Table 3 Impact of the different drug panel sizes on predictive performance on the PRISM dataset using a test set of 52 PDCs

From: Establishing predictive machine learning models for drug responses in patient derived cell culture

Panel size

Rpearson

Rspearman

RMSE

Top 0.5%

Top 1%

Top 2%

Hit rate (top 45)

23

0.530/0.045

0.394/0.035

0.776/0.012

0.067/0.036

0.169/0.059

0.368/0.087

0.559/0.103

45

0.609/0.024

0.444/0.022

0.742/0.009

0.083/0.046

0.215/0.070

0.437/0.078

0.662/0.099

90

0.687/0.019

0.473/0.015

0.684/0.010

0.159/0.064

0.345/0.074

0.567/0.054

0.845/0.063

  1. The reported metrics include Rpearson, Rspearman, RMSE, accuracy for the top 0.5%, 1%, and 2% of predicted drugs, and hit rates among the top 45 recommendations. Results are presented as mean/stdev derived from five experiments.
Back to article page