User:WeepingBritney/Drug design prof review

Here's my feedback. Generally the information in that article is correct and easy to read and has an organization that makes sense. What I don't like about it is:


 * there are no examples shown in some detail, or figures to accompany the material


 * the coverage is quite superficial, and only very cursorily covers one kind of drug design without mentioning other kinds, and does not put the "design" approaches in context of how they are used with experimental approaches. I think the Kuntz paper I sent you does a great job of this.


 * who wrote the article and who is responsible for keeping it up to date? Anonymity isn't good in science because it suggests that no one is taking responsibility for the work and there is no one to contact when there are problems that need to be fixed.


 * similarly, very little of others' word is attributed by references, and there are only two references altogether, which wouldn't provide much of a handle on the field.

As a mini-overview, drug design most often refers to the process of taking a preexisting molecule known to bind to a protein (either a known substrate, or another molecule known to inhibit the protein, say, from screening a large number of compounds experimentally) and improving that compound. Sometimes the improvement is made by comparing a number of known ligands (protein binding molecules), considering which ones bind more tightly and specifically (from experimental results, generally), and mixing and matching the chemical features from them into new designs by using molecular modeling software. These designs are often intended to test hypotheses about which chemical groups are important for binding, so many such compounds would be made and tested, ideally. This can be done in the absence of having a 3-dimensional structure of the protein bound to the inhibitor(s), but obviously having such a structure gives you much more information about the interactions between the protein and ligand, and how they could be improved. Thus, all the large pharmaceutical companies will solve the 3-dimensional structure of the protein (if not yet available publicly) whenever possible. This is not possible yet for most membrane proteins and for a significant subset of soluble proteins, due to experimental difficulties. The structures of some proteins can be modeled based on known structures of close relatives; this is an area in which Dr. Wedemeyer is expert.

Aside from drug design involving optimizing known inhibitor structures (described above), or changing substrates structures so that act as inhibitors, instead, some drug design uses computational software to screen many candidate small organic molecules (100,000's or millions, ideally) and identify that subset which is likely to fit well and make good interactions with the site that should be blocked (inhibited) on the protein. Such "virtual screening" or high-throughput docking, is the area covered by the two papers I sent you. This is a field in which my research group works, and some good things about this approach are that it can discover entirely new classes of molecules that inhibit the protein, and that it provides a predicted structure for each molecule that can bind to the protein. The down side is that the software cannot yet take into account the full flexibility of the protein and each small molecule (due to the computational time that would be involved) and it must also make some assumptions about what kinds of interactions are or are not important in the "scoring function" that assesses the quality of fit between each inhibitor candidate and the protein structure. There tends to be a high false positive rate; typically 80% of molecules identified as good candidates in the top 100 scoring small molecules that were screened turn out to either be insoluble or do not bind the protein, based on follow-up experimental assays. However, the bright side of this picture is that the success rate of experimental screens of large sets of compounds is typically <1% (rather than 20% for computational screening), and this is enormously more expensive, about $1 million for a typical large compound library screen against a protein target.

The other point I wanted to mention is that in virtually all drug discovery projects in major pharmaceutical companies, there is a combination of experimental and computational approaches for discovery, assessment of specificity and affinity, optimization of the compounds, and creating back-up compounds for the drug discovery pipeline. Computational and experiment are not used independently, but together. Many features of drugs, like absorption into the body, excretion of them or their metabolites, metabolism of drugs within the body, and toxic side effects, cannot yet be well predicted based on structures of the proteins and the inhibitors selected as drug candidates. Currently assessment of these bioactivity factors involves a major amount of pharmacological follow-up work, and the vast majority of inhibitor candidates fail at these stages.