experiments on two well-characterized molecular systems: alanine dipeptide and the designed mini-protein chignolin. These systems serve as canonical test cases in the molecular dynamics community, offering a balance between computational tractability and biological relevance. |
| Researcher Affiliation | Collaboration | 1Massachusetts Institute of Technology, Cambridge, MA, USA 2MIT-IBM Watson AI Lab, Cambridge, MA, USA 3North Eastern University, Boston, MA, USA. |
| Pseudocode | No | The paper describes methods like 'Direct Optimization', 'Full Hessian', 'Slow Hessian', and 'Degenerate Hessian' within the 'Methods for Extracting DOF' section, but it does not present these methods in structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain any explicit statements about releasing source code for the described methodology, nor does it provide any links to code repositories. |
| Open Datasets | No | The paper uses "alanine dipeptide" and "chignolin" as molecular systems for simulation, which are well-known in the molecular dynamics community. However, it does not refer to them as 'datasets' in the machine learning sense with associated access information (links, DOIs, etc.). Instead, it describes simulating these systems using force fields. Therefore, no concrete access information for a publicly available or open dataset is provided. |
| Dataset Splits | No | The paper describes simulation setups, such as using a "31 x 31 grid of angles" for generating deformed conformers and running "long open MM simulations for 500ns" for baselines. This refers to simulation exploration and sampling strategy, not to training/test/validation splits of a dataset used for model training or evaluation in the typical machine learning sense. |
| Hardware Specification | No | The paper mentions "computational burden" and discusses the quadratic scaling of "Hessian computation" with system size (e.g., "2sec for alanine-dipeptide, 200sec for chignolin"). However, it does not provide any specific details about the hardware (e.g., GPU/CPU models, memory) used to perform these computations or simulations. |
| Software Dependencies | Yes | We incorporate the AMBER force field, known for its accurate representation of molecular interactions, particularly in proteins. This force field is implemented using the parameters from Open MM (Eastman et al., 2017), and it comprehensively models the following interactions: ... For our baseline we run long open MM simulations for 500ns at 300K with friction coefficient of 1ps-1 and step size of 2fs amounting to 2.5e8 steps. We use the same amber forcefields in both the last step of our method and the baseline simulations in order to maintain consistency. |
| Experiment Setup | Yes | For our baseline we run long open MM simulations for 500ns at 300K with friction coefficient of 1ps-1 and step size of 2fs amounting to 2.5e8 steps. We use the same amber forcefields in both the last step of our method and the baseline simulations in order to maintain consistency. ... We make a 31 Ă 31 grid of angles (θ1, θ2) â [0, 2Ď)2. 4. Generate deformed conformers x = eθ1L1+θ2L2x0. 5. Use openmm.minimize on x and run short 2ps simulations to find stable conformations from the deformed structure. ... For our experiments, we use 16n2 samples for discovery and 16n2 samples for finding the most effective degrees of freedom. |