Direct Comparison of Total Clearance Prediction: Computational Machine Learning Model versus Bottom-Up Approach Using In Vitro Assay

Abstract

The in vitro–in vivo extrapolation (IVIVE) approach for predicting total plasma clearance (CLtot) has been widely used to rank order compounds early in discovery. More recently, a computational machine learning approach utilizing physicochemical descriptors and fingerprints calculated from chemical structure information has emerged, enabling virtual predictions even earlier in discovery. Previously, this approach focused more on in vitro intrinsic clearance (CLint) prediction. Herein, we directly compare these two approaches for predicting CLtot in rats. A structurally diverse set of 1114 compounds with known in vivo CLtot, in vitro CLint, and plasma protein binding was used as the basis for this evaluation. The machine learning models were assessed by validation approaches using the time- and cluster-split training and test sets, and five-fold cross validation. Assessed by five-fold validation, the random forest regression (RF) and radial basis function (RBF) models demonstrated better prediction performance in eight attempted machine learning models. The CLtot values predicted by the RF and RBF models were within two-fold of the observed values for 67.7 and 71.9% of cluster-split test set compounds, respectively, while the predictivity was worse in the time-split dataset. The predictivity of both models tended to be improved by incorporating in vitro parameters, unbound fraction in plasma (f u,p), and CLint. CLtot prediction utilizing in vitro CLint and the well-stirred model, correcting for the fraction unbound in blood, was substantially worse compared to machine learning approaches for the same cluster-split test set. The reason that CLtot is underestimated by IVIVE is not fully explained by considering the calculated microsomal unbound fraction (cfu,mic), extended clearance classification system (ECCS), and omitting high clearance compounds in excess of hepatic blood flow. The analysis suggests that in silico machine learning models may have the power to reduce reliance on or replace in vitro and in vivo studies for chemical structure optimization in early drug discovery.

Novel QSAR Approach for a Regression Model of Clearance That Combines DeepSnap-Deep Learning and Conventional Machine Learning

“…Kosugi and Hosea 24 constructed a prediction model using in-house compounds as a regression model for rat CL. Their models used 330 molecular descriptors and eight algorithms including random forest and radial basis functions.…”

“… 21 In addition to using the features of compounds, Kim et al 33 and Wang et al 37 used biological descriptors related to membrane transporters and constructed a prediction model which improved the prediction performance. Kosugi and Hosea 24 improved the prediction performance of their model using in vitro experimental values as a new feature in addition to molecular descriptors. These findings indicated that the inclusion of different types of features might improve the prediction performance of QSAR models.…”

“…For this reason, it is desirable to construct not only classification models but also regression models that have high performance. Kosugi and Hosea 24 reported a regression model of rat CL that was constructed using conventional ML and 1114 of their in-house compounds. The prediction performance of this model was not sufficient [correlation coefficient ( R 2 ) = 0.555; root mean square error (RMSE) = 0.332].…”

Direct Comparison of the Prediction of the Unbound Brain-to-Plasma Partitioning Utilizing Machine Learning Approach and Mechanistic Neuropharmacokinetic Model

“…Each prediction model was generated with StarDrop (StarDrop v6.5.0, Optibrium Ltd, Cambridge, UK) according to the previously described method ( 23 , 24 ). StarDrop uses 2D SMARTS-based descriptors, which are counts of atom types and functionalities, along with whole molecule properties such as molecular weight (M.W.…”

Development and Biological Characterization of a Novel Selective TrkA Agonist with Neuroprotective Properties against Amyloid Toxicity

“…Based on a rearrangement of the well stirred model [55,56], classification bands can be used to categorize compounds into low, medium or high clearance, assuming an extraction ratio of 0.3 and 0.7 for the low and high boundaries, respectively. Next, this can be scaled to intrinsic clearance (µL/min/mg protein) using the relevant liver weights [57] and microsomal protein concentration [50,58,59] obtained from the literature and computational data. For humans, an CL int < 8.6 µL/min/mg protein defines a low intrinsic clearance classification band, whereas an CL int > 47.0 µL/min/mg protein defines a high intrinsic clearance classification band.…”