Prediction of ADME properties using curriculum learning

Identifier:  TFM:1880

Handle: https://hdl.handle.net/20.500.11797/TFM1880

Authors:  Lagemann, Jens Alexander

Abstract:
This thesis will study the effect that such a curriculum sampler has when trying to predict a variety of molecular properties, using datasets describing absorption, distribution, metabolism and excretion of various samples. Making a quantitative analysis of how models trained in different configurations compare. The results show that the ordering of samples during training is a relevant factor in how well a model learns to generalize, although the type of molecular complexity measures depends on the application.