Effectively identifying compound-protein interactions by learning from positive and unlabeled examples

      Here, we present the first compound-protein interaction (CPI) prediction method -- PUCPI, without negative samples but only positive and unlabeled samples.We believe that there are not really negative CPIs unless you validate them by conducting biology experiments. We combined compound substructures information and protein domains into a single feature vector to represent a CPI.

      Currently the PUCPI program (version 0.1) is available only for the Windows 7 (or above) platform. Because the program is implemented by python, Python 2.7 is also required.

  • Download the datasets: (including human and C.elegans datasets).
  • Download the source code, implemented by python.(PUCPI software ).
  • Please read the readme.txt file.

  • Supplemental data (right click and select save as):

    Other useful links:

    If you find any problem about PUCPI, please contact chengzhanzhan@fudan.edu.cn.