Incremental Machine Learning to Reduce Biochemistry Lab Costs in the Search for Drug Discovery

This paper promotes the use of supervised machine learning in laboratory settings where chemists have a large number of samples to test for some property, and are interested in identifying as many positive instances for the least laboratory testing effort. Rather than traditional supervised learning where the chemists would first develop a large training set and then train a classifier, the paper promotes incrementally re-training from each lab test as it completes and then predicting the next best sample to test, as in the field of reinforcement learning. The method outperformed the 2001 KDD Cup thrombin competition winner, partly due to its reduced risk to concept drift from training set to test set.

• Type: Conference
• Year: 2002
• Where: KDD
• Authors: George Forman
• Keywords: Data Mining, KDD 2002, Lab Test, Test Set, Traditional Supervised Learning