PREDICTION ANALYSIS OF PHARMACOKINETIC PARAMETERS OF SEVERAL ORAL SYSTEMIC DRUGS USING IN SILICO METHOD

Objective: This research aims to observe the pharmacokinetic parameters that can be predicted using a software, discover the best software to predict pharmacokinetic properties, and analyze the correlation between pharmacokinetic parameters used as descriptors with absorption percentage (%ABS) from references. Methods: This research was conducted using Molinspiration, QikProp, admetSAR, SwissADME, Chemicalize, and pkCSM software. This research analyzed 34 oral systemic drug compounds for absorption rate and six descriptors comprising molecular weight (MW), logP, hydrogen bond acceptor (HBA), hydrogen bond donor (HBD), polar surface area (PSA), and pKa. Results: SwissADME showed the most accurate prediction of MW, logP, and HBD. Chemicalize showed the most accurate prediction of HBA, PSA, and pKa. Further, admetSAR showed the most accurate prediction of Caco-2 permeability. The highest R value was obtained from the correlation between %ABS with Caco-2 permeability on 34 drug compounds (R=0.8211). Conclusion: The highest R value was obtained from the correlation between %ABS with Caco2 permeability on 34 drug compounds (R=0.8211), which showed a significant relationship (*p<0.001). This indicates that oral systemic drugs are affected by Caco-2 permeability. Moreover, the result of this research can be considered for the development of oral systemic drugs.


Optimization of predicted pharmacokinetic parameters
By comparing experimental data from the reference with the softwarepredicted data from multiple software, optimization was performed to determine the software that showed the most accurate prediction of the pharmacokinetic parameters used in this research.

Analysis predicted descriptors for oral systemic drugs
The experimental %ABS was correlated with the predicted pharmacokinetic parameters and analyzed using Microsoft Excel. The resulting scatter plot showed the correlation coefficient (R) between the experimental %ABS with the descriptors of oral systemic drugs. Furthermore, SPSS was used to calculate significant values (*p). If *p-value was <0.05, the result was considered statistically significant.

Preparation of experimental pharmacokinetic parameters
Experimental pharmacokinetic parameters were obtained from the study by Zhao et al. and previous studies ( Table 1).
The correlation between reference MW and predicted MW showed R=0.9985; the correlation between reference logP and predicted logP showed R=0.8694; the correlation between reference HBA and predicted HBA showed R=0.8716; the correlation between reference HBD and predicted HBD showed R=0.9253; the correlation between reference PSA and predicted PSA showed R=0.9916; the correlation between reference pKa and predicted pKa showed R=0.6463; and the correlation between reference Caco-2 permeability and predicted Caco-2 permeability showed R=0.8593.

Analysis predicted descriptors for oral systemic drugs
The correlation between %ABS and predicted pharmacokinetic parameters was analyzed using Microsoft Excel. The correlation between %ABS and predicted MW showed R=−0.4773; the correlation %ABS percentage and predicted logP showed R=0.3534; the correlation between %ABS and predicted HBA showed R=−0.7205; the correlation between %ABS and predicted HBD showed R=−0.7046; the correlation between %ABS and predicted PSA showed R=−0.6627; the correlation between %ABS and predicted pKa showed R=−0.5453; and the correlation between %ABS and predicted Caco-2 permeability showed R=0.8211 (Fig. 2). Table 2 indicates the two absorption multiple regression models obtained in this research. Model 1 was created with all compounds with complete parameters and model 2 was created with all compounds with complete parameters but without 100% absorption. From the data, multiple regressions derived better R 2 value were obtained from model 2 than from model 1 (0.792948 and 0.750249, respectively). However, because the standard errors for the models were similar (17.22067 and 17.57382, respectively), the differences were not statistically significant. Further, the weightages of several parameters in model 2 were larger than those in model 1 and LogP, Caco2, and pKa were noticeably larger than the others. Absorption multiple regression results are listed in Table 2.

DISCUSSION
The correlation between reference MW and predicted MW; reference logP and predicted logP; reference HBA and predicted HBA; reference HBD and predicted HBD; reference PSA and predicted PSA; and reference Caco-2 permeability and predicted Caco-2 permeability showed strong correlations with R=0.9985, 0.8694, 0.8716, 0.9253, 0.9916, and 0.8593, respectively; however, the correlation between reference pKa and predicted pKa showed medium correlation with R=0.6463. Therefore, predicted pKa showing R value (<0.7) is the only parameter that exhibits a moderate positive relationship [36]. Several researches mention the accuracy problem of pKa prediction and state that pKa prediction is highly dependent on the dataset [37]. The simplification of the software calculation may also be a limitation of pKa prediction [38]. To improve the algorithm, drug type clustering based on its pKa level should be considered because the algorithm may show different results for acidic and basic drugs. The pKa range of clusterization should be optimized in further research. In addition, the dataset in this experiment contains various compounds that may act as obstacles in accurate pKa prediction for all structures. In this study, pKa of several compounds, such as aminopyrine with anti-inflammatory action; hydrocortisone, methylprednisolone, and prednisolone, which are corticosteroid agents; and nizatidine and ranitidine from H2 receptor antagonist group, could not be accurately predicted. The R value is suggested to reach >0.9 to be considered as accurate prediction.
From this research, we found that the various software programs provided different parameter prediction results. None of the software served as the most accurate prediction tool for all parameters. However, out of seven parameters, Chemicalize and SwissADME accurately predicted three complimentary parameters each. Moreover, Caco2 prediction only can be accurately done using admetSAR.
Analysis descriptors for the 34 oral systemic drugs resulting in the highest R value were the significant correlation between %ABS and Caco-2 permeability (R=0.8211; *p<0.001) (Fig. 2).
The absorption multiple regression models were derived from these data by including the compounds with 100% absorption (model 1) or excluding it (model 2) to observe how the nonlinear function part affects the correlation. Better R 2 values were obtained from model 2 than from model 1; however, the difference was not significant. Further, the weightages of several parameters in model 2 were larger than those in model 1, with LogP, Caco2, and pKa being noticeably larger than the others. This suggests that these three parameters, as opposed to MW and PSA, have higher tendencies to affect absorption.
In general, a model is acceptable if it has R 2 >0.6 [39]. In addition, in this case, good model fitness was observed in both models. This study is limited by its small dataset and usually good prediction is statistically derived from large datasets; therefore, further considerations need to be undertaken such as to selectively include various drugs and also to try several other software programs not included in this study. Nevertheless, from the experiment, both models are acceptable to be used as early in silico tools to assist the prediction of the absorption of systemic oral drugs.

CONCLUSION
Parameter prediction was successfully performed in this research. SwissADME was the most accurate software in predicting MW, logP, and HBD; Chemicalize was the most accurate software in predicting HBA, PSA, and pKa; and admetSAR was the most accurate software in predicting Caco2 permeability. The highest R value was obtained from the significant correlation between %ABS and Caco-2 permeability of 34 drug compounds (R=0.8211; *p<0.001). These results indicate that the %ABS of oral systemic drugs is affected by Caco-2 permeability.