A Review of Medicinal Chemistry Toolkit
The Medicinal Chemistry Toolkit app is a growing suite of resources to support the day to day work of a medicinal chemist. The app was created in conjunction with the book The Handbook of Medicinal Chemistry: Principles and Practice that is based on the very successful Medicinal Chemistry summer school run by the RSC. The app was developed in collaboration with Molecular Materials Informatics.The app has been recently updated and so I thought it might be useful to have a look at the capabilities.
The app is organised to provide medicinal chemists with a toolbox of useful calculators that can be used in the various stages of Drug Discovery.
These tools include
Chemical Formula, MWt, GCLogP, Ligand Efficiency, Ligand Lipophilic Efficiency and AZFilters calculations
Firstly it is important to note that the calculations are done locally on the iPad and not via a web service, this is important for confidentiality when dealing with proprietary compounds. Properties are calculated as you sketch in the molecule so you can actually see the contributions of functional groups as you add them. I found the display seemed optimised for landscape viewing.
The logP is calculated using the updated Ghose-Crippen algorithm as described in J. Chem. Inf. Comput. Sci., 1999, 39 (5), pp 868–873 DOI, the 68 atom type contributions to log P were determined by fitting an extensive training set of 9920 molecules, with r2 = 0.918 and σ = 0.677. The method also includes 4 supplementary atom types intended to catch any atoms not typed. With all these types of calculations the errors will accumulate as you wander into chemical space not described within the original test set, however within a chemical series you will probably find the relative calculations within the series are still very useful. In particularly the contributions due to the introduction of specific functional groups or atoms can be observed. If you want accurate LogP values then you need experimental data.
Ligand Efficiency is intended to act as a guide to the binding energy per atom of a ligand to its target.
LE = 1.37(-logIC50)/Heavy atom count
Ligand Lipophilic Efficiency is intended to help medicinal chemists avoid the trap of simply increasing LogP to increase affinity.
LE = pIC50 - GCliogP
Both measures can be used for selecting hits from a screening campaign and they can be used during the optimisation program to ensure that any added atoms are not simply adding baggage.
The AZFilters are intended to flag undesirable chemical features, these are based on extensive medicinal chemistry experience gained from prosecuting the results of multiple screening campaigns and the process is described in Chemical predictive modelling to improve compound quality, Nature Reviews Drug Discovery 12, 948–962 (2013) DOI. There are over 100 filters and by drawing in a few ugly structures you can rapidly get an idea of the sort of things that are flagged. What is particularly nice is that the offending motif is highlighted by the program and a short text description included. Many companies will have their own filters but these seem to be pretty comprehensive. I'm guessing that now the filtering system is in place other filters might be added in the future.
LogP vs pH Curves
Whilst the Ghose-Crippen algorithm gives an estimate of the LogP many drug like molecules contain ionisable groups (acidic and basic) and predicting the behaviour of such molecules requires a knowledge of the degree of ionisation at different pH. As shown in the image below despite having identical LogP the two compounds have very different LogD at physiological pH (7.4). There seems to be a minor bug, when you change the properties the graph updates, but not the legend which always displays compound 1 "acid".
Free Energy Converter
This app estimates the change in binding expected by a given change in Gibbs free energy at a given temperature (ΔG = −RTln Kd) in both kJ/mol and kcal/mol. The change in binding constant between two molecules (ΔΔKd, in units of M−1) can also be entered into the app to see the derived ΔG in kcal/mol and kJ/mol.
The Cheng-Prusoff equation DOI is probably the most important equation for medicinal chemists involved in the design of inhibitors to understand.
Where Ki is the binding affinity of the inhibitor, IC50 is the functional strength of the inhibitor, [S] is fixed substrate concentration and Km is the concentration of substrate at which enzyme activity is at half maximal. Thus for kinase inhibitors the Ki will vary depending on the concentration of the substrate ATP [S]. It is therefore vitally important to understand the likely range in ATP concentration in the target locations, and to be aware that accumulation of the substrate (ATP) will have an impact on inhibition. This can be part of the reason why shifts in activity are observed when moving from a biochemical assay to a cell-based assay.
In many assays it is often convenient to set the substrate concentration to Km thus
Ki = IC50/2
Plasma Protein Binding
Effectively all drugs bind to plasma proteins, some with very high affinity others with only modest affinity, in the case of human serum albumin the capacity for binding can be huge and so it can have an impact on ligand binding, and a variety of ADME properties. A quick way to get an idea if plasma protein binding may be an issue is to add serum to the in vitro screen, if the apparent affinity drops it is often evidence that the ligands are binding to plasma proteins, you can also add the purified components (e.g. Human serum albumin) to get an idea of which proteins might be involved. By estimating the potency shift due to plasma protein binding you can get a feel as to whether the observed values seem reasonable.
This is series of calculations that uses data from in vitro and preclinical species to predict likely human parameters.
The data required to make the predictions are: * Unbound cell potency * Intrinsic clearance (ideally from human hepatocytes) * Estimated human volume of distribution (estimated from volume of distribution measured in preclinical species and corrected for protein-binding differences across species) * Protein binding in man and the PK species used to estimate human Vss
The tools allow the user to get a handle on clearance, volume of distribution and ultimately likely human dose, it also allows you to experiment with different dosing intervals.
Maximum Absorbable Dose
The maximum absorbable dose (MAD) calculator encodes a simple calculation that indicates any likely absorption problems limited by solubility or permeability as the compound passes through the human gut.
cMAD = Sol x Ka x SIWv x SITT
Where Sol is compound solubility at pH 6.5 (mg/mL), ka = Absorption rate constant (min -1), SIMV = small intestinal water volume (mL) and SITT small intestinal transit time (min).
If the dose is much less than the cMAD then it is likely the drug will be well absorbed and this will not limit efficacy, but if the required dose is above the MAD, there is significant risk the dose will not be completely absorbed and it might be necessary to investigate different formulations.
The Attrition Modeller is based on data from the Steven Paul article in Nature Reviews Drug Discovery (2010), 203-214 DOI, and is a sobering illustration of the costs involved in drug discovery. It also underlines the cost of failed discovery projects.
I've been very impressed by the current version of the Medicinal Chemistry Toolkit app, it now has a set of tools that will be useful in the day to day work of a medicinal chemist. As more features get added I can see this becoming the "swiss army knife" for medicinal chemists. To be honest the app would be useful for biologists as well ;-)
Last Updated 18 May 2015