Computational chemistry is a branch of chemistry that uses computers to assist in solving chemical problems. It uses the results of theoretical chemistry, incorporated into efficient programs, to calculate the structures of molecules and solids.
In theoritcal chemistry, chemists, physicist and mathematics develop algorithms and computer programs to predict atomic and molecular properties and reaction paths for chemical reactions. Computational chemists, in contrast, may simply apply existing computer programs and methodologies to specify chemical question. There are two different aspects to computational chemistry.
Computational studies can be carried out in order to find a starting point for a laboratory synthesis, or to assist in understanding experimental data, such as the position and source of spectroscopic peaks.
Computational studies can be used to predict the possibility of so far entirely unknown molecules or to explore reaction mechanisms that are not readily studied by experimental means.
Computer-aided drug discovery/design methods have played a major role in the development of therapeutically important small molecules for over 2-3 decades . Over the past couple of decades, many powerful standalone tools for computer-aided drug discovery have been developed .
In silico metabolism
After adopting combinatorial chemistry and high throughput biological screening in the past couple of decades, the pharmaceutical industry generated a large collection of potent and selective compounds for numerous targets. However, to become an optimal drug, in addition to potency and selectivity, a compound must have appropriate ADME (absorption, distribution, metabolism and excreation), safety and developability characteristics. Relaying solely on potency in the early stage of drug discovery can result in disproportionate attrition after clinical candidate selection contributing to the exorbitant costs of discovering and developing drugs. Only about one in ten of those diligently chosen, highly potent and selective candidates that enter development reach the market often due to inadequate ADME properties. Therefore, it is extremely important to consider the ADME characteristics of compounds earlier in the discovery process to wager bets on compounds that have a greater potential to survive the development and clinical trail stage of drug development. Increasing the odds of success to one in five (instead of ten) would reduce the total cost of bringing a new therapeutic to the market by 33%.
Experimental determination of ADME and pharmacokinetic (PK) characteristics is both expensive and time consuming, and is not practical for large numbers of compounds, especially when the pharmaceutical industry is under severe pressure to cut costs and improve efficiency. The Price tag to support various ongoing discovery projects in pharmaceutical company for synthesis and high throughput measurement of permeability, solubility, metabolic stability and acute toxicity can run into millions of dollars.
Therefore, much attention is being focused on the application of in silico screens to reliability predict ADME attributes solely from molecular structure. In silico prediction of ADME properties will not only reduce cost and development cycle times by wisely directing resources to essential experimental testing, but also bring forward their consideration earlier at the lead generation stage when compounds are being synthesized and tested almost exclusively to meet pharmacological target potency levels. At the cost of experimental results indicated above, a mere 10-20 % reduction in high throughput experimental measurement of permeability, solubility, metabolic stability, acute toxicity through the use of in silico screens can lead to significant savings. Further, application of in silico screens offers an ideal ‘fail-early-fail-cheaply’ strategy for drug discovery because their application requires nothing more than inputting the basic structural information of a compound into a validation model.
Metabolic Stability Measurement in Drug Discovery
At most major pharmaceutical companies, metabolic stability assays are conducted at the first investigation into the metabolism of a compound. These in vitro assays generally utilize liver microsomes and/or hepatocytes to furnish important information about the rate and/or metabolism. In vitro metabolic studies are important in optimizing pharmacokinetic properties such as in vivo half life, maximum concentration and systemic exposure, because rapid metabolism is often a key factor contributing to poor exposure.
The metabolic stability data is helpful for ranking molecules with respect to their ability to resist metabolism. Though high-throughput automated metabolic stability assay systems have been developed by the major pharmaceutical companies, screening a large number of compounds is still intensive.
Thus, in silico prediction of metabolic stability can be used to rationalize experimental testing and have significant resources. Further, these models allow for prediction of metabolic stability for virtual libraries, thus bringing forwarded their consideration earlier to hit-to-lead stage.
In silico Models for Metabolism Studies
There are several types of in silico tools available for investigations into metabolism. These include knowledge based systems (metabolism databases) rules based or expert systems and quantity structure properties relationship (QSPR) and enzyme structure modeling systems. At its simplest, the partition coefficients, Log P (or its computed equivalent), of a drug in the n-octanol-water system has been shown to loosely correlate with the metabolic stability of a compound. As in silico calculations of Log P values have become readily available, they are being implemented in many of the current metabolism prediction packages. The earliest in silico metabolism tools were the metabolic reaction databases.
In principle, these are databases with the published metabolic reactions and structure of parent compounds and their metabolites. Some of these allow creation of corporate metabolic databases as well. In most cases, these databases can be searched for specific biotransformation (by structure and substructure), biotransformation keywords, and by other user-defined fileds. The key advantage of this knowledge based in silico systems is that they include detailed metabolism findings and original references.
Prediction of Metabolites
The above mentioned knowledge based systems provided the groundwork for the development of rules based in silico predictors of metabolites, also called expert systems.
Commercial rules based programs such as METEOR, META and MetabolExpert iteratively interrogate the chemical bonds of a molecule and apply programmed biotransformation rules in a predetermined hierarchy. As one can imagine, without means of terminating the metabolism tree a very large numbers of metabolites will be generated of predicted. Several of these programs allow the user to specify the number of levels of biotransformation or may use a LogP calculation to terminate the biotransformation process.
Prediction of Sites of Metabolism
The mechanism of oxidation by CYPs is though to be constant across all CUPS. One of the most important steps in the oxidation of drugs (by CYPs) is the ability of the perferyl oxy species (FeO+) to carry out a one-electron oxidation through the abstraction of hydrogen atoms. In silico packages such as Admensa, COMPACT and Metasite calculate the likehood of abstracting a hydrogen atom from all sites on a molecule, and then quantify which sites are most likely to be oxidized. The greatest advantage of these types of predictions is the ability to quantify the most likely major “hot spots” on the molecule.
Prediction of Substrate Binding
The ability of a molecule to properly dock on the active site of enzyme plays a major role in accurately determining the site(s) of metabolism on a molecule. Therefore, modeling of the P450 active sites has long been a goal for metabolism prediction. Some of the first attempts to model the active site of metabolizing enzymes used pharmacophore, site directed mutagenesis, and protein homology. Using homology models built from bacterial and mammalian P450s, in silico studies of docking potential substrates into the active site have been performed with mixed success. Now that the crystal structure for human CYP3A4, CYP2C9, CYP2C8 and CYP2A6 are available docking experiment with these models should more reliably predict the sites of metabolism on a molecule that the predictions form the homology models.
Prediction of Metabolic Inhibition
One of the causes of drug-drug interactions is the inhibition of metabolic pathways. Enzyme inhibition by a drug leads to a decrease in metabolism and intrinsic clearances, and an alteration in pharmacokinetics of a co-administered drug. Any knowledge around the potential of drug-drug interactions is useful for a quantitative assessment of the ability a new molecular entity to inhibit the metabolism of another drug.
In silico methods to predict drug-drug interactions are mostly limited to competitive inhibitors because they rely primarily on the binding models in the used as templates with alignment and orientation of core structure in an active site. However, the quality of prediction depends on the structural similarity to the molecules that were used to build the model. Inhibitors of CYP2D6 and CYP2C9 have been predicted using the CoMFA method. For inhibitors of CYP3A4, CYP2C9 and CYP2D6, methods such as CATALYST and GRIND have been used to reduce the bias in the alignment of inhibitors.
Prediction of Enzyme Induction
The induction of drug metabolizing enzymes is an additional way in which co-administered drugs can affect the clearance and pharmacokinetics of a given drug. Induction or increased expression of the drug metabolizing enzymes leads to an increase in the rate of metabolism and ultimately, to increase intrinsic clearance. Therefore, enzyme induction leads to decreased drug exposure which may results in therapeutic failure.
Most often, the induction of enzymes occurs through the activation of nuclear receptors such as the Arylhydrocarbon receptor (Ahr), the Pregnane X receptor (PXR) and the constitutive Androstane Receptor (ACR). The activation of nuclear receptors PXR and CAR are responsible for the induction of several drug metabolizing enzymes including CYP3A, UGT1A1, SULT1A and CYP2C9. On the other hand, induction of CYP1A, SULT1A1 and UGT1A1 has been associated with activation of AhR.
Prediction of Metabolic Stability
As mentioned above, metabolic stability influences both oral bioavailability and half life of a drug. There is good correlation between in vitro metabolic stability and in vivo clearance. Therefore, the assessment of metabolic stability of compounds is being demanded and earlier in discovery projects. To meet these needs, in silico models are commonly employed.
As described above, many different in silico tools focused on studying different aspects of metabolism have been developed. In our experience, the choice of the in silico tool by a user should primarily be guided by the type of information and level of accuracy that is desired. It is imperative that the developers of these in silico tools continually refine and validate them to reliably predict and quantify the metabolic fate of drug in humans. On the other hand, the chemists, biologists and ADME scientist on project teams to evaluate and use the existing in silico tools and to challenge their developers to demand tools that will rationally and efficiently move the discovery projects forward.
Metabol Expert is an ideal program for a quick prediction of the metabolic fate of compound in the drug discovery process during the dispositional research phase. Metabol Expert is a unique tool for initial estimation of the structural formula of metabolites. Metabol Expert is a rule based system with open architecture, in other words, the chemists, metabolism researchers, drug disposition experts and environmental managers can understand, expand, modify or optimize the data on which the metabolic structural estimation relies. Commercialized by Compudrug in 1987, Metabol Expert is composed of a database, a knowledge base and several prediction tools. The basic biotransformation database contains 179 biotransformations, 112 of which are derived from Testa and Jenner, the others are based on frequently occurring metabolic pathways.
The transformation knowledge-base is composed of if-then type rules. Each is composed of our components, the structure changed during the transformation, the new substructure formed, a list of substructure at least one of which must be present in the molecule for the biotransformation form occurring. These rules have been derived from the literature by experts and are input into the system by means of graphical tools. The system is open so that new rules can be added or existing ones modified or deleted. There are two types of predictions in Metabol Expert. In the first type, the system tries to match basic transformations automatically. There is a filter so that biotransformation sequences can be arrested after generation of a specified number of metabolites.
Basic transformations are classified as phase I or phase II. In the event of a phase II metabolite being generated, the sequences are arrested and the metabolite is not included in the next level. If a transformation results in the generation of two metabolites, then both are passed into the next level. The second type of analysis is an extended prediction model in which metabolites generated from basic transformations is compared to a list of transformations in a learned tree for a given species and the analogues are then listed in order of similarity. The program then attempts to quantify predictions based in the information in the learned tree.
MexAlert was developed to be an ideal assistant for high-throughput screening. It is advantageous to consider metabolism still before synthesis of the compounds, in order to exclude unwanted metabolic pathways, leading, for example, to first pass effect or to formation of toxic intermediates. MexAlert predicts first pass metabolic pathways by quickly identifying sites on the molecule where Phase II metabolic transformations (in other words, conjugation) may occur. It is rule based system; the rules are selected from among the Phase II transformations in the animal knowledge base, and modified according to in vivo experimental example of first-pass effect pathways.
In silico toxicity prediction
Attrition during the drug development process is a serious economic problem for the pharmaceutical industry and it is often due to inappropriate ADME/Tox characteristics. IT has been estimated that 20-40% of the drug failure in investigational drug development phases are due to safety issues, not continuing multiple incidents of adverse effects of existing drugs. The early drug discovery process needs to address in parallel not only potency but also pharmacokinetics and toxicological properties.
Van de Waterbeemd and his colleagues at Pfizer have called this approach ‘property-based design’, emphasizing the importance of the critical combinations of physical and structural properties that contribute to ‘druglikeness’. Ideally this process should begin early in discovery, using computational models to screen both virtual libraries and available compound collections to identify compounds with the desired properties (good potency, ADME and low toxicity). Often molecular size and lipophilicity have an important effect on all three properties. High biological activity is frequently associated with high LogP, but this may also raise the probability of high toxicity.
In the 1990s, drug companies invested heavily in combinatorial chemistry and high-throughput screening (HTS) as a source of leads for new targets. Most screen actives turned out to be large and/or hydrophobic, clearly contraindicating to the principles of minimal hydrophobicity. The most visible outcome was not even that they were toxic (because most compounds did not advance that far in development), but that they were either insoluble or non-absorbable. This overshadowed any toxicological consideration in the early stages of drug discovery, bringing forward compounds solubility and permeability as the most urgent problems to address. In the late 1990s the concepts of drug-likeness and lead likeness emerged and simple rules were formulated e.g. “rule of 5” to warn chemists when compounds were well outside the property space normal for orally active drugs.
These rules are now widely used in Virtual Screening to remove undesirable compounds from consideration prior to their synthesis or acquisition. In addition to simple property filters toxicological issues should also be considered because otherwise any specific hazardous sub structural effects are ignored. Some of the harshest reactivity effects are identified and removed using predefined alert substructures (e.g., acid halides) (sometimes called “garbage filters”). The problem is the most of such alert substructures are “chameleonic” in nature, i.e., they may not necessarily cause toxic effects depending on other functional groups and overall molecular structures (e.g., acid halides). To fix this, all chameleonic” substructures (from “garbage” filters and beyond) must be supplemented with class specific QSAR’s for different health effects, yielding toxicological expert systems. Such systems can be used in virtual screening along with “drug –likeness” filters to subdivide compounds into “safe”, “hazardous” and “Questionable”. Promising compounds need further toxicological evaluation, but this cannot be done by predicative methods alone due to multiple knowledge gaps in their training sets and limited numbers of considered toxicological end points.
Sometimes further evaluation is done during lead optimization, when chemical structures are covalently modified and tested. At this stage various toxicological predictions are used to prioritize both compounds that are to be tested and in vitro methods can reliably predict drug’s effect on a whole animal system, yet Animal Tests are not usually done until drug development candidates are identified. By this time drug development cost can reach substantial levels, economic risks become considerable.
Thus any predictive tools that help to identify adverse effects in animals prior to conducting such testes are highly desirable. If a known toxic pharmacophore is identified and closely related to the pharmacophore required for activity at the therapeutic target then series specific SAR is needed to aid design of molecules with an adequate separation between the therapeutic dose and the dose threshold for the toxic effect. Since it is essential to have an in vitro surrogate that can model the desired in vivo effect on a significant number of compounds. Some of the more important end points for which in vitro surrogates that can model the desired in vivo effect on a significant numbers of compounds. Some of the more important end points for which in vitro surrogates have been used are: QT prolongation in heart due to blockade of the hERG potassium channel, hepatotoxicity due to compounds that produce phospholipidosis and hepatotoxicity due to induction of CYP450 enzymes.
To summarize, in drug design toxicity predictions can be useful for three purpose – virtual screeing, prioritization of compounds and in vitro tests, and prediction of health effects in whole animal systems. Although many toxicologists are understandably leave us with no choice but to make the attempt using available animal toxicity database.