Tutorial: optimizing an antibody against a ligand

Sometimes our starting docked structure is not very good, and the binder may loose the target after some nanoseconds. This is more frequent when simulating docked ligands. locuaz supports the addition of positional restraints so users can get started with their optimizations, until a better binder is found.

In this tutorial we will use what we learnt in previous tutorials, plus some other new tricks, to optimize a nanobody against a tirosol molecule, like the one on Figure 1.

Figure 1: our starting complex, a tirosol molecule docked to a nanobody.

As usual, activate your locuaz environment and get the necessary files.

Necessary files

As always we’re going to need a starting PDB and as in Tutorial: using Tleap topologies, we’ll also need a set of tleap related files in order to rebuild the topology of our system after each mutation.

1. tir.pdb: the PDB file of the pre-equilibrated complex. As usual, target chains go first, also, remember that since we are using Tleap, residues should be numbered on a continuous progression.

2. tleap: Tleap dir with the script to build the topology of the system each time a mutation is performed. Remember to avoid solvating and creating a box in this file, since the solvent will already be present. Another thing to notice is the usage of addions. We keep this commands since Tleap will be responsible of keeping neutrality of the system. Avoid using addions2 since we need it to replace water molecules each time it ads ions, to keep the N of the system constant. You’ll also find lig.frcmod and lig.prep, the auxiliary tirosol parameters.

3. config_ligand.yaml: the input file to run the protocol.

4. mdp directory: minimization, NVT and NPT GROMACS input files.

If you are finding it hard to get a PDB of your system with chainID information, check the FAQ.

The configuration file

We will focus on the new options that didn’t show up on the previous tutorials.

protocol

protocol:
    epochs: 10
    new_branches: 3
    constant_width: false
    memory_size: 4
    failed_memory_size: 6

• constant_width: when this value is set to false, new_branches means the number of branches (new mutations) that are obtained from each previous branch. Check Platform DAGs for more info.

• We’re setting new_branches to 3 because we will input 4 bins in the next section.

creation

creation:
    sites: 2
    sites_interfacing: true
    sites_interfacing_probe_radius: 3.0
    sites_probability: uniform
    aa_bins: ["CDEST", "AGIMLV", "PFWY", "RNQHK"]
    aa_bins_criteria: without
    aa_probability: uniform

As we said, we’re dealing with a rather small system, so we’re going to mutate 2 sites at the same time and on each branch we’re going to try 3 different amino acids from 3 different bins so we can sample the solution space quickly.

mutation

mutation:
    mutator: dlpr
    allowed_nonstandard_residues: [UNL]

• allowed_nonstandard_residues: locuaz cleans up the PDBs that go into the mutator program, so they don’t have any non-protein residues that may cause the mutator to error. We have to add our target ligand as an exception to this rule so that it’s included in the side-chain optimization that’s performed after inserting a mutation.

pruning

pruning:
    pruner: metropolis
    kT: 0.593

• pruner: metropolis: uses the Metropolis-Hastings criteria to decide if a new branch passes to the next epoch. Does not work with multiple scorers.

• kT: product between the boltzmann constant and a temperature. The current value corresponds to a temperature of 300K.

md

md:
    gmx_mdrun: "gmx mdrun"
    mdp_names:
        min_mdp: min.mdp
        nvt_mdp: short_nvt.mdp
        npt_mdp: short_npt_posres.mdp
    mps: true
    numa_regions: 1
    use_tleap: true
    maxwarn: 2
    box_type: octahedron
    npt_restraints:
        posres: 50
        posres_water: 50

• mps: when set to true, locuaz will use the NVIDIA Multi-Process Server (MPS), to run multiple MD simulations per GPU. This usually decreases the speed of each run, but considerably increases the total throughput. Useful when using a variable width DAG protocol which may make the number of branches explode. Check this blog post for more info.

• numa_regions: when using MPS, locuaz will automatically set these options: ngpus, mpi_procs, omp_procs and pinoffsets. To be able to do this effectively, it needs to know CPU affinity of each GPU, which should follow the NUMA layout. Check the FAQ if you don’t know how many regions you have.

• npt_restraints: This is where we set the value for our positional restraints. Remember also to define the -DPOSRES and -DPOSRES_WATER flags in your NPT mdp file so these take effect.

scoring

scoring:
    scorers: [autodockvina]
    allowed_nonstandard_residues: [UNL]
    nthreads: 6
    mpi_procs: 1

• allowed_nonstandard_residues: when scoring, the NPT trajectory is split in “sanitized” PDB frames. That is, they receive a treatment to make sure the scorers don’t error out when meeting unexpected artifacts, like non-standard residues. All scorers but gmxmmpbsa use this PDBs. Since we want to score the interaction between our nanobody and a tirosol molecule, we need to add it to this list of residue names, so locuaz doesn’t remove it from the PDB frames.

Running the protocol

Nothing new here, we just run the protocol with our config file:

mamba activate locuaz
python /home/user/locuaz/locuaz/protocol.py config_ligand.yaml

It’s educational to look at the DAG with the branch names that locuaz draws. Given the elevated branching, it’s difficult to see the whole DAG, but check Figure 2 for a part of it.

Figure 2: section of the Directed Acyclic Graph (DAG) of a sample optimization against the tirosol molecule. We can see 3 epochs, the epoch 0, the initial one. Then a branch from epoch 1 (the one in the middle), and 6 branches from epoch 2 at the bottom. The edges that go out from one of these bottom branches belong to the epoch 3 (not shown). We can see that the branch from epoch 1 was approved and moved on to be mutated again on 2 sites, generating 6 branches in total, of which only 1 was approved and mutated again another 6 times.