( 1 of 1 ) |
United States Patent | 5,526,281 |
Chapman ,   et al. | June 11, 1996 |
Explicit representation of molecular shape of molecules is combined with neural network learning methods to provide models with high predictive ability that generalize to different chemical classes where structurally diverse molecules exhibiting similar surface characteristics are treated as similar. A new machine-learning methodology that can accept multiple representations of objects and construct models that predict characteristics of those objects. An extension of this methodology can be applied in cases where the representations of the objects are determined by a set of adjustable parameters. An iterative process applies intermediate models to generate new representations of the objects by adjusting said parameters and repeatedly retrains the models to obtain better predictive models. This method can be applied to molecules because each molecule can have many orientations and conformations (representations) that are determined by a set of translation, rotation and torsion angle parameters.
Inventors: | Chapman; David (San Francisco, CA); Critchlow; Roger (San Francisco, CA); Jain; Ajay N. (San Carlos, CA); Lathrop; Rick (Cambridge, MA); Perez; Tomas L. (Cambridge, MA); Dietterich; Tom (Corvalis, OR) |
Assignee: | Arris Pharmaceutical Corporation (South San Francisco, CA) |
Appl. No.: | 382990 |
Filed: | October 28, 1994 |
U.S. Class: | 364/496; 364/578; 395/920 |
Intern'l Class: | G06F 017/00; G06F 015/18 |
Field of Search: | 364/496,497,578 395/920,924,21,22,23 |
5025388 | Jun., 1991 | Cramer, III et al. | 364/496. |
5167009 | Nov., 1992 | Skeririk | 395/27. |
5260882 | Nov., 1993 | Blanco et al. | 364/499. |
5265030 | Nov., 1993 | Skolnick et al. | 364/496. |
T. A. Andrea, et al., Applications of Neural Networks in Quantitative Structure-Activity Relationship of Dihydrofolate Reductase Inhibitors, Journal of Medicinal Chemistry, vol. 34, No, 9, 1991, pp. 2824-2836. Good, et al., Structure-Activity Relationships from Molecular Similarity Matrices, Journal of Medicinal Chemistry, vol. 36, No. 4, Feb. 19, 1993, pp. 433-438. Aoyama, et al., Neural Networks Applied to Quantitative Structure-Activity Relationship Analysis, Journal of Medicinal Chemistry, 1990, vol. 33, No. 9, pp. 2583-2590. Aoyama, et al., Obtaining the Correlation Indices Between Drug Activity and Structural Parameters Using a Neural Network, Chemical Pharmaceutical Bulletin, vol. 39, No. 2, pp. 372-378, 1991. Oinuma, et al., Neural Networks Applied to Structure-Activity Relationships, Journal of Medicinal Chemistry, vol. 33, No. 3, 1990, pp. 905-908. Viswanadhan, et al., Mapping the Binding Site of the Nucleoside Transporter Protein: A 3D-QSAR Study, Biochemica et Biophysica Acta., 1039 (1990) 356-366. Ghose, et al., Use of Physicochemical Parameters in Distance Geometry and Related Three-Dimensional Quantitative Structure-Activity Relationships: A Demonstration Using Escherichia coli Dihydrofolate Reductase Inhibitors, Journal of Medical Chemistry, vol. 28, No. 3, 1985, pp. 333-346. Smellie, et al., Fast Drug-Receptor Mapping by Site-Directed Distances: A Novel Method of Predicting New Pharmacological Leads, J. Chem. Inf. Comput. Sci., vol. 31, No. 3, 1991, pp. 386-392. Cramer, III, et al., Comparative Molecular Field Analysis (COMFA), J. Am. Chem. Soc., vol. 110, No. 18, 1988, pp. 5959-5967. Crippen, et al., Distance Geometry and Molecular Conformation, Research Studies Press Ltd., 8, Ligand Binding, pp. 361-429. Hopfinger, A QSAR Investigation of Dihydrofolate Reductase Inhibition By Baker Triazines Based Upon Molecular Shape Analysis, J. Am. Chem. Soc., 1980, 102, 7196-7206. Doweyko, The Hypothetical Active Site Lattice. An Approach to Modelling Active Sites from Data on Inhibitor Molecules, Journal of American Chemistry, 1988, 31 (7), pp. 1396-1406. Simon, et al., Mapping of Dihydrofolate-Reductase Receptor Site by Correlation with Minimal Topological (Steric) Differences, J. Theor. Biol., 1977, 66, pp. 485-495. Crippen, Deduction of Binding Site Structure from Ligand Binding Data, Annals New York Academy of Sciences, vol. 439, 1984, pp. 1-11. Boulu, et al., Voronoi Binding Site Models: Calculation of Binding Modes and Influence of Drug Binding Data Accuracy, Journal of Computational Chemistry, 10 (5), 673-632 (1989). Crippen, Voronoi Binding Site Models, Journal of Computational Chemistry 8 (7), 943-955 (1987). Boulu, et al., Voronoi Binding Site Model of a Polycyclic Aromatic Hydrocarbon Binding Protein, J. Med. Chem., 1990, 33, 771-775. Blankley, Introduction: A Review of QSAR Methodology, Academic Press, Inc., 1983, pp. 1-17. Nicklaus, et al., QSAR of Conformationally Flexible Molecules: Comparative Molecular Field Analysis of Protein-Tyrosine Kinase Inhibitors, Journal of Computer-Aided Molecular Design, 6 (1992) 487-504. |
TABLE 1 ______________________________________ Predictive accuracy of musk model in a 20-fold cross-validation hold-out test (standard error is in brackets). True False True False Pos. Neg. Neg. Pos. % Correct ______________________________________ Adaptive alignment 36 3 57 6 91[2.8] Fixed alignment 36 3 47 16 81[3.9] ______________________________________
TABLE 2 __________________________________________________________________________ Predictive accuracy of musk model across structural classes. Numbers in brackets are standard error. The counts reported in rows 2-4 are for adaptive alignment. Structural (1)4-substituted (2)1- (3)6-substituted Class: dihydroindanes indanones tetrahydronapthalenes (4) benzopyrans __________________________________________________________________________ Number of molecules 13 21 27 14 True positives 7 6 9 4 False negatives 0 0 4 3 True negatives 6 13 13 5 False positives 0 2 0 1 Percent correct 100[0.0] 90[6.5] 85[6.8] 71[12.1] (adaptive alignment) Percent correct 85[9.9] 76[9.3] 74[8.4] 57[13.2] (fixed alignment) Number misranked 0 1 1 2 Percent correct 100[0.0] 95[4.8] 96[3.8] 86[9.3] (by ranking) __________________________________________________________________________