Opportunities and Considerations in the Application of Artificial Intelligence to Pharmacokinetic Prediction

Abstract

The improvement in the ability of the pharmaceutical industry to predict human pharmacokinetic behavior are attributable to major technological shifts from 1990 to the present day. The opportunity for the application of AI/ML based approaches in the pharmaceutical industry is driven by the abundance of data sets that exist within individual pharmaceutical and biotech companies and the availability, within these environments, of abundant computing power. This chapter seeks to describe opportunities for artificial intelligence to contribute to the assessment and evaluation of the dug metabolism and pharmacokinetic (DMPK) properties of novel compounds across the drug discovery and development continuum. Many initiatives are already underway with respect to the application of AI/ML in predicting pharmacokinetic profiles so the question is not whether AI will influence pharmacokinetic prediction but rather how to best utilize and incorporate this and how to evaluate the value added from these applications. Since our understanding of the underlying biology of the in vitro and in vivo systems with respect to ADME, one of the key challenges to AI-based methods will be the ability to adapt to data sets that change in quality over time.

Equations

• 1. Calculation of intrinsic clearance (Cl_int) from microsomal half-life (t _1/2 in minutes)

  $$ {\mathrm{Cl}}_{\mathrm{int}}=0.693\times \frac{1}{t_{\frac{1}{2}}}\times \frac{\mathrm{g}\ \mathrm{of}\ \mathrm{liver}\ \mathrm{weight}}{\mathrm{kg}\ \mathrm{body}\ \mathrm{weight}}\times \frac{\mathrm{mL}\ \mathrm{incubation}\ \mathrm{volume}}{\mathrm{mg}\ \mathrm{microsomal}\ \mathrm{protein}}\times \frac{45\ \mathrm{mg}\ \mathrm{microsomal}\ \mathrm{protein}}{\mathrm{g}\ \mathrm{of}\ \mathrm{liver}\ \mathrm{weight}} $$

  (1)

• 2. The well-stirred model of hepatic clearance. Q is hepatic blood flow, f _u is the fraction unbound in blood, Clint as defined in Eq. 1

  $$ {\mathrm{Cl}}_{\mathrm{h}}=\frac{Q\times {f}_{\mathrm{u}}\times {\mathrm{Cl}}_{\mathrm{int}}}{Q+{f}_{\mathrm{u}}\times {\mathrm{Cl}}_{\mathrm{int}}} $$

  (2)

• 3. The parallel tube model of hepatic clearance. Q is hepatic blood flow, f _u is the fraction unbound in blood, Clint as defined in Eq. 1

  $$ {\mathrm{Cl}}_{\mathrm{h}}=Q\times \left(1-{\mathrm{e}}^{\frac{-{\mathrm{Cl}}_{\mathrm{int}}\times {f}_{\mathrm{u}}}{Q}}\right) $$

  (3)

• 4. Reduced χ ² statistic. N is the number of observation, Δ the difference between observed and predicted values, and σ ² the variance of the observation at a specific time-point

  $$ {\chi}^2=\frac{1}{N}\sum \limits_{i=1}^N\left(\frac{\Delta^2}{\sigma_i^2}\right) $$

  (4)

• 5. Mean-fold error. N is the number of observations

  $$ \mathrm{Mean}\ \mathrm{fold}\ \mathrm{error}={10}^{\left[\frac{1}{n}\sum \left(\log \mathrm{fold}\ \mathrm{error}\right)\right]} $$

  (5)