Probabilistic reasoning in intelligent systems: networks of plausible inference
Probabilistic reasoning in intelligent systems: networks of plausible inference
Factorial Hidden Markov Models
Machine Learning - Special issue on learning with probabilistic representations
Dynamic bayesian networks: representation, inference and learning
Dynamic bayesian networks: representation, inference and learning
ICML '06 Proceedings of the 23rd international conference on Machine learning
A Bayesian model that links microarray mRNA measurements to mass spectrometry protein measurements
RECOMB'07 Proceedings of the 11th annual international conference on Research in computational molecular biology
A generalized mean field algorithm for variational inference in exponential families
UAI'03 Proceedings of the Nineteenth conference on Uncertainty in Artificial Intelligence
EURASIP Journal on Bioinformatics and Systems Biology
Hi-index | 0.02 |
Methods suggested for reconstructing regulatory networks can be divided into two sets based on how the activity level of transcription factors (TFs) is inferred. The first group of methods relies on the expression levels of TFs assuming that the activity of a TF is highly correlated with its mRNA abundance. The second treats the activity level as unobserved and infers it from the expression of the genes the TF regulates. While both types of methods were successfully applied, each suffers from drawbacks that limit their accuracy. For the first set, the assumption that mRNA levels are correlated with activity is violated for many TFs due to post-transcriptional modifications. For the second, the expression level of a TF which might be informative is completely ignored. Here we present the Post-Transcriptional Modification Model (PTMM) that unlike previous methods utilizes both sources of data concurrently. Our method uses a switching model to determine whether a TF is transcriptionally or post-transcriptionally regulated. This model is combined with a factorial HMM to fully reconstruct the interactions in a dynamic regulatory network. Using simulated and real data we show that PTMM outperforms the other two approaches discussed above. Using real data we also show that PTMM can recover meaningful TF activity levels and identify post-transcriptionally modified TFs, many of which are supported by other sources.