Improved genome-scale multi-target virtual screening via a novel collaborative filtering approach to cold-start problem
Hansaim Lim, Paul Gray, Lei Xie, and Aleksandar Poleksic

COSINE is a computer program for predicting protein-chemical interactions. Building upon the so-called "one-class collaborative filtering", our algorithm incorporates position-specific weighting and imputation of interaction matrix entries, as well as “dual regularization” from both drugs and targets. While COSINE can also predict drug off-targets, it is specifically tailored to address the “cold start” problem. In other words, the program is capable of finding targets for “new” chemicals, those with little to no available interaction data. The program has been extensively tested on both gene-specific and cross-gene family benchmarks. These tests clearly show that our program outperforms current state-of-art methods for the same problem.

Source code (Matlab)
Benchmarking test sets