Generate Typical Protein Kinase Catalytic Domain in Inactive Conformations (CIDO/CODI/CODO)
author: Peter M.U. Ung @ MSSM
vers: 3.
- Step 1: compare the input kinase FASTA sequence to an internal library of protein kinase with atomic structures available
- Step 2: use the kinase structures that is most similar to the input kinase sequence as Base structure
- Step 3: align input kinase sequence to the selected kinase structures and a set of template kinase structures in alternative conformation (CIDI/CIDO/CODI/CODO/wCD)
- Step 4: generate homology models of input kinase based on the Base kinase structure and the template structures in alternative conformation
- Step 5: calculate the volume of the active site, rank and select ones with the largest volume
CIDO (aC-helix-in/DFG-out)
-
This is the typical DFG-out conformation described in the literature, where DFG-motif has a 180-degree flip and brings the DFG-Phe phenyl ring out from the hydrophobic deep pocket in the kinase, leaving the deep DFG-pocket vacant and available for small molecule binding.
-
This range of movement can be well-captured by one population (set) of kinase structures in DFG-out conformation, where DFG-motif has a clear 180-degree flip, while the set captures the range of small-lobe movement observed. Because the TK family has been heavily studied and with the most number of known CIDO structures, there are 2 sets of CIDO templates, one for TK family only and another one for all other kinases.
-
When doing molecular docking to these CIDO models, I found that default settings for OpenEye FRED works very well, while Schrödinger GLIDE will need a 0.75x adjustment to the GRID van der Waals radius to get similar results. In both cases, should also set a hydrophobic sphere in the DFG-pocket to enforce docking to that site while having H-bond constraints to the hinge residue.
CODO (aC-helix-out/DFG-out)
- Similar to CIDO conformation, a collection of kinase structures that are found to have CODO conformation is used as template for both TK and non-TK kinases. Only one population of models will be generated.
CODI (aC-helix-out/DFG-in)
- This is the C-helix-out conformation described in the literature as the "CDK-like" or "Src-like" inactive conformation. This conformation encompasses a range of aC-helix movements - translational, rotational, angular - that is more heterogeneous than DFG-motif movement (mostly just DFG-motif flip).
- Due to the large range of aC-helix movemnents, there are actually several (4+) sub-populations of CODI conformations. 4 of them are often seen in multiple kinases in different families, while some are family-specific. To generalize the modeling of CODI conformation, 4 sub-populations (sets) are used in the modeling process as compared to CIDO model generation.
- Notably, MEK has been seen to adopt a unique CODI conformation when bound to type-III inhibitor and ATP together, while cMET adopts another unique CODI conformation held up by an electrostatic interaction on its activation loop. These case-specific CODI conformations are not included in the general CODI sub-populations used for modeling but are also available for use.
CIDI (aC-helix-in/DFG-in)
- Different from CIDO/CODI/CODO model generation, modelling CIDI conformation uses the traditional approach where only 1 structure of the most related CIDI xtal structure is used in the homology modeling. The best-matched CIDI xtal structure is found via the result of Konformation, available in the Kinome Conformation Atlas KinaMetrix. Currently (2020.04) there are ~ 4,800 kinase xtal structures in RCSB PDB and ~ 4,000 have CIDI conformation, so there should be enough templates for use. Internally, it is using the same chemeric multi-piece template approach just like the other conformations, but since it is only using the best-matched xtal structure as the only template, and the multi-piece template is a copy of the same xtal multiple times, the end result is the same as doing a single-piece template modeling.
Website: KinaMetrix.com
Reference for Modeller homology modeling: Sali, Blundell. Comparative protein modelling by satisfaction of spatial restraints. J. Mol. Biol. (1993) 234(3), 779-815.
Reference for Modi-Dunbrack alignment: Modi V, Dunbrack RL. A structurally-validated multiple sequence alignment of 497 human protein kinase domains. Scientific Reports (2019) 9, 19790.
Reference for Muscle MSA alignment: MUSCLE: a multiple sequence alignment method with reduced time and space complexity. BMC Bioinf. (2004) 5, 113.
Reference for T-Coffee MSA alignment: Notredame, et al. T-coffee: a novel method for fast and accurate multiple sequence alignment. J. Mol. Biol. (2000) 302(1), 205-217.
Reference for BioPython: Biopython: freely available Python tools for computational molecular biology and bioinformatics. Bioinformatics (2009) 25, 1422-3.
######################################################################################
- Primary script: Modeling multiple kinases with simple default settings (multi-steps)
> 1_auto_DFGmodx.py
[ FASTA file of kinase(s) to be modeled ]
[ Conformation: cido | codi | codo | cidi | wcd ] ** capital or lowercase
[ Number of models to generate ]
[ Number of top-models to take ]
[ CPU per Model generation run ]
[ Step: 0 - Setup homology modeling (default) ]
[ 1 - Edit paths when change from local to HPC directory ]
## [ 1 - Setup homology modeling (_TEMP.y.fasta corrected) ] no used anymore
## [ #( rename *.y1.fasta to corrected *.y1.corr.fasta ) ]
[ 2 - Run homology modeling ]
[ 3 - Run volume calculation and selection ]
[ Optional Conformation: III\t- CODI MEK1-like for type-III inh ]
[ MET\t- CODI cMET-like altered A-loop ]
[ EGFR\t-CODI EGFR-like altered A-loop ]
e.g.> ./1_auto_DFGmodx.py \
ulk4_and_kit.fasta \
cido 10 5 5 \
0 # "0" will stop after generate initial setup
# "1" will alter the directory paths to match the ones in "x_variables.py"
# "2" will *only* generate actual models + volumes with successful setup
# "3" will *only* use generated models to calculate volumes
For general purpose and simply automated setup/running of DFGModx, use 1_auto_DFGMod.py. This is the wrapper script of the backbone script 1_run_single_DFGmodx.py with all default settings for modeling kinase structures in different conformations. The entire modeling takes 2-3 steps:
-
Step 0: generate the initial setup files and folders for all kinases being modeled with default setting
-
Step 2: generate kinase models + volumne calculation with default settings, results in the /1_dfgmod folder nameing convention: _TEMP.{0}.y1.fasta, where {0} = model output prefix.
-
step 3: an optional step to run only the homology model binding site volume calculation. Generally don't need to do this unless Step 2 failed.
-
Special, Step 1: This is used when the setup files (.setup, *.pir, *.pdb) are all ready and only need to do the production run (-mod, -vol) on a different computer and the paths/directories are different, e.g. from local WorkStation to HPC cluster. This updates all the directory paths to match the new computer's directories according to the new "x_variables.py"
> ./1_auto_DFGmodx.py kinome.fasta cido 10 5 5 1
######################################################################################
- Optional script: Run DFGmodx on individual kinase with more options (multi-step)
> 1_run_single_DFGmodx.py
[ Setup Script: formatted setup file | Pickle ]\n
[ Mode: -full Perform modified alignment, DFGmodel, and selection
-pir Run modified alignment only
-mod Run DFGmodel only
-vol Select top models only
-set Generate template setup script\n
-none * use new paths in "x_variables"; Workstation -> HPC
** Always use -pir then -mod, and ALWAYS check .pir for additional chain
breaks. Only use -full if you have checked the kinase and it has no
additional chain break or missing residues\n
[ Opt: -force Forced restart of /dfgworking directory ]
[ Opt: -restart Forced restart of /1_result directory ]
[ Opt: -pass Pass along; always use this unless otherwise ]
[ Opt: -paths * Update directory paths in setup files; use with "-none" ]
e.g.> ./1_run_single_DFGmodx.py setup.file -pir -pass
This is the backbone script to manually setup and modeling of a single kinase. To run this, several steps are needed:
-
Step 0: generate the initial template setup script when there is nothing
> ./1_run_single_DFGmodx.py -set -
Step 1: after modifying the setup file, generate the .pir file used by Modeller modeling. Recommended to check .pir file for errors before proceeding to model generation
> ./1_run_single_DFGmodx.py setup.txt -pir -pass -
Step 2: generate models using the .pir file. If no error occurs, volume calculation will start right after modeling.
> ./1_run_single_DFGmodx.py setup.txt -mod -pass -
Step 3: optional step, use if volume was not calculated right after the modeling step
> ./1_run_single_DFGmodx.py setup.txt -vol -pass -
Special: This is used when the setup files (.setup, *.pir, *.pdb) are all ready and only need to do the production run (-mod, -vol) on a different computer, i.e. from local WorkStation to HPC cluster. This updates all the directory paths to match the new computer's directories according to the new "x_variables.py"
> ./1_run_single_DFGmodx.py setup.txt -none -paths
###################################################################################
- Optional script: Check PDB structure by comparing PDB-sequence to official FASTA sequence
> 0_validate_PDB_fasta_seq.py
[ List of PDB file to be validated ]
[ No-gap FASTA sequences downloaded from RCSB for the PDBs ]
e.g.> ./0_validate_PDB_fasta_seq.py
pdb_structures_to_be_checked.list
correspond_complete_PDB_seqs.fasta
- PDB sequence downloaded from RCSB (or supplied) reflects the full-length sequence used in crystallography, but actually resolved PDB structure can have unresolved residues, resulting in Xtal FASTA sequence that is shorter by a few residues. Sometimes there are extra residues in Xtal structures that are not included in the published FASTA sequences too.
- This script compares the FASTA sequences published in RCSB (or supplied) and the xtal-FASTA extracted from the actual PDB structures. If the FASTA sequences have different lengths that indicates a discrepancy, they and aligned with Clustalo to show where the missing/inserted residues are. Use this detection result to correct the sequence in the FASTA database manually.
####################################################################################
- Optional script: Superimpose all kinase structures to a common reference frame
> 1_pymol_alignment.py
[Template PDB] * usually 1ATP.pdb
[Template Residues to superimpose] * 'resi 122-138+162-181' PyMOL naming
[List of Model PDBs]
[Output PyMOL session prefix]
[Name of Superimposed PDB List]
[Number of Model] (for Modeller PDB, B999000xx)
e.g.> ./1_pymol_alignment.py 1atp.pdb 'resi 122-138+162-181' model.list align_output _tmp.list 1
This script uses PyMOL as the main driver to superimpose all kinase PDB structures to a common reference frame, often 1ATP.pdb large-lobe, then save the superimposed stuctures with .mod.pdb ending. All structures used in DFGmodx, should use this to superimpose the structures to the common frame.
- PyMOL has a pretty good structure superimpose algorithm that is independent of the atom number and sequence matching, which is great for superimposing structures from different kinases with different lengths and residues.
- The 'super' function is sequence-independent while 'align' is sequence-dependent, and 'super' is the primary function to be used. Becuase some kinases have different C-lobe helices arrangement, it is safest to align to two upmost conserved region: hinge region + D-helix, and catalytic loop + beta-hairpin, '1atp and resid 122-138+162-181'.
- If the input structures are homology models from Modeller, they will have .B999000 in the filename. For these model structures, the filename will be condensed to generate {0}.B*.mod.pdb filename. If there are 10+ files, the first 9 models will have zero, i.e. .B0*, to match up for easier sorting.
- If the first round of 'super' superimposing fails due to 'Insufficient Number of Atom for Structure Superposition', that is a problem of older PyMOL that should be fixed by now but if that happens, 'align' will be automatically used instead to superimpose the structures. Although 'align' gives slightly different superimposed structure than 'super', the difference is minor.
###################################################################################
> 2_pir_result_info.py
[ fasta file with Kinase name ]
[ conformation: cidi -- cidi_s_templ or cidi_y_templ ]
[ cido -- cido_s_templ or cido_y_templ ]
[ codi -- codi_templ ]
[ codo -- codo_templ ]
[ wcd -- wcd_templ ]
[ output prefix ]
> 3_extract_dope_vol.py
[ Fasta Database ]
[ Prefix: cidi|cido|codi|codo|wcd ]
[ Top Number of model of Volume ]
[ Output Prefix ]
- A testing script trying to correlate z-DOPE score of a particular conformation model and its binding site volume as calculated by POVME. Not really useful but it uses plotnine to generate a nice looking plot.
######################################################################################
/a_examples
|-----/1_run_DFGmodx_cido
| |..... run_DFGmodx_cido_step_1.sh
| |..... run_DFGmodx_cido_step_2.sh
| |..... run_DFGmodx_cido_correction.sh
| |..... troubleshoot_common_issues.txt
| |..... ulk4_kit.fasta
| |...../correct_results/
| |...../KIT
| |...../ULK4
| |...../0_
| |...../1_
| |.....
| |.....
| |.....
| |.....
| |.....
|-----/2_run_DFGmodx_codi
|..... run_DFGmodx_codi_step_1.sh
|..... run_DFGmodx_codi_step_2.sh
|..... run_DFGmodx_codi_correction.sh
|..... troubleshoot_common_issues.txt
|..... ulk4_kit.fasta
|...../correct_results/
|...../KIT
|...../ULK4
|...../0_
|...../1_
|..... ...
-
In this example, 2 kinases (ULK4 and KIT) are used.
-
KIT modeling should go without a hitch since it has a fairly typical sequence.
-
Standard sequence alignment will fail for ULK4 because it has 1 fewer large-lobe helix than other typical kinases and will generate bad models (zDOPE > +0.0), while the pseudokinase nature makes the automated alignment of conserved region bad, hence will need to do a alignment correction before proceeding.
-
At the end of every step, the log file should be checked for WARNING and ERROR flags, especially when many kinases are being modeled at the same time. Most common errors are misalignment of the conserved region, mismatching between sequence and template structures.
###################################################################################
- Required packages:
Blast+ # 2.6.0+ standalone program or conda
POVME # 2.1 source codes already integrated into DFGmodx
PyMOL # 1.8+ standalone program or conda
Modeller # 9.20+ standalone program or conda
Clustalo # 1.2 standalone program or conda
Muscle # 3.8.51 standalone program or conda
T-Coffee # 13+ standalone program or conda
Python # 3.7.2+
numpy # 1.17.3
pandas # 0.25.3
pathos # 0.2.3
biopython # 1.74 conda
zlib # 1.2.11
bzip2 # 1.0.6
tqdm #
tarfile #