DMCloud

DMcloud is a macromolecular structure modeling tool using point cloud matching with diffusion model for cryo-EM maps at 3-15A resolution

Installation

Clone the Repository

First, clone the DMCloud repository:

git clone https://github.com/kiharalab/DMCloud.git
cd DMCloud

Download Model Weights

Download the pre-trained diffusion model weights:

mkdir -p best_model
wget -O best_model/diffusion_best.pth.tar https://github.com/kiharalab/DMCloud/releases/download/weights/diffusion_best.pth.tar

Prerequisites

First, install a conda package manager if you don't have one:

• Miniforge (recommended): Download from https://github.com/conda-forge/miniforge
• Miniconda: Download from https://docs.conda.io/en/latest/miniconda.html

Environment Setup

Using conda:

conda env create -f environment.yml
conda activate dmcloud

Or using mamba (faster alternative):

mamba env create -f environment.yml
mamba activate dmcloud

Note: If you installed Miniforge, mamba is included by default. For Miniconda users, install mamba with: conda install mamba -n base -c conda-forge

Usage

python DMcloud.py --Source <protein_structures> --Target <ldp_points_or_map_file> --OutPath <output_directory> [options]

Examples

Example 1: Basic Fitting with MRC Map (with Diffusion)

# Fit AlphaFold2 model to cryo-EM map using automatic diffusion to generate LDP points
# Uses default config file (config/diffusion.json)
python DMcloud.py --Source example/af2_model.pdb --Target example/1461_complex.mrc --OutPath ./results/ --contour 0.52

Example 2: Scoring Mode for Existing Models

# Score existing fitted models without running fitting
python DMcloud.py --Source fitted_model.pdb --Target example/1461_complex.mrc --OutPath ./scoring_results/ --ScoreMode True --ScoreWindow 3

Parameters

Parameter	Type	Default	Required	Description
--Source	str	-	Yes	Protein structures in PDB or CIF format. For multiple files, split filenames by comma or use pattern like DATA/*.pdb
--Target	str	-	Yes	LDP points in CIF format, or MRC/MAP file (diffusion will be run automatically to generate LDP points)
--OutPath	str	./FitModels/	No	Final output directory
--ransac_dcut	float	3.0	No	Corresponding distance cutoff for RANSAC
--icp_dcut	float	3.0	No	Corresponding distance cutoff for ICP
--clash_dist	float	2.0	No	Clash definition distance
--n_ransac_iter	int	2000000	No	Number of RANSAC iterations
--n_iter	int	1	No	Number of iterations
--MinClstSize	int	200	No	Minimum size of fitting
--LocalRadius	float	8.0	No	Local radius of FPFH
--LocalNN	int	5	No	Number of nearest neighbors for FPFH
--LocalSegment	int	100	No	Size of the local segment for RANSAC
--LocalSegmentDistance	float	15.0	No	Radius of the local segment for RANSAC
--MinSim	float	50.0	No	Minimum similarity of FPFH
--Alpha	float	1.0	No	DBSCAN weight = exp(-alpha*deviation)
--AveLDDT_cut	float	0.4	No	Average LDDT cutoff
--NrRate	float	0.95	No	Non-redundant ratio of models (A&B)/(AorB)>X
--DownsampleSpace	float	10.0	No	Sampling space for local fitting
--FillGaps	float	5.0	No	Fill gaps within X angstroms from aligned positions
--Threads	int	8	No	Maximum number of threads in CP
--CPtime	int	600	No	Maximum CP computing time (seconds)
--CPpen	int	1400	No	Penalty for conflict models
--pLDDT	float	50.0	No	pLDDT filtering. If negative, use all points. Set to 100 for all points
--DomEps	float	5.0	No	Domain splitting eps parameter
--DomSamples	int	5	No	Domain splitting samples parameter
--SaveTemp	bool	False	No	Save temporary files
--pkl	str	-	No	Resume computation from the pkl file of fitting
--cp_pkl	str	-	No	Resume computation for CP part
--cp_size	int	2000	No	Number of models per iteration
--ScoreMode	bool	False	No	Scoring mode for the given model
--ScoreWindow	int	0	No	Window size of the score ±N (0 = only that residue)

Diffusion-Related Parameters (for MRC/MAP files)

Parameter	Type	Default	Required	Description
--diffusion_config	str	config/diffusion.json	No	Config file for diffusion parameters (uses default config/diffusion.json when target is MRC/MAP file)
--diffusion_gpu	str	0	No	GPU ID for diffusion (when target is MRC/MAP file, defaults to 0)
--contour	float	0.0	No	Contour level for input map (when target is MRC/MAP file)

Important Notes

• Config File Path Resolution: When using MRC/MAP files as targets, DMcloud automatically uses config/diffusion.json as the default configuration. The config path is resolved relative to the DMcloud.py script location, so the tool works correctly when called from any directory.
• Cross-Directory Usage: You can run DMcloud from any directory, and it will automatically find the default config file relative to its installation location.

License

License: GPL v3. (If you are interested in a different license, for example, for commercial use, please contact us.)

Contact

Contact: Genki Terashi ([email protected]) Daisuke Kihara ([email protected])

---

Standalone Backbone Diffusion Module

Usage

python diffusion.py -m <input_map_path> --config <config_path> -g <gpu_id> -o <output_directory> -c <contour_level>

Parameters

Parameter	Type	Default	Required	Description
-m, --map	str	-	Yes	Input map path
--config	str	config/diffusion.json	No	Config path specifying diffusion and LDP parameters (defaults to config/diffusion.json)
-g, --gpu	str	-	No	GPU ID to use
-o, --output	str	-	No	Output directory
-c, --contour	float	0.0	No	Contour level for input map (suggested: 0.5 × author_contour)