Evaluation of CYP۲D۶ polymorphisms and other parameters on the plasma concentration of the tomoxifen and its metabolites by machine learning approaches

Background and aims: tamoxifen is one of the useful drugs in the prevention and treatment ofbreast cancer. Effectiveness of this drug is still under investigation in the CYP۲D۶ genetic polymorphismpersons. It is hypothesized that poor metabolizer genotypes do not sufficiently converttamoxifen to its active metabolites in the therapeutic concentrations and hence leads to the lowefficacy and the risk of relapsing of the tumor. The aim of this study is to find any relationshipbetween the tamoxifen metabolites plasma concentration and the genetic polymorphism of theCYP۲D۶ or the other parameters.Method: In the present study, ۸۴ estrogen receptor positive breast cancer patients which werepreviously involved in the determination of their CYP۲D۶ genetic polymorphisms and treatedwith tamoxifen in Sari, Iran were used. Totally, ۲۲ independent parameters for each patient suchas plasma concentration of endoxifen and ۴-hydroxytamoxifen, number of delivery, number ofabortion, size of tumor, duration of OCP usage, number of radiotherapy, duration of tamoxifenconsumption, number of chemotherapy, Her۲ status, Category of polymorphism, detail of polymorphism,polymorphism subgroup and etc., were included for analysis. All of the preprocessingsteps such as encoding, data type conversion, partitioning, missing value handling, outlier detectionand normalization were performed by KNIME Analytics Platform. Various machine learningmethods such as logistic regression, decision tree, random forest, tree ensemble, gradient boostedtree and neural networks were also constructed and used in the KNIME Analytics Platform.Results: Results showed that polymorphism subgroup has been used as root split in the decisiontree algorithm to define the target parameter ۴-hydroxytamoxifen with an overall accuracy of۶۹.۵%. results of the random forest and tree ensemble methods with an overall accuracy of ۷۳.۹%and ۷۸% respectively, indicated that the total duration of tamoxifen consumption, number of radiotherapyand duration of OCP usage are affecting the ۴-hydroxytamoxifen plasma concentration.Conclusion: polymorphism subgroups (EM/EM, EM/IM, EM/PM, IM/PM, PM/PM) which definethe normal, intermediate or poor metabolizers was the first and main feature to correlatethe plasma concentration of the ۴-hydroxytamoxifen in the decision tree method. The plasmaconcentration of ۴-hydroxytamoxifen was also dependent on the total duration of tamoxifen consumptionand tamoxifen plasma concentration but the plasma concentration of endoxifen wasfound to be dependent primarily on the plasma concentration of the ۴-hydroxytamoxifen. Since,tamoxifen is converted naturally to ۴-hydroxytamoxifen by CYP۲D۶ and then to endoxifen byCYP۳A۴ isoenzymes, thus confirming our accurate predictions. The other main factors affectingthe plasma concentration of the tamoxifen metabolites are the duration of OCP usage, number ofradiotherapy and Her۲ status.