Prediction of phenotypes using recommender systems

A non-transitory computer readable medium configured to store instructions, the instructions, when executed by one or more processors, cause the one or more processors to perform steps comprising:

• retrieving survey responses of a set of individuals regarding a set of phenotypes of the individuals, the set of individuals comprising a target individual;
• constructing a matrix that arranges the set of individuals in a first dimension of the matrix and the set of phenotypes in a second dimension of the matrix, at least a subset of elements in the matrix being numerical representations of the individuals' phenotypes obtained from the survey responses;
• identifying one or more undetermined phenotypes of the target individual, the matrix having no value at particular elements corresponding to the one or more undetermined phenotypes of the target individual;
• determining a sparsity of the matrix, the sparsity of the matrix corresponding to a number of undetermined values in the matrix;
• determining, responsive to the sparsity higher than a threshold, at least one of the undetermined values corresponding to a particular individual based on corresponding values of other individuals who are similar to the particular individual to reduce the sparsity of the matrix;
• and
• determining a prediction of the one or more undetermined phenotypes of the target individual by performing collaborative filtering using a machine learning technique, the collaborative filtering based on other phenotypes of the target individual and based on at least another individual's phenotypes as represented in numerical representations in the matrix, performing the collaborative filtering using the machine learning technique comprises:
• performing a factorization of the matrix to estimate two or more lower dimensionality matrices that generates an approximation of the matrix, and
• using coordinate descent to iteratively adjust values in elements of the two or more lower dimensionality matrices to reduce differences between the approximation and the matrix.

A system comprising:

• one or more processors;
• a storage medium configured to store instructions, the instructions, when executed by the one or more processors, cause the one or more processors to perform steps comprising:
• retrieving survey responses of a set of individuals regarding a set of phenotypes of the individuals, the set of individuals comprising a target individual;
• constructing a matrix that arranges the set of individuals in a first dimension of the matrix and the set of phenotypes in a second dimension of the matrix, at least a subset of elements in the matrix being numerical representations of the individuals' phenotypes obtained from the survey responses;
• identifying one or more undetermined phenotypes of the target individual, the matrix having no value at particular elements corresponding to the one or more undetermined phenotypes of the target individual;
• determining a sparsity of the matrix, the sparsity of the matrix corresponding to a number of undetermined values in the matrix;
• determining, responsive to the sparsity higher than a threshold, at least one of the undetermined values corresponding to a particular individual based on corresponding values of other individuals who are similar to the particular individual to reduce the sparsity of the matrix;
• and
• determining a prediction of the one or more undetermined phenotypes of the target individual by performing collaborative filtering using a machine learning technique, the collaborative filtering based on other phenotypes of the target individual and based on at least another individual's phenotypes as represented in numerical representations in the matrix, performing the collaborative filtering using the machine learning technique comprises:
• performing a factorization of the matrix to estimate two or more lower dimensionality matrices that generates an approximation of the matrix, and
• using coordinate descent to iteratively adjust values in elements of the two or more lower dimensionality matrices to reduce differences between the approximation and the matrix.