Molecular property modeling using ranking
View Patent ↗Methods and articles of manufacture for modeling molecular properties using data regarding the partial orderings of compound properties, or by considering measurements of compound properties in terms of partial orderings are disclosed. One embodiment provides for constructing such partial orderings from data that is not already in an ordered form by processing training data to produce a partial ordering of the compounds with respect to a property of interest. Another embodiment of the invention may process the modified training data to construct a model that predicts the property of interest for arbitrary compounds.
1. A method for training a molecular properties model, comprising:
ranking a training set of property measurements for a plurality of molecules, wherein each measurement assigns a value for a property of interest relative to a single molecule;
generating a representation of the molecules included in the training set that is appropriate for a selected machine learning algorithm;
providing the training set to a selected machine learning algorithm that optimizes a function of the rank order of the molecules in the training set relative to the property of interest, wherein, for at least two molecules in said training set, said function penalizes incorrectly ordered molecules when said function is evaluated on said at least two molecules;
executing on a computer system the machine learning algorithm to generate a trained molecular properties model;
selecting at least one additional molecule;
generating a representation of the at least one additional molecule appropriate for the molecular properties model;
generating on a computer system and with the molecular properties model a prediction about the at least one additional molecule regarding the property of interest; and
determining the accuracy of the prediction for the at least one additional molecule by performing laboratory experimentation using a physically realized sample of the test molecule.
2. The method of claim 1 , wherein the molecular properties model generates predictions related to a property of interest selected from at least one of a pharmacokinetic property, pharmacodynamic property, physiological activity, pharmacological activity, toxicity, selectivity, binding affinity, pKa, a property of a specific atom or bond in a molecule, melting point, solubility, a membrane permeability, or a force-field parameter.
3. The method of claim 1 , wherein the selected machine learning algorithm comprises a Boosting algorithm, RankBoost algorithm, Alternating Decision Trees algorithm, Support Vector Machines algorithm, a Perceptron algorithm, Winnow, a Hedge Algorithm, decision trees, neural networks, genetic algorithms or genetic programming.
4. The method of claim 1 , wherein the selected machine learning algorithm is configured to minimize, either directly or indirectly, an area above, or below, a receiver operator characteristic curve.
5. The method of claim 1 , further comprising, determining an accuracy of the prediction for the additional molecule by performing a research study using physical samples of the additional molecule.
6. The method of claim 1 , wherein generating the representation of the molecules included in the training set comprises: generating a vector representation of the molecules, wherein the vector representation is configured to encode the structure the molecules included the training set; or comprises generating an n-point pharmacophore representation of the molecules included in the training set.
7. The method of claim 1 , wherein a threshold value or cutoff molecule is selected for the molecular properties model.
8. The method of claim 1 , wherein the step of generating a representation of the at least one additional molecules comprises generating a representation of at least two or more additional molecules, and wherein the prediction comprises a ranked ordering of the two or more additional molecules, relative to one another and to the property of interest.
9. The method of claim 1 , wherein at least two of the molecules in the training set are alternative representations of the same physical molecule and the property of interest is a property of said alternative representations.
10. A physical computer-storage medium containing a program which, when executed by a processor, performs a method for training a molecular properties model, comprising:
receiving a ranked training set of property measurements for a plurality of molecules, wherein each measurement assigns a value for a property of interest relative to a single molecule;
generating a representation of the molecules included in the training set that is appropriate for a selected machine learning algorithm;
providing the training set to a selected machine learning algorithm that optimizes a function of the rank order of the molecules in the training set relative to the property of interest, wherein, for at least two molecules in said training set, said function penalizes incorrectly ordered molecules when said function is evaluated on said at least two molecules;
executing the machine learning algorithm to generate a trained molecular properties model;
selecting at least one additional molecule;
generating a representation of the at least one additional molecule appropriate for the molecular properties model; and
generating with the molecular properties model a prediction about the at least one additional molecule regarding the property of interest.
11. The method of claim 1 , comprising the additional step that the accuracy of the prediction for the at least one additional molecule is determined by performing a computational simulation configured to model the property of interest for the additional molecule.
12. The method of claim 1 , further comprising using the predictions regarding the additional molecules to identify at least one molecule; and physically realizing said molecule.
13. The method of claim 1 , wherein the set of property measurements include virtual data.
14. The method of claim 1 , wherein the plurality of molecules include virtual molecule(s).
15. The method of claim 1 , wherein the at least one additional molecule includes at least one virtual molecule.
16. The method of claim 1 , wherein molecules in the training set are weighted, and the machine learning algorithm is configured to minimize a function of the weighted rank order of the molecules.
17. The method of claim 1 , wherein the providing the training set to a selected machine learning algorithm comprises constructing a pseudo-partial ordering of molecules, wherein the pseudo-partial ordering includes at least a representation of a first and second molecule, ordered relative to one another and the property of interest.
18. The method of claim 17 , wherein the molecular properties model generates predictions related to a property of interest selected from at least one of a pharmacokinetic property, pharmacodynamic property, physiological activity, pharmacological activity, toxicity, selectivity, binding affinity, pKa, a property of a specific atom or bond in a molecule, melting point, solubility, a membrane permeability, or a force-field parameter.
19. The method of claim 17 , wherein the selected machine learning algorithm comprises a classification learning algorithm.
20. The method of claim 17 , wherein the selected machine learning algorithm comprises a kernel based learning algorithm.
21. The method of claim 17 , wherein the selected machine learning algorithm comprises a Boosting algorithm, RankBoost algorithm, Alternating Decision Trees algorithm, Support Vector Machines algorithm, a Perceptron algorithm, Winnow, a Hedge Algorithm, decision trees, neural networks, genetic algorithms or genetic programming.
22. The method of claim 17 , wherein the selected machine learning algorithm is configured to minimize, either directly or indirectly, an area above, or below, a receiver operator characteristic curve.
23. The method of claim 17 , wherein the selected machine learning algorithm is configured to minimize, either directly or indirectly, a function of the rank ordering of molecules in the pseudo partial ordering.
24. The method of claim 17 , further comprising, determining an accuracy of the prediction for the additional molecule by performing a research study using physical samples of the additional molecule.
25. The method of claim 17 , wherein constructing the representation of the molecules included in the pseudo partial ordering of molecules comprises: generating a vector representation of the molecules, wherein the vector representation is configured to encode the structure of the molecules included the pseudo-partial ordering; or comprises generating an n-point pharmacophore representation of the molecules included in the pseudo-partial ordering.
26. The method of claim 17 , wherein a threshold value or cutoff molecule is selected for the molecular proprieties model.
27. The method of claim 17 , wherein the at least one additional molecule comprises two or more additional molecules, and wherein the prediction comprises a ranked ordering of the two or more additional molecules, relative to one another and to the property of interest.
28. The method of claim 17 , wherein at least two of the molecules in the training set are alternative representations of the same physical molecule and the property of interest is a property of the alternative representations.