Data processing: artificial intelligence – Neural network – Learning task
Reexamination Certificate
2000-02-15
2003-07-01
Voeltz, Emanuel Todd (Department: 2121)
Data processing: artificial intelligence
Neural network
Learning task
C706S020000
Reexamination Certificate
active
06587845
ABSTRACT:
FIELD OF THE INVENTION
The invention pertains to the field of using computational methods in predictive chemistry. More particularly, the invention utilizes a neural network with associated algorithmic functions, and the quantum mechanical properties of the molecules investigated or portions of those molecules, to optimize the prediction of bioactivity or the mode of chemotherapeutic action for molecules of interest.
BACKGROUND OF THE INVENTION
The role of medicinal chemist has not been altered in several decades. Their efforts to identify chemotherapeutic compounds, and thereafter devise more potent variations to them for medicinal use has long been one involving the arduous task of testing one compound at a time to determine individual bioactivity. This slow throughput system is made even more costly by the fact that historically over 10,000 compounds must be individually tested and evaluated for every one that actually reaches market as a therapeutic agent, as discussed in SCRIP, World Pharmaceutical News, Jan. 9, 1996, (PJB Publications). These facts have driven many scientists and pharmaceutical houses to shift their research from traditional drug discovery (e.g. individual evaluation) towards the development of high throughput systems (HTP) or computational methods that will bring to bear increasingly powerfull computer technology for the drug discovery process. To date none of these systems have been proven to significantly shorten discovery and optimization time for the development of chemotherapeutic agents.
The first attempts to develop computational methods to predict the inhibitory potency of a given molecule prior to synthesis have been broadly termed quantitative structure activity relationship (QSAR) studies. These techniques require the user to define a functional relationship between a specific molecular property and a given molecular action. In the QSAR approach, or any approach where an individual is responsible for adjusting a mathematical model, the investigator must use variations in the structure of a molecule as the motivation for changing the value of coefficients in the computer model. For a chemical reaction as complex as an enzymatically mediated transformation of reactants to product, often an important part of therapeutic or medicinal activity, it is not possible to predict a priori all the effects a change to a substrate molecule will have on enzymatic action. This fact has made the QSAR approach to drug discovery exceptionally impracticable and inefficient.
Accordingly, a need exists to optimize the prediction of bioactivity in chemical compounds such that the discovery and development of therapeutically valuable compounds is made more rapid and efficient.
SUMMARY OF THE INVENTION
Briefly stated, the invention described herein provides a neural network approach to the prediction of chemical activity in at least one molecule of interest. The example provided herein demonstrates how this methodology is useful in the prediction of bioactivity for a molecule capable of acting as an enzymatic inhibitor. This same methodology is also applicable to a variety of compounds of interest using the same training protocols and the same quantum mechanical properties of given molecules, or portions thereof discussed herein.
The neural network provided herein is comprised of an input layer having at least one neuron where input data is sent and then given a vector value, a hidden layer having at least one neuron such that when data is received from the input layer that vector data is multiplied by a set weight and thereafter generates a weight matrix having the dimensions n by m where n is the length of an input vector and m is the number of hidden layer neurons available, and an output layer consisting of at least one neuron where the weight matrix data is sent before it is then sent to a transfer function. The transfer function is a non-linear equation that is capable of taking any value generated by the output layer and returning a number between −1 and 1.
Feed-forward neural networks with back-propagation of error, of the type disclosed herein (see pages 7-10), are trained to recognize the quantum mechanical electrostatic potential and geometry at the entire van der Waals surface of a group of training molecules and to predict the strength of interactions, or free energy of binding, between an enzyme and novel inhibitors of that enzyme. More generally, the input for the functions of the neural network are the quantum mechanical electrostatic potentials of various molecules of interest. The predictive value of the system is gained through the use of a “training” process for the neural network using the known physicochemical properties of at least one training molecule, such as Inosine-Uridine Preferring Nucleoside Hydrolase (IU-NH). IU-NH is a nucleoside hydrolase from first isolated from the organism
Crithidia fasciculata
. The neural network is given input generated from the known electrostatic potential surfaces of the known molecules and attempts to predict the free energy of binding for that training set of molecules. When the neural network is able to accurately predict the free energy of binding of the training set of molecules, then the same neural network can be used with high accuracy to determine the free energy of binding, and hence the chemical characteristics, of unknown molecules.
Among the novel aspects of the present invention is the utilization in the current invention of the quantum mechanical electrostatic potential of the molecule of interest at the van der Waals surface of that molecule as the physicochemical descriptor. The entire surface for each molecule, represented by a discrete collection of points, serves as the input to the neural network. In this way the invention utilizes quantum mechanical means to describe the molecular activity of a compound of interest. With improved knowledge of molecular activity the method described herein provides for enhancing the predictive value of neural networks with regard to phyisicochemical properties of compounds of interest either with regard to therapeutic compounds or compounds that would have other commercial or scientific value. The neural networks provided herein are useful in modeling chemical interactions that are non-covalent in nature. That is, as long as a reaction is mediated by electrostatic forces, including Van der Waals forces, the neural networks provided herein, and the associated algorithms, are accurate predictors of chemical activity and interaction. In this way they will save time and money in drug discovery and chemical evaluation processes. Specifically, with regard to enzymatic action, the neural networks herein described are able to determine chemical configurations that will optimize known chemotherapeutics and allow the discovery of new compounds that need to have specific binding characteristics or activity. These new compounds can be developed by modeling the quantum characteristics of specific molecular moieties with a trained neural or double neural network.
According to an exemplary embodiment of the invention, a computational method has been developed to predict the free energy of binding for inhibitor or untested enzyme molecules.
REFERENCES:
patent: 4461619 (1984-07-01), Hendry et al.
patent: 5214715 (1993-05-01), Carpenter et al.
patent: 5631469 (1997-05-01), Carrieri et al.
patent: 5727128 (1998-03-01), Morrison
patent: 5825645 (1998-10-01), Konar et al.
patent: 5933819 (1999-08-01), Skolnick et al.
patent: 6119111 (2000-09-01), Gross et al.
patent: 6185548 (2001-02-01), Schwartz et al.
patent: 6289328 (2001-09-01), Shaffer
patent: 6295514 (2001-09-01), Agrafiotis et al.
patent: 6314414 (2001-11-01), Keeler et al.
Su et al.; “Neural Model Predictive Control of Nonlinear Chemical Processes”. Proceedings of the 1993 IEEE International Symposium on Intelligent Control, Aug. 1993, p. 358-363.*
Lambert et al.; “Application of Feedforward and Recurrent Neural Networks to Chemical Plant Predictive Modeling”. IJCNN-91-Seattle International Joint Conferenc
Booker Kelvin
Braunheim Benjamin B.
Ganz Law PC
Ganz, Esq. Bradley M.
Todd Voeltz Emanuel
LandOfFree
Method and apparatus for identification and optimization of... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for identification and optimization of..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for identification and optimization of... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3059638