Hao Lou1,2, Michael J Hageman3,4. 1. Department of Pharmaceutical Chemistry, University of Kansas, Lawrence, Kansas, 66047, USA. lou0@ku.edu. 2. Biopharmaceutical Innovation and Optimization Center, University of Kansas, Lawrence, Kansas, 66047, USA. lou0@ku.edu. 3. Department of Pharmaceutical Chemistry, University of Kansas, Lawrence, Kansas, 66047, USA. mhageman@ku.edu. 4. Biopharmaceutical Innovation and Optimization Center, University of Kansas, Lawrence, Kansas, 66047, USA. mhageman@ku.edu.
Abstract
PURPOSE: One knowledge gap related to subcutaneous (SC) delivery is unpredictable and variable bioavailability. This study was aimed to develop machine learning methods to predict whether mAb's bioavailability was ≥70% or below, without completely knowing the mechanism and causality between inputs and outputs. METHODS: A database of mAb SC products was built. The model training and validation were accomplished based on this database and a set of the inputs (product properties) were mapped to the output (bioavailability) using different machine learning algorithms. Dimensionality reduction was undertaken using principal component analysis (PCA). RESULTS: The bioavailability of the mAb products being investigated varied from 35% to 90%. The tree-based methods, including random forest (RF), Adaptive Boost (AdaBoost), and decision tree (DT) presented the best predictability and generalization power on bioavailability classification. The models based on Multi-layer perceptron (MLP), Gaussian Naïve Bayes (GaussianNB), and k nearest neighbor (kNN) algorithms also provided acceptable prediction accuracy. CONCLUSION: Machine learning could be a potential tool to predict mAb's bioavailability. Since all input features were acquired using theoretical calculations and predictions rather than experiments, the models may be particularly applicable to some early-stage research activities such as mAb molecule triage, design/optimization, mutant screening, molecule selection, and formulation design.
PURPOSE: One knowledge gap related to subcutaneous (SC) delivery is unpredictable and variable bioavailability. This study was aimed to develop machine learning methods to predict whether mAb's bioavailability was ≥70% or below, without completely knowing the mechanism and causality between inputs and outputs. METHODS: A database of mAb SC products was built. The model training and validation were accomplished based on this database and a set of the inputs (product properties) were mapped to the output (bioavailability) using different machine learning algorithms. Dimensionality reduction was undertaken using principal component analysis (PCA). RESULTS: The bioavailability of the mAb products being investigated varied from 35% to 90%. The tree-based methods, including random forest (RF), Adaptive Boost (AdaBoost), and decision tree (DT) presented the best predictability and generalization power on bioavailability classification. The models based on Multi-layer perceptron (MLP), Gaussian Naïve Bayes (GaussianNB), and k nearest neighbor (kNN) algorithms also provided acceptable prediction accuracy. CONCLUSION: Machine learning could be a potential tool to predict mAb's bioavailability. Since all input features were acquired using theoretical calculations and predictions rather than experiments, the models may be particularly applicable to some early-stage research activities such as mAb molecule triage, design/optimization, mutant screening, molecule selection, and formulation design.
Entities:
Keywords:
bioavailability; machine learning; material attribute; monoclonal antibody; subcutaneous
Authors: Linda J Chan; Jürgen B Bulitta; David B Ascher; John M Haynes; Victoria M McLeod; Christopher J H Porter; Charlotte C Williams; Lisa M Kaminskas Journal: Mol Pharm Date: 2015-02-18 Impact factor: 4.939
Authors: David S Wishart; Yannick D Feunang; An C Guo; Elvis J Lo; Ana Marcu; Jason R Grant; Tanvir Sajed; Daniel Johnson; Carin Li; Zinat Sayeeda; Nazanin Assempour; Ithayavani Iynkkaran; Yifeng Liu; Adam Maciejewski; Nicola Gale; Alex Wilson; Lucy Chin; Ryan Cummings; Diana Le; Allison Pon; Craig Knox; Michael Wilson Journal: Nucleic Acids Res Date: 2018-01-04 Impact factor: 16.971