Yield Control in Drug Manufacturing

Summary:

There are two methods in drug production: fermentation and chemical synthesis. In the synthetic method, the process starts from some simple materials (raw materials), and it takes many steps (each being a chemical reaction with added chemical compound) in a "synthesis flow" represented by a flow chart. Since this flow is usually very long, the overall yield is rather low. For instance, if the yield is 90% at each step, the overall yield is reduced to 81% (0.9x0.9) in 2 steps respectively. Therefore, it is very important to find a method to optimize the yield by increasing the yield of each step in a synthetic drug production.

Zaptron’s techniques and software have been proved very powerful in controlling and increasing production yield in drug manufacturing.

Feature Selection:

How to select and use the design parameters to describe the molecular structure of a drug is a complicated problem. People use quantum chemical calculation to find the partial charge values on different atoms, or to calculate the electric field, represented by a graph of contours, around drug molecules. People also use molecular mechanics to calculate a few dynamic parameters of drug molecules. Still some people use semi-empirical methods to find a few parameters to describe the electronic or geometric structure of drug molecules.

Up to now, features in drug designs are extracted by human intelligence For example, some medical chemists believe that for some molecules the structure of "three oxygen atoms spanned by a certain angle" favors the bio-activity of some anti-tumor drugs. Of course, this is found by human brain, not by computers. Molecular structure is very complicated and it is often described by a large number of design parameters. The difficulty is how to choose the right set, or a reduced set, of design parameters that correlate the bio-activity with the structure of drug molecules for the best result. It seems that such empirical rules can also be discovered by the Process Master software that has implemented a few powerful methods for feature selection and feature reduction purpose.

The separability criteria, implemented in Zaptron’s Process Master software, are rather useful in selecting key factors that influence the bio-activity of a drug. People often use nonlinear regression in drug design. Zaptron’s data mining software has been proved by real-world examples to be very useful in simplifying the selection of nonlinear terms in regression. It has been compared favorably against the popular JMP and RS-1 software. The former uses far less terms in mathematical models, and produces lower prediction error (PRESS).

Traditional Methods:

As part of informatics, chemometrics and phamakocinetics, traditional techniques in pattern recognition, such as linear and nonlinear regression, PLS and ANN, have been widely applied to materials and drugs design for many years. In pattern recognition applications, PLS (partial least square) method are usually used to find quantitative structure-activity relationships. However non-linearity exists among target and factors, and PLS often fails to give meaningful results.

Yield – is defined as the productivity (amount of product) per unit of raw materials.

Manufacturing Productivity – is defined as the amount of product in each batch of production (fermentation).

Zaptron’s Methods:

Zaptron’s MasterMiner™ software has a number of major advantages over those based on pure ANN or pure regression (such as PCA) that are currently used in drug designs. In addition, the separability test by MasterMiner™ on data and classification on two classes of compounds (bio-active and bio-inactive)   has also been proved to be very effective in practice.

Intelligent Drug Designs

Design Issues (targets):

The major issue (target) in drug design is to discover quantitative relationships between a drug’s molecular structure, the way drug molecules are arranged, and its bio-activity, the effect of medicine or toxicity. It is also desirable to find relationships in molecular-molecular interactions, for example, the drug molecule-DNA molecule interaction.

The bio-activity of a drug is experimentally measured by conducting animal tests. The molecular structure is usually described by a number of design parameters or features (factors), such as the electrostatic field around the drug molecule, the electric charge of different parts of the molecule, the arrangement of water molecules around the molecule, the relative geometric relation with DNA, and so on.

Another very important issue in drug design is related to the diffusion capability of the molecules of a drug in the human body, where the inner part of each cell is like a water solution, and the wall of each cell, made of fat, has the nature of "oil". A drug can diffuse quickly if it has suitable solubility both in "water" and in "oil". This issue is usually expressed by the octanol-water distribution of drug molecules.

Sometimes we have no experimental data about this distribution, but want to use the molecular structure of a drug to predict the octanol-water distribution by a number of distribution coefficients between octanol and water. How to effectively calculate these coefficients is an important task. Therefore, the relationship between the molecular structure and solubility of a drug is also a target in drug research.

Distribution Coefficient

Oil (fat) and water cannot dissolve each other and they form two distinct layers. If a third matter (substance) is added to this "two layer" system, it will be partially dissolved in oil and partially dissolved in water. The ratio of the concentration of this third matter in oil to that in water is called "coefficient of distribution" or "distribution coefficient". If a matter can dissolve in water but not in oil, the distribution coefficient is 0, If it dissolves in oil only, the coefficient will be infinity. Since the human body is a mixture of water and fat, the drug diffusion process in human body will depend on the distribution coefficient of this drug in water and in oil. It should be not too large or too small, otherwise its diffusion will be prevented by water or by oil. Since octanol is "oil-like", drug scientists use the distribution coefficient in water and in octanol to correlate the diffusion ability of a drug in human body. For instance, if a drug can kill bacteria very effectively, but it cannot reach them in human body, it is of no use. Therefore, this is a very important factor in drug designs.

Zaptron’s Solutions to Drug Designs:

Zaptron’s techniques and software have been proved valuable in several kinds of drug designs.

Real-World Examples

back to Home

Email to info@zaptron.com.
© 1998-2000 ZAPTRON Systems, Inc.