From Biological Spectra (multiple protein binding data) to pharmacological profiling!

An ideal drug cures a decease and does not kill a patient (or even lab animals in the course of preclinical testing). Usual drug discovery paradigm is based on studying a compound's properties against a specific, normally decease-related (protein) target. The ability of a compound to bind (inhibit) a specific target is called efficacy.

Even if the efficacy is good, another important property of a compound is its toxicity. Toxicity is related to the compound physical properties, such as solubility etc, as well by its ability to bind to and hence inhibit various vital human proteins (and may be even DNA and RNA).

Common sense suggests that an ideal compound binds its specific drug related target and does not bind to anything else. Anything in between is toxic, at least to a some extent. For example, most of important properties utilize ATP molecules, which means that human body contains a lot of ATP-bindig proteins. If you make a drug attacking an ATP-binding site of a "bad" protein, most probably, a lot of "good" and useful proteins will be also affected. In that case your compound should be toxic. This is indeed the case for many cancer drugs attacking ATP-binding sites of kinases.

The latter statement is the foundation of our approach. Although it's quite conceptually simple, it's useless unless it can be supplemented by a meaningful mathematical model. Let us dwell into some more details to see how the whole thing can be made working.

Let us overview important properties of a drug candidate. First there is a bunch of physical properties, such as solubility, differential solubility, LogP (namely the difference between water and lipid solubility) etc. These quantities are easy to measure, are of direct physical meaning and can be pretty easily calculated (with or without QUANTUM software).

Another set of characteristics defines a compound ability to penetrate through cell membranes and its biochemical in liver. These are quantities deturmining bioavailability, half life, volume of distribution etc. None of such quantities can be evaluated using the simple physical properties alone. For example, drug absorbtion depends on the molecule interaction with proteins actively transporting the molecules through the cell membranes.

The bottom line: bioavailability and other quantities require understanding of a compound binding properties to a selected number of proteins participating in a compound transport and metabolism.

So the conclusion is that IF YOU KNOW WHICH PROTEINS ARE IMPORTANT, AND IF YOU CAN CALCULATE HOW YOUR COMPOUND BINDS TO THEM, YOU KNOW THE COMPOUND PHARMACOLOGICAL AND TOXICOLOGICAL PROPERTIES

Now the only problem how to identify those "important" proteins.

Fortunately, there are thousands of molecules with known properties. What we can do is the following:

- take a molecule
- calculate its binding to any human protein with known 3d structure
- use the obtained binding affinities (numbers) as a molecule's binding profile fingerprint (the Biological Spectrum), characterizing the ability of the molecule to interact with the whole human proteome

Now assume we know such Biological Spectra for 1000s molecules with well known properties. This means we can now datamine the fingerprints->known properites relations. The basic premise is, of course, that the molecules with similar fingerprints have similar properties.

We have a number of proofs of such technology working. The most recent one is the prediction of active transport drug absorption properties for drug like molecues based on the binding data against human brain hexokinase type I-related protein. We prove that the binding energy of a compound against the protein may serve to distinguish between the passively and actively transported molecules and even help to calculated the drug absorbtion quantitatevely.

From binding data to pharmacokinetics: a novel approach to active drug absorbtion prediction

Oral administered drugs are mainly absorbed in the small intestine. Here, depending on drug composition and size, absorption can happen through a variety of processes . Through the epithelial cells and the lamina pro- pria the drug passes from the lumen into the blood stream in the capillaries. On its way it might be metabolised, transported away from the tract where absorption is possible or accumulate in organs other than those of treatment. Besides a fundamental interest in understanding the basic mechanisms by which a drug is assimilated by the human body, the kinetics of drug absorption is also a topic of much practical interest. A detailed knowledge of this process, resulting in the prediction of the drug absorption profile, can be of much help in the drug development stage .

To this end, several kinetic models for drug absorption within the body have been introduced (see e.g. ). They necessarily introduce some simplifications belonging to the category of the so-called three-compartment models where the substances (such as drugs or nutrients) move between three volumes (e.g. the human organs). In fact the models require two kinds of molecular properties. First are purely physical characteristics, such as solubility, differential solubility, LogP etc. These quantities are easy to measure or to calcualte, have direct physical meaning and sufficient to predict absorbtion profile of passively absorbed drugs. Actively transported molecules interact with protein transporters and therefore prediction for actively transporting compounds require a lot of separate knowledge of binding to and kinetics of the transporting proteins.





The major objective of this investigation was to develop a drug absorbtion prediction approach based on entirely different paradigm, thus avoiding difficulties of both knowledge-based and QSAR-based models, and therefore capable of better predictions. Recently it was observed that experimental values of molecular activities against a large proteins set can be used for predicting broad biological effects . In this investigation we take advantage of this concept and develop a novel quntitative method for identification of actively transported drugs. To do that we performed a docking study of a few hundreds small molecules (mostly drugs) against a diversified 510 proteins set representing human proteom. Using available absorbtion data for each of the molecules we obtained a support vector classifier capable to identify proteins which affinity for drugs correlates well with the active absorption of these drugs in 81% cases. The observation helped us improve our passive absorbtion model by adding non-liner fluxes associated with the transporting protein to obtain also a quantitative model of the passively absorbed drugs.

Ref: arXiv:0810.2617 [ps, pdf, other]

Title: From protein binding to pharmacokinetics: a novel approach to active drug absorption prediction

Authors: P.O. Fedichev, T.V. Kolesnikova, A.A. Vinnik

Comments: 9 pages, 5 eps figures

Subjects: Quantitative Methods (q-bio.QM); Biomolecules (q-bio.BM)

Drug likeness: what do bioavailability and toxicity properties tell us about druglikeness?

Druglikeness is a qualitative concept used in drug design for an estimate on how "druglike" a prospective compound is. Usually it is estimated from the molecular structure, often even before the substance is synthesized and tested.

A good drug should show good availability, low toxicity and high potency. The quantitative measures of such properties are bioavailability (BA, measured in %), Maximum Recomended Daily Dose (MRDD, mmol/L) and IC50 against a drug's target.

The product of toxicity and availability, MRDD*BA, gives an upper bound on target IC50 and hence is an indication of a drug quality. The Figure above represents the distribution of such product for slightly over 100 drugs. As it can be seen from the Graph, most of drug compounds have the product small, roughly below 2*10^-5mol/L. Hence, small value of MRDD*BA product may be regarded as an indication of druglikeness.

In fact the situation gets even more interesting if the same druglikeness parameter is plotted in log-scale (see the Figure on the right). Since MRDD*BA limits drug's IC50 against its target, we can deduce that most drugs are centered around pIC50 = 5 (which means that the target pIC50 should exceed 5).