Data and Network Science for Noisy Heterogeneous Systems

Data in many growing fields has an underlying network structure that can be taken advantage of. In this dissertation we apply data and network science to problems in the domains of systems biology and healthcare. Data challenges in these fields include noisy, heterogeneous data, and a lack of ground truth.

The primary thesis of this work is that the application of data mining and network science to data with these challenges must be carefully joined with domain knowledge. In the fields of systems biology and healthcare, data mining is increasingly being used to create models that represent the current state of understanding of important problems. These models are used to determine the direction of future work and to evaluate novel approaches. Therefore, any systematic bias in the models can be detrimental to scientific progress. For these same reasons, data mining has enormous potential to contribute to advances in our understanding.

Through our study of data and network science in this context we innovate new methods and highlight open and important problems. We strongly advocate the use of multiple measures for relationships in data in addition to heterogeneous data for the construction of network models, as relationships are often a matter of degree and no single measure or data set can capture everything about a problem.