The simplest exercise in QSAR is canonical ordering which starts with choosing a group of chemicals, and a selected property or biological activity for each. In this slide, nine chemicals are listed with their boiling points. If we think we understand how chemical structure relates to boiling point, we would expect that we could choose those molecular descriptors which would place the chemicals in the same order as would a measured behaviour such as boiling point.
In this slide, the chemicals are sorted by increasing boiling point. Can we identify molecular descriptors that create the same order. If not, we do not understand the inter- and intramolecular forces that control boiling point. If QSAR can order them properly, the task is then to find chemicals that fit between these values an test the QSAR model. Through numerous iterations , theoretical explanations can be evaluated for relevance, and the important molecular descriptors are discovered. This came approach can be used for toxicity data.
In this example, I am illustrating that there are many toxicity mechanisms, and if all the chemicals having the same mechanism are compiled, it would not be unusual for the potency of those chemicals to range over 8-10 orders of magnitude. Even if the range were much less, the first challenge for QSAR would be to identify a molecular descriptor that places the chemicals in the same order as the potency measures (LC50). To illustrate, I am using aquatic lethality with fish just to move away from the rodent inhalation example, but keep in mind that a fish test is just an inhalation test with aquatic organisms.
For many mechanisms, uptake of the chemicals is controlled by passive transport and one would expect the octanol/water partition coefficient to covary with passive transport. When the entire range of potency values are plotted vesus Log Ko/w, the chemicals remain in the same order and quantitative relationship between LC50 and Ko/w can be derived exactly like that for the rodent inhalation data.
General Concepts in QSAR for Using the QSAR Application Toolbox Part 2
General Concepts in QSAR for Using the QSAR Application Toolbox <br />
Part IITrend Analysis and Filling Data Gaps in Hazard Assessment<br />
Some Important Lessons<br />We saw that vapor pressure correlates with rodent LC50s, but hundreds of other molecular descriptors may not <br />The trend in the data will be evident only when the toxicity mechanism of all chemicals is the same <br />Chemicals with other mechanisms (i.e. acrolein, phosgene) will appear as statistical outliers to the major trend because they do not produce lethality as do the aliphatic ethers.<br />The QSAR approach uses trend analysis to evaluate mechanisms and to test chemical similarity in terms of common toxicity mechanisms or modes of action<br />
QSAR and Trend Analysis<br />QSAR research asks simple questions about why chemicals behave the way they do in all kinds of tests<br />If we wanted to model the boiling point of chemicals, we might start with a list of chemicals and their boiling points<br />In the following slide, chemicals were picked to give a wide range of boiling points<br />
Let’s Look at the Order in Boiling Point<br />Boiling Point (°C)<br />Chemical<br />Amyl amine<br />Amyl chloride<br />Dibromobenzene<br />Ethyl bromide<br />n-Heptanol<br />Methacrolein<br />Methyl-p-anisylketone<br />n-Octane<br />n-Nonane<br />103-4<br />98-9<br />219-2<br />38.4<br />192<br />68<br />267-9<br />126<br />151<br />
QSAR and Trend Analysis<br />If we reorder the chemicals to put the boiling points in ascending order, we have the next slide<br />The goal of QSAR would be to calculate molecular descriptors for each chemical and look for those giving the same order <br />If a molecular descriptor used in this canonical ordering produced an order different from that for measured boiling points, that descriptor would clearly not reflect the intra- and intermolecular forces controlling the boiling point <br />
Look for Order in Molecular Descriptors<br />Chemical<br />Boiling Point (°C)<br />Ethyl bromide<br />Methacrolein<br />Amyl chloride<br />Amyl amine<br />n-Octane<br />n-Nonane<br />n-Heptanol<br />Dibromobenzene<br />Methyl-p-anisylketone<br />38.4<br />68<br />98-9<br />103-4<br />126<br />151<br />192<br />219-2<br />267-9<br />
QSAR and Trend Analysis<br />In the next example, we might assume that we have compiled a set of toxicity values (LC50 in fish) for chemicals that are expected to be nonpolar narcotics<br />The actual toxicity values (chemical potency) for a single mechanism can range as much as six orders of magnitude<br />
10+2<br />10 0<br />10_2<br />Acute <br />Toxicity<br />(Moles/L)<br />It is not uncommon to find endpoint<br />values spanning 6-10 orders for a <br />single toxicity mechanism <br />10_4<br />10_6<br />10-8<br />1<br />2<br />3<br />4<br />5<br />Chemical Class or Mechanism<br />Exploring Mechanisms with Simple Endpoints<br />
QSAR and Trend Analysis<br />As with boiling points, the QSAR approach evaluates molecular descriptors and identifies those descriptors that yield the same order as the LC50 values would provide <br />In this example, Log Kow explains most of the variance, as a correlation (or trend) emerges between LC50 and Log Kow<br />With a good trend between the structure and biological activity for one toxicity mechanism, the similarity of other chemicals can be judged from consistency with the trend<br />As we shall see, this type of trend analysis is the centerpiece of defining and defending chemical categories<br />
Current Limitations in QSAR<br />The general QSAR approach is most reliable for chemicals where the parent chemical structure is the actual toxicant<br />One limitation has been predicting the effects of chemicals that are metabolically converted to more active (potent) metabolites<br />Predicting metabolic activation in many test species is a limitation being overcome with metabolic simulators (virtual livers, kidneys, skin, lung etc.)<br />Once the metabolites are predicted, the same library of toxicity models can be used on parent and metabolites to identify the most toxic form of the chemical<br />
Current Limitations in QSAR<br />The general QSAR approach works well for short-term bioassays where steady-state exposures are achieved<br />Long-term toxic effects, particularly low-incident effects such as cancer, can result from a chemical perturbation of biological functions but are influenced by many other biological factors as well<br />QSAR models for long-term effects will be limited until modeling both chemistry and disease progression over time can be integrated<br />One promising approach to overcome this limitation is the use of adverse outcome pathways<br />