DSR: Dynamical Surface Representation as Implicit Neural Networks for Protein

Authors: Daiwen Sun, He Huang, Yao Li, Xinqi Gong, Qiwei Ye

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental results demonstrate that our model accurately captures protein dynamic trajectories and can interpolate and extrapolate in 3D and time. We design experiments from three aspects to verify the ability to learn SDF from the raw point clouds, the ability to reconstruct the protein dynamical surface, and the generalization of the model in terms of temporal interpolation and extrapolation.
Researcher Affiliation Academia 1Institute for Mathematical Sciences, School of Mathematics, Renmin University of China, Beijing, China 2Beijing Academy of Artificial Intelligence, Beijing, China 3School of Life Sciences, Tsinghua University, Beijing, China
Pseudocode No The paper provides an "Overview of DSR" in Figure 2, which visually illustrates the training and inference processes. However, it does not include any formal pseudocode blocks, algorithms, or structured steps labeled as such.
Open Source Code Yes Codes are available at https://github.com/Sundw-818/DSR, and we have a project webpage that shows some video results, https://sundw-818.github.io/DSR/.
Open Datasets Yes 500ns Trajectory Data. This data set[28] currently contains two 100ns atomistic molecular dynamics trajectories of Abeta (40 residues)... URL https://doi.org/10.7910/DVN/ERYOZS. MDAnalysis Data. MDAnalysis Data collects a set of data resources pertaining to computational biophysics... Ad K equilibrium dataset[30]... URL: https://figshare. com/articles/Molecular_dynamics_ trajectory_for_benchmarking_MDAnalysis/5108170, doi, 10:m9. GPCRmd. The GPCRmd (http://gpcrmd.org/)... DRYAD_MD. This dataset contains the trajectory data of 23 proteins simulated by Jumper JM[38]... URL https://doi.org/10.5061/dryad.h9f8sb7.
Dataset Splits No The paper describes how data frames were selected for training (e.g., "used the first 1500 frames... and diluted them by ten times... selecting 150 frames for each trajectory for training"). It also discusses evaluation through interpolation and future prediction, which serve as test cases. However, it does not explicitly specify a separate validation dataset split (e.g., 80/10/10 split or specific number of samples for validation) used for hyperparameter tuning or early stopping.
Hardware Specification No The paper does not explicitly describe the hardware used for running experiments. It mentions "computational resources" generally but does not specify GPU models, CPU types, or other detailed hardware specifications.
Software Dependencies No The paper mentions using "Py Mol" and "Py Torch Autograd" and "python scikit-image package" but does not provide specific version numbers for these software components, which are necessary for reproducible dependency listing.
Experiment Setup Yes For representing shapes we used level sets of MLP f(x, t; θ) = R4 Rdz R, called DSR Model, with 8 layers, each containing 512 hidden units, and a single skip connection from the input to the middle layer. We set the loss parameters to λ = 0.1, τ = 1, α = 1e 3. And the model architecture is as Figure 2. The activation function between fully connected layers is softplus activation: x 7 1 β ln 1 + eβx , where β = 100. The initial latent code vector z of size 192, were sampled from N(0, 1.02).