Non-linear modelling of biological activity of chemical compounds
AssigneeCamitro Corporation
InventorTodd J.A. Ewing
Life of this patent
Estimating the term…
Abstract
Models predict activity of chemical compounds by employing “transformed” descriptor values. The descriptors are transformed via transformation functions that convert the raw descriptor values to new values better representing the contribution of the descriptors to the activity in question. Typically, these transformation functions are non-linear parametric functions such as Gaussian functions or sigmoid functions. Typically, the model will employ at least two different descriptors, each transformed by its own non-linear parametric function.
Claims104
A method of predicting the activity of a compound, the method comprising:
- (a) providing a model for predicting activity of compounds based on one or more descriptors of physicochemical properties of compounds;
- (b) characterizing the compound by one or more descriptor values;
- (c) transforming the one or more descriptor values via one or more non-linear parametric functions, each of which provides an analytical relationship between the activity and the corresponding descriptor value;
- (d) using the model to predict the activity of the compound from the transformed descriptors.
A computer program product comprising a machine readable medium on which is provided program instructions for predicting the activity of a compound, the program instructions comprising instructions for
- (a) providing a model for predicting activity of compounds based on one or more descriptors of physicochemical properties of compounds;
- (b) characterizing the compound by one or more descriptor values;
- (c) transforming the one or more descriptor values via one or more non-linear parametric functions, each of which provides an analytical relationship between the activity and the corresponding descriptor value;
- (d) using the model to predict the activity of the compound from the transformed descriptors.
A method of predicting the activity of a compound, the method comprising:
- (a) providing a model for predicting activity of compounds based on two or more descriptors of physicochemical properties of compounds, wherein the model comprises (i) a first non-linear transformation function which transforms values of a first descriptor and provides an analytical relationship between the activity and the first descriptor and (ii) a second non-linear transformation function which transforms values of a second descriptor and provides an analytical relationship between the activity and the second descriptor;
- (b) characterizing the compound by a first descriptor value and a second descriptor value;
- and
- (c) predicting the activity of the compound by using the first and second descriptor values as arguments in the model.
A computer program product comprising a machine readable medium on which is provided program instructions for predicting the activity of a compound, the program instructions comprising the following instructions:
- (a) providing a model for predicting activity of compounds based on two or more descriptors of physicochemical properties of compounds, wherein the model comprises (i) a first non-linear transformation function which transforms values of a first descriptor and provides an analytical relationship between the activity and the first descriptor and (ii) a second non-linear transformation function which transforms values of a second descriptor and provides an analytical relationship between the activity and the second descriptor;
- (b) characterizing the compound by a first descriptor value and a second descriptor value;
- and
- (c) predicting the activity of the compound by using the first and second descriptor values as arguments in the model.
A method of predicting the activity of a compound, the method comprising:
- (a) providing a model for predicting activity of compounds based on two or more descriptors of physicochemical properties of compounds;
- (b) characterizing the compound by the two or more descriptor values;
- (c) generating one or more orthogonal, linear transformations of the descriptor values;
- (d) transforming the linear transforms via one or more non-linear parametric functions, each of which provides an analytical relationship between the activity and the corresponding linear transform;
- (e) using the model to predict the activity of the compound from the linearly transformed descriptors and subsequent non-linear transforms of the linearly transformed descriptors.
A method of creating a multivariate model for predicting the activity of compounds, the method comprising:
- (a) for each compound for a plurality of compounds, obtaining an indication of the activity of the compound and values of one or more descriptors of the compound;
- (b) for each of the descriptors, using activity versus descriptor data for the plurality of compounds to identity a non-linear parametric function representing an analytical relationship between the activity and the descriptor data;
- (c) for each of the descriptors, transforming the descriptor via the non-linear parametric function;
- and
- (d) creating the multivariate model of activity as a function of the descriptors and the non-linear parametric function transforms of the descriptors.
A method of creating a model for predicting the activity of compounds, the method comprising:
- (a) for each compound for a plurality of compounds, obtaining an indication of the activity of the compound and values of two or more descriptors of the compound;
- and
- (b) creating the multivariate model of activity as a function of the first and second descriptors by using the values of the two or more descriptors and the indication of the activity for each of the plurality of compounds and fitting the descriptors and indications of activity to two or more non-linear parametric functions of the descriptors, which non-linear parametric functions are present in the model, wherein each of the non-linear parametric functions provides an analytical relationship between the activity and the corresponding descriptor value.
A method of creating a model for predicting the activity of a compound, the method comprising:
- (a) for each compound for a plurality of compounds, obtaining an indication of the activity of the compound and values of two or more descriptors of the compound;
- (b) generating one or more orthogonal, linear transformations of the descriptor values;
- (c) for each of the linear descriptor transforms, using activity data versus the linear transform for the plurality of compounds to identify a non-linear parametric function, which provides an analytical relationship between the activity and the corresponding linear descriptor transform;
- (d) for each of the linear descriptor transforms, transforming it via the non-linear parametric function;
- and
- (e) creating the multivariate model of activity as a function of the linearly transformed descriptors and the subsequent non-linear transforms of these linearly transformed descriptors.
A method of creating a multivariate model for predicting the activity of compounds, the model including at least one non-linear parametric transformation function that transforms a descriptor of the compounds, the method comprising:
- (a) fitting activity versus descriptor data for a training set of compounds using an optimization function comprising (i) a penalty term that drives a parameter of the transformation function toward a boundary value, and (ii) an error term that compares a predicted activity with an actual activity for members of the training set;
- (b) identifying one or more descriptors whose parameters were driven toward the boundary value by the penalty function to values below a threshold of significance to the model;
- (c) eliminating the one or more descriptors identified in (b) from use with the multivariate model.
A computer program product comprising a machine readable medium on which is provided program instructions for creating a multivariate model for predicting the activity of compounds, the model including at least one nonlinear parametric transformation function that transforms a descriptor of the compounds, the program instructions comprising instructions for
- (a) fitting activity versus descriptor data for a training set of compounds using an optimization function comprising (i) a penalty term that drives a parameter of the transformation function toward a boundary value, and (ii) an error term that compares a predicted activity with an actual activity for members of the training set;
- (b) identifying one or more descriptors whose parameters were driven toward the boundary value by the penalty function to values below a threshold of significance to the model;
- (c) eliminating the one or more descriptors identified in (b) from use with the multivariate model.
Dependencies are inferred from the claim text.