Limits...
A novel HMM-based method for detecting enriched transcription factor binding sites reveals RUNX3 as a potential target in pancreatic cancer biology.

Levkovitz L, Yosef N, Gershengorn MC, Ruppin E, Sharan R, Oron Y - PLoS ONE (2010)

Bottom Line: Using a novel experimental paradigm to distinguish between normal and PAC cells, we find that RUNX3 mRNA (but not RUNX1 or RUNX2 mRNAs) exhibits time-dependent increases in normal but not in PAC cells.These increases are accompanied by changes in mRNA levels of putative RUNX gene targets.The integrated application of DEMON and a novel differentiation system led to the identification of a single family member, RUNX3, which together with four of its putative targets showed a robust response to a differentiation stimulus in healthy cells, whereas this regulatory mechanism was absent in PAC cells, emphasizing RUNX3 as a promising target for further studies.

View Article: PubMed Central - PubMed

Affiliation: Department of Physiology and Pharmacology, Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel.

ABSTRACT

Background: Pancreatic adenocarcinoma (PAC) is one of the most intractable malignancies. In order to search for potential new therapeutic targets, we relied on computational methods aimed at identifying transcription factor binding sites (TFBSs) over-represented in the promoter regions of genes differentially expressed in PAC. Though many computational methods have been implemented to accomplish this, none has gained overall acceptance or produced proven novel targets in PAC. To this end we have developed DEMON, a novel method for motif detection.

Methodology: DEMON relies on a hidden Markov model to score the appearance of sequence motifs, taking into account all potential sites in a promoter of potentially varying binding affinities. We demonstrate DEMON's accuracy on simulated and real data sets. Applying DEMON to PAC-related data sets identifies the RUNX family as highly enriched in PAC-related genes. Using a novel experimental paradigm to distinguish between normal and PAC cells, we find that RUNX3 mRNA (but not RUNX1 or RUNX2 mRNAs) exhibits time-dependent increases in normal but not in PAC cells. These increases are accompanied by changes in mRNA levels of putative RUNX gene targets.

Conclusions: The integrated application of DEMON and a novel differentiation system led to the identification of a single family member, RUNX3, which together with four of its putative targets showed a robust response to a differentiation stimulus in healthy cells, whereas this regulatory mechanism was absent in PAC cells, emphasizing RUNX3 as a promising target for further studies.

Show MeSH

Related in: MedlinePlus

Schematic of the DEMON's algorithm work flow.a. Retrieving a list of co-expressed genes from high-throughput experiments. b. For each HMM-promoter pair a score is computed as the ratio between the probability to emit the promoter sequence using the TFBS HMM and the probability to emit the promoter sequence using a background HMM. The sum of scores for each TF is used for computing a single score reflecting the TF's overall abundance in the input promoter set. c. Randomly selecting 100 promoter data sets with the same size as the original data set. Scores are calculated as before for those data sets. d. Each TF is assigned with an empirical p-value defined as the percentage of random cases in which it scored higher.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC3008686&req=5

pone-0014423-g002: Schematic of the DEMON's algorithm work flow.a. Retrieving a list of co-expressed genes from high-throughput experiments. b. For each HMM-promoter pair a score is computed as the ratio between the probability to emit the promoter sequence using the TFBS HMM and the probability to emit the promoter sequence using a background HMM. The sum of scores for each TF is used for computing a single score reflecting the TF's overall abundance in the input promoter set. c. Randomly selecting 100 promoter data sets with the same size as the original data set. Scores are calculated as before for those data sets. d. Each TF is assigned with an empirical p-value defined as the percentage of random cases in which it scored higher.

Mentions: Each HMM contains states for a unique motif, and background states that model inter-motif segments (Fig. 1). DEMON scores each promoter for the appearance of any given motif. This score reflects the probability that the sequence was generated based on the HMM describing the motif, vs. the probability that it was generated based on a simple background model. Given a target set of co-regulated genes, the scores of the promoters are summed up for each HMM, and compared to sums of scores obtained with random target sets. This comparison is used to assign a p-value for each motif that reflects its abundance in the promoter regions of the target set (see Fig. 2 and Methods).


A novel HMM-based method for detecting enriched transcription factor binding sites reveals RUNX3 as a potential target in pancreatic cancer biology.

Levkovitz L, Yosef N, Gershengorn MC, Ruppin E, Sharan R, Oron Y - PLoS ONE (2010)

Schematic of the DEMON's algorithm work flow.a. Retrieving a list of co-expressed genes from high-throughput experiments. b. For each HMM-promoter pair a score is computed as the ratio between the probability to emit the promoter sequence using the TFBS HMM and the probability to emit the promoter sequence using a background HMM. The sum of scores for each TF is used for computing a single score reflecting the TF's overall abundance in the input promoter set. c. Randomly selecting 100 promoter data sets with the same size as the original data set. Scores are calculated as before for those data sets. d. Each TF is assigned with an empirical p-value defined as the percentage of random cases in which it scored higher.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC3008686&req=5

pone-0014423-g002: Schematic of the DEMON's algorithm work flow.a. Retrieving a list of co-expressed genes from high-throughput experiments. b. For each HMM-promoter pair a score is computed as the ratio between the probability to emit the promoter sequence using the TFBS HMM and the probability to emit the promoter sequence using a background HMM. The sum of scores for each TF is used for computing a single score reflecting the TF's overall abundance in the input promoter set. c. Randomly selecting 100 promoter data sets with the same size as the original data set. Scores are calculated as before for those data sets. d. Each TF is assigned with an empirical p-value defined as the percentage of random cases in which it scored higher.
Mentions: Each HMM contains states for a unique motif, and background states that model inter-motif segments (Fig. 1). DEMON scores each promoter for the appearance of any given motif. This score reflects the probability that the sequence was generated based on the HMM describing the motif, vs. the probability that it was generated based on a simple background model. Given a target set of co-regulated genes, the scores of the promoters are summed up for each HMM, and compared to sums of scores obtained with random target sets. This comparison is used to assign a p-value for each motif that reflects its abundance in the promoter regions of the target set (see Fig. 2 and Methods).

Bottom Line: Using a novel experimental paradigm to distinguish between normal and PAC cells, we find that RUNX3 mRNA (but not RUNX1 or RUNX2 mRNAs) exhibits time-dependent increases in normal but not in PAC cells.These increases are accompanied by changes in mRNA levels of putative RUNX gene targets.The integrated application of DEMON and a novel differentiation system led to the identification of a single family member, RUNX3, which together with four of its putative targets showed a robust response to a differentiation stimulus in healthy cells, whereas this regulatory mechanism was absent in PAC cells, emphasizing RUNX3 as a promising target for further studies.

View Article: PubMed Central - PubMed

Affiliation: Department of Physiology and Pharmacology, Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel.

ABSTRACT

Background: Pancreatic adenocarcinoma (PAC) is one of the most intractable malignancies. In order to search for potential new therapeutic targets, we relied on computational methods aimed at identifying transcription factor binding sites (TFBSs) over-represented in the promoter regions of genes differentially expressed in PAC. Though many computational methods have been implemented to accomplish this, none has gained overall acceptance or produced proven novel targets in PAC. To this end we have developed DEMON, a novel method for motif detection.

Methodology: DEMON relies on a hidden Markov model to score the appearance of sequence motifs, taking into account all potential sites in a promoter of potentially varying binding affinities. We demonstrate DEMON's accuracy on simulated and real data sets. Applying DEMON to PAC-related data sets identifies the RUNX family as highly enriched in PAC-related genes. Using a novel experimental paradigm to distinguish between normal and PAC cells, we find that RUNX3 mRNA (but not RUNX1 or RUNX2 mRNAs) exhibits time-dependent increases in normal but not in PAC cells. These increases are accompanied by changes in mRNA levels of putative RUNX gene targets.

Conclusions: The integrated application of DEMON and a novel differentiation system led to the identification of a single family member, RUNX3, which together with four of its putative targets showed a robust response to a differentiation stimulus in healthy cells, whereas this regulatory mechanism was absent in PAC cells, emphasizing RUNX3 as a promising target for further studies.

Show MeSH
Related in: MedlinePlus