CryoMDM Supplementary: Data and Supplementary Materials

# CryoMDM Supplementary Package This Zenodo package provides datasets and supplementary materials for the cryoMDM study on the SARS-CoV-2 spikeprotein. The associated manuscript is currently under peer review, and its title may change during the reviewprocess. The contents of this dataset correspond to the version described in the related bioRxiv preprint below. ## Related preprintCryoMDM; Molecular simulation Driven structural Matching Approach to Estimate Variety of Atomic Three-dimensional ModelBased on Noisy cryoEM Single Particle Images. bioRxiv (posted 2024-11-01). doi:10.1101/2024.10.31.621264 ## Manuscript (under peer review)Current working title: “Atomic protein intermediates revealed from single-particle cryo-EM images by simulation-guided inference”. --- ## What this package containsThis package includes: 1. **Experimental cryo-EM single-particle image data** prepared from RELION 2D classification 2. **Target images** selected for structure inference (raw and VAE-enhanced) 3. **3D structural ensembles estimated by the cryoMDM method** - Global-search outputs - Local-search (PaCS) outputs4. **FMO (Fragment Molecular Orbital) calculation results** for the estimated structures 5. **RMSD analysis scripts/logs** comparing STEP1 and STEP2 6. **Supplementary videos (MP4)** reconstructed by cryoMDM --- ## Important terminology: RELION 2D classes vs VAE latent-space clustersThis package uses two different types of grouping, which should not be confused: - **RELION 2D classes (“RELION class”)** Classes obtained by **RELION 2D classification** of cryo-EM particle images. In the experimental data archive, particle images belonging to each RELION 2D class are saved individually as PNG files (e.g., in folders like `PNG_class*` after extraction). - **VAE latent-space clusters (“VAE cluster”)** Clusters obtained by applying a **VAE** to the PNG particle images and clustering in the **VAE latent space**. cryoMDM structure inference is performed based on these VAE clusters, and a target image (e.g., a cluster-center image) is selected for each cluster. **Note on naming (important):** In this package, some folders/files use the prefix `class#` (e.g., `3-Local-Search/class0/` and `PaCS-top10_cycle_minimized_class25_rank9.tgz`). Here, `class#` refers to the **VAE cluster index (historical naming)** and **does not** refer to a RELION 2D class. --- ## Directory / archive overview### `0-CryoEM-Experimental-Data.tar`Experimental cryo-EM particle images prepared from RELION 2D classification. Particle images for each **RELION 2D class** were exported and saved as **individual PNG files**, which were used as inputsfor the VAE and latent-space clustering. ### `1-Target-Images.tar`Target images selected **per VAE cluster** for structure inference, provided in both:- raw target images- VAE-enhanced target images ### `2-Global-Search.tar`Global-search outputs organized **per VAE cluster** (e.g., `cluster0/`, `cluster1/`, …). Each cluster folder includes, for example:- `output.json`: global-search output/metadata- `sorted_score.dat`: matching scores sorted in descending order (basis for ranking) ### `3-Local-Search.tar`Local-search (PaCS) outputs. Folder names are `class0/`, `class1/`, …; in this package these `class#` folders correspond to**VAE cluster indices (historical naming)** and are **not** RELION 2D classes. Local-search results provide explicit ranked PDB structures (**rank1–rank10**) per `3-Local-Search/class#/` folder. ### `4-FMO.tar`FMO calculation results and RMSD analyses, including:- STEP1 results (maximum force criterion: 500 kJ mol⁻¹ nm⁻¹)- STEP2 results (recalculation with stricter minimization for selected cases)- RMSD analyses for STEP1 and STEP2 ### `5-Supplementary-Video.tar`Supplementary MP4 videos reconstructed by cryoMDM:- `Cluster0_MD3_toUp_0012.mp4`- `Cluster25_MD1_Up_0051.mp4`- `Cluster33_MD1_Down_0017.mp4`- `Cluster5_MD2_Down_0020.mp4` --- ## Ranking scheme (VAE cluster-based)Structures are provided per **VAE cluster**. Within each cluster, structures are ranked by a matching score; **rank=1** indicates the highest score. --- ## What is inside each FMO archive (common to STEP1/STEP2)Each `.tgz` archive contains a full set of files for the FMO calculation of a single structure(corresponding to a specific VAE-cluster index and rank), including: - `*.pdb`: target 3D structure - `*.inp`: FMO input file - `*.ajf`: job configuration file - `*.sh`: run script - `*.sh.<jobid>.out`: standard output log - `output.<jobid>/...`: output directory (e.g., `stdout.*`) --- ## FMO calculations (STEP1/STEP2) and energy minimization settingsPrior to FMO, structure optimization (energy minimization) was performed using **Gromacs 2022.3** with: - Force field: **AMBER99SB-ILDN** - Minimization: **Steepest Descent** - Position restraints: protein heavy atoms, force constant **1000 kJ mol⁻¹ nm⁻²** - Convergence criterion for STEP1: maximum force **≤ 500 kJ mol⁻¹ nm⁻¹** Using the minimized structures as initial structures for FMO, electronic charges did not converge for **54 structures**. For these cases, the minimization criterion was tightened (maximum force **≤ 100 kJ mol⁻¹ nm⁻¹**) and recalculated as**STEP2**, which reduced the number of non-converged cases to **17**. **Note:** This package includes these **17 non-converged structures** as-is. --- ## Data formatThe data are provided as `.tar` archives to comply with Zenodo’s file-number limitation. To extract, run: `tar -xf <archive>.tar` --- ## [e-mail] ## LicenseCreative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (**CC BY-NC-ND 4.0**) ## KeywordscryoMDM; SARS-CoV-2; spike protein; cryo-EM; single particle analysis; RELION; 2D classification; VAE; latent space clustering;target images; structural ensemble; global search; local search; PaCS; FMO; Fragment Molecular Orbital; quantum chemistry;energy minimization; Gromacs; AMBER99SB-ILDN; RMSD; supplementary video