摘要
Studies on drug combinations are becoming more and more popular in the past few decades, with the development of computer and algorithms. One of the most common methods in optimizing drug combinations is regression of a polynomial model based on certain number of experimental observations. In this paper, we study how to determine the degree of polynomials in different circumstances of drug combination optimization. Using cross-validation, we have found that in most cases, a high degree results in failures of accurate prediction, named overfitting. An anti-noise test has also revealed that polynomial model with high degree tends to be less resistant to random errors in the observations.
Studies on drug combinations are becoming more and more popular in the past few decades, with the development of computer and algorithms. One of the most common methods in optimizing drug combinations is regression of a polynomial model based on certain number of experimental observations. In this paper, we study how to determine the degree of polynomials in different circumstances of drug combination optimization. Using cross-validation, we have found that in most cases, a high degree results in failures of accurate prediction, named overfitting. An anti-noise test has also revealed that polynomial model with high degree tends to be less resistant to random errors in the observations. © 2014 Chinese Association of Automation.
