Indicator Regularized Non-Negative Matrix Factorization Vs. a Novel Combinatorial Heuristic Matrix Factorization: a Comparison of Matrix Factorization Methods as theBuilding Block of Drug Repurposing

Drug repurposing is one of the ponderable computational methods suggested for rapid developing of a drug.This approach can be formulated as a matrix factorization (MF) method. First, a MF method is applied todifferent types of drug-target relation and similarity matrices. Then, the unseen data will be analysed toevalutae the predictive power of the model. Some types of data in bioinformatics are binary, e.g. binaryrepresentation of impacts of antiviral drugs on viruses. MF of such data is more complicated than theconventional numerical values. Most of conventional algorithms for binary matrix factorization use gradientbasedmethods. These methods utilize relaxation approach to solve continuous binary problems. Therelaxation approach turns the binary constraint into a box constraint. Despite the widespread use, thesemethods do not perform well on sparse data sets, and also unable to return a proper approximation of problem.To avoid facing these limitations, we propose a binary matrix factorization method utilizing combinatorialoptimization and modular arithmetic. In our prposal, binary values are used in each step and the results comefrom modular multiplication. We compare our results with Indicator Regularized Non-Negative MatrixFactorization (IRNMF) which is a gradient based method. Both methods are appled to the same antiviralviruseinteraction matrix and Five-fold cross-validation (CV) are used to report the performance. The resultsindicate that better performance and lower error value are the advantages of the suggested method in comparewith IRNMF.

کلیدواژه ها:

Matrix Factorization ، Drug repurposing ، Antiviral Drugs ، Heuristic Factorization