Semi-Supervised Multi-Task Learning for Predicting Interactions between HIV-1 and Human Proteins

 

 
Y. Qi, NEC Laboratories America
O. Tastan, Carnegie Mellon University
J.G. Carbonell, Carnegie Mellon University
J. Klein-Seetharaman, Carnegie Mellon University
J. Weston, Google Research NY
 
Contact Author: Yanjun Qi (qyj@cs.cmu.edu)

 


Paper PDF

Online Link: URL

ECCB 2010 Talk Slide

 

Abstract

 

 

Motivation: Protein-protein interactions (PPIs) are critical for virtually every biological function. Recently, researchers suggested to use supervised learning for the task of classifying pairs of proteins as interacting or not. However, its performance is largely restricted by the availability of truly interacting proteins ({\em labeled}).  Meanwhile there exist a considerable amount of protein pairs where an association appears between two partners, but not enough experimental evidence to support it as a direct interaction ({\em partially labeled}).

 

Results: We propose a semi-supervised multi-task framework for predicting PPIs from not only {\em labeled}, but also {\em partially labeled} reference sets. The basic idea is to perform multi-task learning on a supervised classification task and a semi-supervised task.  The supervised classifier trains a multi-layer perceptron network for PPI predictions from {\em labeled} examples. The semi-supervised auxiliary task shares network layers of the supervised classifier and trains with {\em partially labeled} examples. Semi-supervision could be utilized in multiple ways. We tried three approaches in this paper, (1) classification (to distinguish partial positives with negatives); (2) ranking (to rate partial positive more likely than negatives); (3) embedding (to assure data clusters get similar labels).  We applied this framework to identify the set of interacting pairs between HIV-1 and human proteins. Our method improved upon the state-of-the-art method for this task indicating the benefits of semi-supervised multi-task learning using auxiliary information.

 

 

 

 

Feature Details:

 

Due to the length limitation of the main text, we put some details in this supporting website.

 

á          Prior work of this paper: O. Tastan, Y. Qi, J.G. Carbonell, J. Klein-Seetharaman, “Prediction of Interactions between HIV-1 and Human Proteins by Information Integration“, Pacific Symposium on Biocomputing 14: (PSB-2009) Jan. 2009 (Supplementary Web)

 

á          Details about feature sets used !

 

á          Download all features used in this paper ! ==> (18 features for all possible HIV-Human Pairs)

 

á          Download IDs of the pairs : ==> (HIV Id: Human ID) for all hiv-Human pairs in the above feature file !

 

 

Data and Reference Sets Sharing:  

 

á          The top ranked 2500 predictions between HIV-1 and Human proteins could be downloaded here

 

á          The experts labelled positive PPIs pairs is shared @ here !

 

á          The NIAID GroupI HIV-1 to human protein pairs (partial positive) are downloadable here  !

 

á          The NIAID GroupII HIV-1 to human protein pairs could be downloaded too !

 

 

 

Validation of Predictions:

 

á          Validation of top ranked HIV-1 to human protein interaction pairs

á          Human gene lists from functional screens related to HIV-1 (published recently) are shared here for readersÕ convenience.