Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

PyDTI -- a Python library for drug-target interaction prediction

version 1.0, June 08 2015

This package is written by:

Yong Liu


School of Computer Engineering, Nanyang Technological University, Singapore Joint NTU-UBC Research Centre of Excellence in Active Living for Elderly (LILY), Nanyang Technological University, Singapore

For any questions regarding to this library, please feel free to contact the author.

PyDTI is implemented by Python 2.7.9, which can be downloaded from: PyDTI requires several other Python packages, including Numpy, scikit-learn, Scipy, and pymatbridge (a bridge between Python and Matlab).

The original packages can be found here:

Note that pymatbridge is only required by KBMF2K. The 64-bit Windows binaries of Numpy, scikit-learn, and Scipy can also be found at:

  1. Add the folder "$PYTHON_ROOT$/Scipts/" to the system path. Please replace "$PYTHON_ROOT$" with the root folder of Python in your system.

  2. Install the packages using pip utility. Open a console and type the following to install

pip install numpy scipy scikit-learn

For the KBMF2K method, please download the matlab code implemented by M. Gonen from and put the matlab code in the sub-folder "kbmf2k/". To connect the Matlab code and Python, we need to define a Matlab function:

function predictR = kbmf(args)
	Kx = args.Kx;
	Kz = args.Kz;
	Y = args.Y;
	R = args.R;
	state = kbmf_regression_train(Kx, Kz, Y, R);
	prediction = kbmf_regression_test(Kx, Kz, state);
	predictR =;

Save this function into a Matlab file named kbmf.m and put this file into the subfolder "kbmf2k/".

To get the results of different methods, please run by setting suitable values for the following parameters:

--method 			set DTI prediction method
--dataset: 			choose the benchmark dataset, i.e., nr, gpcr, ic, e
--folder:			set the the folder that contains the datasets (default "datasets/")
--csv:				choose the cross-validation setting, 1 for CVS1, 2 for CVS2, and 3 for CVS3, (default 1)
--specify-arg:		0 for choosing optimal arguments, 1 for using default/specified arguments (default 1)
--method-opt:		set arguments for each method (method ARGUMENTS have the form name=value)
--predict-num:		0 for not predicting novel DTIs, a positive integer for predicting top-N novel DTIs (default 0)

Here are some examples:

(1) run a method with default arguments
	python --method="nrlmf" --dataset="nr"
	python --method="nrlmf" --dataset="nr" --cvs=2
	python --method="nrlmf" --dataset="nr" --cvs=2 --specify-arg=1

(2) run a method with specified arguments

	python --method="nrlmf" --dataset="nr" --cvs=1 --specify-arg=1 --method-opt="r=100"

	python --method="nrlmf" --dataset="nr" --cvs=1 --specify-arg=1 --method-opt="c=5 K1=5 K2=5 r=100 lambda_d=0.125 lambda_t=0.125 alpha=0.25 beta=0.125 theta=0.5"

	You can refer to lines 47-58 in the for the default parameters of each DTI prediction method.

(3) choose the optimal parameters for a method

	python --method="nrlmf" --dataset="nr" --cvs=1 --specify-arg=0

(4) predict the top-100 novel DTIs

	python --method="nrlmf" --dataset="nr" --predict-num=100 --method-opt="r=100"
  1. You can run for the statistical comparision between NRLMF and other baseline methods on all datasets, under different cross-validation settings. Note that you should first obtain the auc and aupr results of each method.



No description, website, or topics provided.







No releases published


No packages published