Comparative analysis of four disease prediction models of Parkinson's disease.

Parkinson's disease (PD) is a multi-factorial disorder with high-penetrant mutations accounting for small percentage of PD. Our previous studies demonstrated individual association of genetic variants in folate, xenobiotic, and dopamine metabolic pathways with PD risk. The rational of the study was to develop a risk prediction model for PD using these genetic polymorphisms along with synuclein (SNCA) polymorphism. We have generated additive, multifactor dimensionality reduction (MDR), recursive partitioning (RP), and artificial neural network (ANN) models using 21 SNPs as inputs and disease outcome as output. The clinical utility of all these models was assessed by plotting receiver operating characteristics curves where in area under the curve (AUC) was used as an index of diagnostic utility of the model. The additive model was the simplest and exhibited an AUC of 0.72. The MDR model showed significant gene-gene interactions between SNCA, DRD4VNTR, and DRD2A polymorphisms. The RP model showed SHMT C1420T as important determinant of PD risk. This variant allele was found to be protective and this protection was nullified by MTRR A66G. Inheritance of SHMT wild allele and SNCA intronic polymorphism was shown to increase the risk of PD. The ANN model showed higher diagnostic utility (AUC = 0.86) compared to all the models and was able to explain 56.6% cases of sporadic PD. To conclude, the ANN model developed using SNPs in folate, xenobiotic, and dopamine pathways along with SNCA has higher clinical utility in predicting PD risk compared to other models.

Clinical utility of genetic variants of glutamate carboxypeptidase II in predicting breast cancer and prostate cancer risk.

In view of documented evidence showing glutamate carboxypeptidase II (GCPII) inhibitors as promising anti-cancer agents, certain variants of GCPII modulate breast and prostate cancer risk, and we developed an artificial neural network (ANN) model of GCPII variants and ascertained the risk associated with eight genetic variants of GCPII. In parallel, an in silico model was developed to substantiate the ANN simulations. The ANN model with modified sigmoid function was used for disease prediction, whereas the hyperbolic tangent function was used to predict folate hydrolase 1 (FOLH1) and prostate specific membrane antigen (PSMA) expression. PyMOL models of GCPII variants were developed, and their affinity toward the folylpolyglutamate (FPG) ligand was tested using glide score analysis. Of the eight genetic variants of GCPII, p.P160S alone conferred protection against both of the cancers. This variant exhibited higher affinity toward FPG compared with wild GCPII (-2.06 vs. -1.69); and positive correlation was observed between the P160S variant and circulating folate (r = 0.60). The ANN model for disease prediction showed significant predictability associated with GCPII variants toward breast cancer (area under the curve (AUC): 1.00) and prostate cancer (AUC: 0.97), with clear distinguishing ability of healthy controls (AUC: 0.96). The ANN models for the expression of FOLH1 and PSMA explained 60.5% and 86.7% of the variability, respectively. Thus, GCPII variants are potential contributors of risk toward breast cancer and prostate cancer. Risk modulation appeared to be mediated through changes in the expression of FOLH1 and PSMA.