Merizo is a fast and accurate deep learning method for domain segmentation in complex protein structures. Notably, it makes use of invariant point attention (IPA) to read a protein structure into a latent representation. Domains are predicted via an affinity learning approach whereby the embeddings of residues belonging to the same domain, are encouraged towards similar embeddings and discouraged if belonging to different domains.
If you use Merizo, please cite the paper:
Lau et al., 2023. Merizo: a rapid and accurate protein domain segmentation method using invariant point attention. Nature Communications. (link)
Packages needed to run Merizo can be installed either individually or concurrently using the requirements.txt file:
pip install -r requirements.txt
Individual dependencies:
pip install torch==2.0.1
pip install scipy
pip install matplotlib
pip install einops
pip install networkx
pip install rotary-embedding-torch
pip install natsort
pip install numpy<2
If your machine has low disk space consider starting by installing CPU-only torch with
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
The standard usage for Merizo is:
python predict.py -i /path/to/file.pdb
Multiple pdb files can be supplied to -i:
python predict.py -i /path/to/file_1.pdb /path/to/file_2.pdb /path/to/file_3.pdb
or:
python predict.py -i /path/to/files/*.pdb
Merizo can be ran either in standard mode, or an iterative segmentation mode where domains larger than a certain size are continuously passed through the network until convergence (i.e. the assignment no longer changes). This length threshold is set to 200 residues (roughly just smaller than the average domain in CATH) in predict.py.
The iterative mode should be used for processing any long models (~ >800 residues), as well as models from the AlphaFold Protein Structure Database.
python predict.py -d cpu -i examples/2xdqA.pdb
python predict.py -d cpu -i examples/*.pdb --save_domains --save_pdf --save_fasta
python predict.py -d cpu -i examples/2xdqA.pdb --save_domains --plddt_filter
python predict.py -d cpu -i examples/AF-Q96PD2-F1-model_v4.pdb --iterate
By default if no device flag (-d) is specified, if a GPU is available, it will be used. If no GPU is available, Merizo will default to using CPU. On Apple Silicon devices, the -d mps option can be given to use the Metal Performance Shaders (MPS) backend.
-h, --help show this help message and exit
-i INPUT [INPUT ...], --input INPUT [INPUT ...]
Specify path to pdb file input. Can also take multiple inputs (e.g. '/path/to/file.pdb' or '/path/to/*.pdb').
-d DEVICE, --device DEVICE
Hardware to run on. Options: 'cpu', 'cuda', 'mps'.
--save_pdf Include to save the domain map as a pdf.
--save_pdb Include to save the result as a pdb file. All domains will be included unless --conf_filter or --plddt_filter is used.
--save_domains Include to save parsed domains as separate pdb files. Also saves the full pdb.
--save_fasta Include to save a fasta file of the input pdb.
--output_headers Control if the table returned to stdout has a header line
--conf_filter CONF_FILTER
(float, [0-1]) If specified, only domains with a pIoU above this threshold will be saved.
--plddt_filter PLDDT_FILTER
(float, [0-100]) If specified, only domain with a plDDT above this threshold will be saved. Note: if used on a non-AF structure, this will correspond to crystallographic b-factors.
--iterate If used, domains under a length threshold (default: 200 residues) will be re-segmented.
--max_iterations MAX_ITERATIONS
(int [1, inf]) Specify the maximum number of re-segmentations that can occur.
--shuffle_indices Shuffle domain indices - increases contrast between domain colours in PyMOL.
By default, Merizo will only output a summary of the parsed PDB file which is printed to the terminal:
filename nres nres_dom nres_ndr ndom pIoU runtime result
2xdqA.pdb 425 425 0 3 0.98407 1.89492 6-18_296-459,19-156,157-162_192-295
Residue ranges are delimited by -, e.g. residues 1 to 200 are 1-200. Separate domains are delimited by ,, e.g. 1-200,201-400, and discontinuous segments are delimited by _, e.g. 1-200,201-300_350-400.
nres, nres_dom and nres_ndr are the total number of residues, number of residues in domains and number of non-domain residues respectively.
If processing a large number of files, it may be useful to redirect the output into a file using > for reference. This can be performed by:
python predict.py -d cpu -i examples/*.pdb > examples/results.txt
PDB files are generated when using either the --save_pdb or --save_domains flags.
If the --save_pdb flag is used, the original PDB with the domain assignment will be saved with a .pdb2 file extension. The domain index of each residue is written to the occupancy column. Values in the b-factor column (used for plDDT in AlphaFold2) will be preserved. NDRs are assigned an occupancy of 0.0.
ATOM 1 N PRO A 6 -8.641 99.395 86.382 1.00 0
ATOM 2 CA PRO A 6 -9.798 100.225 85.975 1.00 0
ATOM 3 C PRO A 6 -9.490 101.162 84.771 1.00 0
ATOM 4 O PRO A 6 -10.415 101.802 84.223 1.00 0
...
ATOM 217 N GLN A 32 12.896 69.675 61.036 2.00 0
ATOM 218 CA GLN A 32 13.321 68.978 62.230 2.00 0
ATOM 219 C GLN A 32 12.070 68.331 62.894 2.00 0
ATOM 220 O GLN A 32 12.125 67.232 63.353 2.00 0
If the --save_domains flag is used, a PDB file of each domain structure will be saved separately, as well as one file for the entire structure. The --conf_filter and --plddt_filter flags can be used to designate pIoU and plDDT thresholds to use for the saved files. Domains with pIoU lower than the value provided by --conf_filter or plDDT lower than the value provided by --plddt_filter will not be saved.
If the --save_pdf option is provided, the domain map of the assignment will be saved in a graphical format. This will generate a single .pdf file that contains both the predicted domain map as well as a pIoU map which describes the network confidence for each domain.
We recommend using PyMOL for visualising the PDB outputs of Merizo. When the .pdb2 file is opened in PyMOL, the domain assignment can be visualised using the spectrum q command in PyMOL applies colouring to each domain to help differentiate them.
Tips:
- When a larger number of domains are in the structure, the contrast between colours used by
spectrum qmay not be very high, making neighbouring domains hard to differentiate between. Using the--shuffle_indicesflag in Merizo may help increase the contrast between different domains/colours. - NDRs are assigned to domain index 0. After
spectrum q, their positions will always be in dark blue. To manually apply a colour to NDRs, thecolor white, q < 1command can be used to colour all NDRs white.