Molecular docking gives you binding affinity, however cannot defferentiate between activator or inhibitor. Molecular dynamics gives you the binding characteristics. If your set has atleast 20 cases with experimental IC50 you can try QSAR.

Irreversible inhibitors usually form covalent bonds with the protein in the active site. I don't know if there is a way of predicting this computationally. It is a matter of understanding the chemistry of the enzyme reaction and the chemistry of the inhibitor.

As Adam says, you need to know the chemistry. You can sometimes get an idea of what is targeted by looking at structures, eg. compounds that covalently bind to cysteine residues (a popular covalent-linking approach) often contain specific warheads. See eg. this paper: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3583020/ (Gray et al. Chem Biol. 2013 Feb 21; 20(2): 146–159).

If you know the reaction mechanism you may be able to use QM to assess activation energies for that reaction. There is no reaction-agnostic procedure that I know of (I do not think eg. HOMO-LUMO gaps will be of much use).

For a drug, reactivity of such warheads will need to be considered as tox can result if a warhead is too reactive. But it sounds like that is a worry for later.

As previously said by others, irreversible inhibitors are usually covalent inhibitors, though exceptions are also possible. I do not know any high throughput and reliable method to predict whether a given ligand would behave as an irreversible inhibitor. In fact, from an in silico approach, you are not even sure whether a particular ligand will actually bind the target protein in first place. The final answer comes from experimental validation of your predictions. I will recommend, however, to filter out from your compound database potentially reactive compounds before carrying out the virtual screening per se. This can be done using already-available filters, such as those included in MOE platform. By doing so, you will reduce the chances of getting compounds with potentially reactive groups and, therefore, potential candidates of irreversible inhibitors, in your final list of selected compounds.

Best regards, 

Jorge

You first need to check whether there is any cysteine residue in the pocket, which could potentially be reached by a ligand. Then you would take a non-covalent (reversible) inhibitor of the target of interest and graft a reactive warhead on it. It must be oriented with respect to the cysteine sulfur in a way allowing the reaction. It is unlikely to identify the best orientation of the warhead by modeling alone. You would rather need a few trial-and-error iteration with a medicinal chemist. Probably the hardest part would be to develop a right assay to test the compounds. Most biochemical techniques usually cannot distinguish between covalent and non-covalent inhibitors. The assay technology might involve either washout or mass spectrometry.

I suggest that you first use virtual screening software to make the hit compounds , than assay the results by performing molecular dynamic .

As pointed before, irreversible inhibitors are usually covalent inhibitors. All the docking programs use forcefields. Due to that, no docking program can predict covalent bond formation. Of course, using Molecular Dynamics wont solve the problem either.

My suggestion is that you do other type of simulation and study the electronic density using methods/theories like AIM (Atoms in Molecules). It is possible from the Laplacian of the electronic density characterize the type of bonding.

Of course the methods will be very computationally expensive (compare to docking) as they will take into account the electrons. Some methods are: semi-empirical (Using MOPAC with MOZYME keyword is possible to optimize the whole complex), Quantum Mechanic/Molecular Mechanics (use QM for the active site and MM for the rest of the system) or Fragment Molecular Orbital (FMO) were you can use any ab initio or DFT you want.

Thank you all for your time..