(657a) Deducing Hidden Transcription Networks And Regulatory Signals From Gene Expression Data

Transcriptional regulation forms a bipartite network between genes and transcription regulators. Two key features of bipartite networks are the network architecture and regulatory signals. Such information is often implicitly imbedded in the network output, which in the case of transcriptional regulation would be the transcriptome. Here we develop a technique, Network Component Mapping (NCM), that deduces bipartite network connectivity and regulatory signals from data without the need for prior information. Network and regulatory signal inference in the absence of prior information is particularly attractive for situations when gene expression data is available from a poorly understood organism, or a poorly characterized environment. We demonstrate the utility of our approach by analyzing UV-vis spectra from mixtures of metabolites and gene expression data from Saccharomyces cerevisiae. From UV-vis spectra, hidden mixing networks and pure component spectra (sources) were deduced to a higher degree of resolution with our method than other current bipartite techniques. Analysis of S. cerevisiae gene expression from two separate environmental conditions (zinc and DTT treatment) yielded transcription networks consistent with ChIP-chip derived network connectivity. Due to the high degree of noise in DNA microarray data the transcription network for many genes could not be inferred. However, with comparatively clean expression data, our technique was able to deduce hidden transcription networks and instances of combinatorial regulation. For noisy data, NCM yields the sparsest network capable of explaining the data. In addition, partial knowledge of the network topology can be utilized by NCM as constraints.