Transferable crowd-based network models

Abstract

Specific inference problems have shown the potential of approximating a reference network with high accuracy using crowd-based predictions. In gene regulatory network (GRN) inference a crowd-based prediction is an aggregation of network predictions. We build crowd-based network models using the weighted average in logistic (WAL) regression. As an example we use the five networks of size 100 of the DREAM4 in silico network inference challenge. A simple crowd-based network model can be built by combining pre- diction lists of few dissimilar well-performing classifiers with an appropriate noise model. Adding similar predictions, bad predictions and/or short prediction lists are possibilities of extending a crowd-based model. WAL regression-based network models are applicable across networks.