Limits...
Annotation of genomics data using bidirectional hidden Markov models unveils variations in Pol II transcription cycle.

Zacher B, Lidschreiber M, Cramer P, Gagneur J, Tresch A - Mol. Syst. Biol. (2014)

Bottom Line: To overcome these limitations, we introduce bidirectional HMMs which infer directed genomic states from occupancy profiles de novo.Application to RNA polymerase II-associated factors in yeast and chromatin modifications in human T cells recovers the majority of transcribed loci, reveals gene-specific variations in the yeast transcription cycle and indicates the existence of directed chromatin state patterns at transcribed, but not at repressed, regions in the human genome.We anticipate bidirectional HMMs to significantly improve the analyses of genome-associated directed processes.

View Article: PubMed Central - PubMed

Affiliation: Gene Center and Department of Biochemistry, Center for Integrated Protein Science CIPSM, Ludwig-Maximilians-Universität München, Munich, Germany Institute for Genetics, University of Cologne, Cologne, Germany.

No MeSH data available.


Related in: MedlinePlus

Application of bdHMM to chromatin modifications in human T cells identifies direction of chromatin states.Example of chromatin state annotation of ChromHMM and bdHMM (bottom tracks) with RefSeq gene annotation and input signal. State direction matches gene orientation of annotated convergent genes and divergent genes. The log-transformed signal (Ernst and Kellis, 2010) of all 41 data tracks is shown in black on top. Binarized input signal is shown for 18 acetylation marks in blue, 20 methylation marks in red and CTCF/PolII/H2A.Z in brown.bdHMM transitions between promoter-associated states 1–11 are shown for forward and reverse states. The asymmetric, transposed structure of these two submatrices (i.e., transition probabilities favor one direction for pairs aij and aji) uncouple the two reading directions.The symmetric ChromHMM transition matrix hides the underlying directed flow of chromatin states.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4300491&req=5

fig06: Application of bdHMM to chromatin modifications in human T cells identifies direction of chromatin states.Example of chromatin state annotation of ChromHMM and bdHMM (bottom tracks) with RefSeq gene annotation and input signal. State direction matches gene orientation of annotated convergent genes and divergent genes. The log-transformed signal (Ernst and Kellis, 2010) of all 41 data tracks is shown in black on top. Binarized input signal is shown for 18 acetylation marks in blue, 20 methylation marks in red and CTCF/PolII/H2A.Z in brown.bdHMM transitions between promoter-associated states 1–11 are shown for forward and reverse states. The asymmetric, transposed structure of these two submatrices (i.e., transition probabilities favor one direction for pairs aij and aji) uncouple the two reading directions.The symmetric ChromHMM transition matrix hides the underlying directed flow of chromatin states.

Mentions: We evaluated the performance of bdHMM on sequencing data and large genomes, by applying bdHMM to a dataset of 41 chromatin marks in human T cells (Ernst & Kellis, 2010). The chromatin mark data had been binarized into presence/absence of each mark at a resolution of 200 bp bins and analyzed with a standard HMM approach (ChromHMM) (Ernst & Kellis, 2012). To handle the binarized chromatin marks data defined by Ernst and Kellis (2010), we extended bdHMM and included binary (Bernoulli) emission distributions. We fixed the emission distributions during bdHMM learning, allowing a direct comparison of bdHMM states to HMM states. Moreover, this ensured that differences in the result are only due to differences in the modeling of state transitions. We developed a directionality score (Materials and Methods) to decide that in the bdHMM, 35 out of a total of 51 ChromHMM states are modeled as directed state pairs and 16 ChromHMM states are modeled as undirected states. Consistently, we identified directed chromatin states around transcribed, but not at repressed or repetitive regions (Supplementary Fig S5). Up to state directionality, 83% of state annotations agreed between the two methods (Materials and Methods). Comparison of the ChromHMM with the bdHMM transitions revealed that in ChromHMM, transition probabilities between two states are similar in both directions (Fig6C), whereas the bdHMM can resolve the true order of chromatin states (Fig6A and 6B, Supplementary Fig S6). For example, transitions from state 6 into states 2 and 3 are high for the forward direction, but low for the reverse drection. In contrast, transitions from states 2 and 3 into state 6 are high in reverse, but low in forward, direction (Fig6B). However, all of these transitions are high in the symmetric ChromHMM model (Fig6C), demonstrating that bdHMM adds previously unexploited and valuable information to HMM-based analyses by uncoupling the underlying state directionality of genomic processes. Analysis of promoter and transcribed state frequencies at the TSS showed that state annotations matched the reading (sense) direction of the transcribed loci with up to 85% (Supplementary Fig S7). Promoter states showed pronounced peaks in sense direction at the TSS, which are further downstream followed by high frequencies of (sense) 5' proximal transcribed states. We conclude that bdHMM significantly improves the annotation of the human epigenome, because it correctly recovers the flow of chromatin states as they occur during transcription.


Annotation of genomics data using bidirectional hidden Markov models unveils variations in Pol II transcription cycle.

Zacher B, Lidschreiber M, Cramer P, Gagneur J, Tresch A - Mol. Syst. Biol. (2014)

Application of bdHMM to chromatin modifications in human T cells identifies direction of chromatin states.Example of chromatin state annotation of ChromHMM and bdHMM (bottom tracks) with RefSeq gene annotation and input signal. State direction matches gene orientation of annotated convergent genes and divergent genes. The log-transformed signal (Ernst and Kellis, 2010) of all 41 data tracks is shown in black on top. Binarized input signal is shown for 18 acetylation marks in blue, 20 methylation marks in red and CTCF/PolII/H2A.Z in brown.bdHMM transitions between promoter-associated states 1–11 are shown for forward and reverse states. The asymmetric, transposed structure of these two submatrices (i.e., transition probabilities favor one direction for pairs aij and aji) uncouple the two reading directions.The symmetric ChromHMM transition matrix hides the underlying directed flow of chromatin states.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4300491&req=5

fig06: Application of bdHMM to chromatin modifications in human T cells identifies direction of chromatin states.Example of chromatin state annotation of ChromHMM and bdHMM (bottom tracks) with RefSeq gene annotation and input signal. State direction matches gene orientation of annotated convergent genes and divergent genes. The log-transformed signal (Ernst and Kellis, 2010) of all 41 data tracks is shown in black on top. Binarized input signal is shown for 18 acetylation marks in blue, 20 methylation marks in red and CTCF/PolII/H2A.Z in brown.bdHMM transitions between promoter-associated states 1–11 are shown for forward and reverse states. The asymmetric, transposed structure of these two submatrices (i.e., transition probabilities favor one direction for pairs aij and aji) uncouple the two reading directions.The symmetric ChromHMM transition matrix hides the underlying directed flow of chromatin states.
Mentions: We evaluated the performance of bdHMM on sequencing data and large genomes, by applying bdHMM to a dataset of 41 chromatin marks in human T cells (Ernst & Kellis, 2010). The chromatin mark data had been binarized into presence/absence of each mark at a resolution of 200 bp bins and analyzed with a standard HMM approach (ChromHMM) (Ernst & Kellis, 2012). To handle the binarized chromatin marks data defined by Ernst and Kellis (2010), we extended bdHMM and included binary (Bernoulli) emission distributions. We fixed the emission distributions during bdHMM learning, allowing a direct comparison of bdHMM states to HMM states. Moreover, this ensured that differences in the result are only due to differences in the modeling of state transitions. We developed a directionality score (Materials and Methods) to decide that in the bdHMM, 35 out of a total of 51 ChromHMM states are modeled as directed state pairs and 16 ChromHMM states are modeled as undirected states. Consistently, we identified directed chromatin states around transcribed, but not at repressed or repetitive regions (Supplementary Fig S5). Up to state directionality, 83% of state annotations agreed between the two methods (Materials and Methods). Comparison of the ChromHMM with the bdHMM transitions revealed that in ChromHMM, transition probabilities between two states are similar in both directions (Fig6C), whereas the bdHMM can resolve the true order of chromatin states (Fig6A and 6B, Supplementary Fig S6). For example, transitions from state 6 into states 2 and 3 are high for the forward direction, but low for the reverse drection. In contrast, transitions from states 2 and 3 into state 6 are high in reverse, but low in forward, direction (Fig6B). However, all of these transitions are high in the symmetric ChromHMM model (Fig6C), demonstrating that bdHMM adds previously unexploited and valuable information to HMM-based analyses by uncoupling the underlying state directionality of genomic processes. Analysis of promoter and transcribed state frequencies at the TSS showed that state annotations matched the reading (sense) direction of the transcribed loci with up to 85% (Supplementary Fig S7). Promoter states showed pronounced peaks in sense direction at the TSS, which are further downstream followed by high frequencies of (sense) 5' proximal transcribed states. We conclude that bdHMM significantly improves the annotation of the human epigenome, because it correctly recovers the flow of chromatin states as they occur during transcription.

Bottom Line: To overcome these limitations, we introduce bidirectional HMMs which infer directed genomic states from occupancy profiles de novo.Application to RNA polymerase II-associated factors in yeast and chromatin modifications in human T cells recovers the majority of transcribed loci, reveals gene-specific variations in the yeast transcription cycle and indicates the existence of directed chromatin state patterns at transcribed, but not at repressed, regions in the human genome.We anticipate bidirectional HMMs to significantly improve the analyses of genome-associated directed processes.

View Article: PubMed Central - PubMed

Affiliation: Gene Center and Department of Biochemistry, Center for Integrated Protein Science CIPSM, Ludwig-Maximilians-Universität München, Munich, Germany Institute for Genetics, University of Cologne, Cologne, Germany.

No MeSH data available.


Related in: MedlinePlus