Table 1. Datasets.

	Dataset	Original size	Size after filtering 1	Data source
I. Datasets of all drugs
Detrimental drug interactions 2
Severe	A	21831	10323	Drugs.com
Moderate	B	112976	92958	Drugs.com
Minor	C	13143	17973	Drugs.com
Beneficial drug interactions 3 , 4	D	429	293	DCDB, TTD
II. Cancer-related datasets
Detrimental drug interactions 2
Severe	E	1053	817	Drugs.com
Moderate	F	6857	5700	Drugs.com
Minor	G	273	241	Drugs.com
Beneficial drug interactions 3 , 5	H	55	33	DCDB, TTD
III. Negative datasets used in ROC analysis
All FDA-approved drugs 6	I	848253	733542	Drugbank
Random drugs 7	J	427350	426425	-

¹We filtered the available drug pairs by leaving out the drug combinations where the components have exactly the same targets, or the components were structurally similar, as described in Methods. The drugs with no available targets were also discarded

²Taken from Drugs.com (November 11, 2013) as described in the methods

³Taken from the Drug Combination Database (March 8, 2012) and the Therapeutic Target Database (July 23, 2012) as described in the methods

⁴All approved drug combinations were included

⁵All approved drug combinations that are used in cancer treatment.

⁶We made all possible binary combinations of FDA-approved drugs (taken from DrugBank, 12^th September of 2012), and then leaved out all pairs that were listed as beneficial or detrimental combinations.

⁷We constructed random drugs corresponding to the number of targets of all individual drugs. We generated 25 random drugs for each target count (37). From this pool we made the all possible binary combinations. In each case, we randomly selected a negative set of the size which was 5 times greater than the positive dataset [51].