Limits...
iRegulon: from a gene list to a gene regulatory network using large motif and track collections.

Janky R, Verfaillie A, Imrichová H, Van de Sande B, Standaert L, Christiaens V, Hulselmans G, Herten K, Naval Sanchez M, Potier D, Svetlichnyy D, Kalender Atak Z, Fiers M, Marine JC, Aerts S - PLoS Comput. Biol. (2014)

Bottom Line: Identifying master regulators of biological processes and mapping their downstream gene networks are key challenges in systems biology.Next, we use iRegulon on more challenging types of gene lists, including microRNA target sets, protein-protein interaction networks, and genetic perturbation data.Similarly we map a repressive network with no indication of direct p53 regulation but rather an indirect effect via E2F and NFY.

View Article: PubMed Central - PubMed

Affiliation: Laboratory of Computational Biology, KU Leuven Center for Human Genetics, Leuven, Belgium.

ABSTRACT
Identifying master regulators of biological processes and mapping their downstream gene networks are key challenges in systems biology. We developed a computational method, called iRegulon, to reverse-engineer the transcriptional regulatory network underlying a co-expressed gene set using cis-regulatory sequence analysis. iRegulon implements a genome-wide ranking-and-recovery approach to detect enriched transcription factor motifs and their optimal sets of direct targets. We increase the accuracy of network inference by using very large motif collections of up to ten thousand position weight matrices collected from various species, and linking these to candidate human TFs via a motif2TF procedure. We validate iRegulon on gene sets derived from ENCODE ChIP-seq data with increasing levels of noise, and we compare iRegulon with existing motif discovery methods. Next, we use iRegulon on more challenging types of gene lists, including microRNA target sets, protein-protein interaction networks, and genetic perturbation data. In particular, we over-activate p53 in breast cancer cells, followed by RNA-seq and ChIP-seq, and could identify an extensive up-regulated network controlled directly by p53. Similarly we map a repressive network with no indication of direct p53 regulation but rather an indirect effect via E2F and NFY. Finally, we generalize our computational framework to include regulatory tracks such as ChIP-seq data and show how motif and track discovery can be combined to map functional regulatory interactions among co-expressed genes. iRegulon is available as a Cytoscape plugin from http://iregulon.aertslab.org.

Show MeSH

Related in: MedlinePlus

Evaluation of iRegulon and comparison to other methods.The TF recovery (y-axis) corresponds to the fraction of TFs correctly detected among all TFs for which a motif from our library can be associated. A. Positive sets consist of the top 200 genes ranked by the maximum signal value of the ChIP-Seq peak in the corresponding search space. Control sets are negatives from ENCODE (genes without a ChIP-Seq peak); TF neighborhoods (TFNB; all TFs within 5 Mb around a query TF); and random signatures (RND). The color (from red to yellow) and order of stacked bars indicate the number of times the queried TF was identified in the 1st rank (top1), 2nd rank (top2), 3rd rank (top3), 4th rank (top4), 5th rank (top5) and 6th to 10th rank (top10). White color indicates the number of detected TFs (motif enrichment ≥3) but with rank >10. B. Positives are mixed with negative genes (noise) from 0% to 100% of noise. The lines represent the sensitivity (Sn), Specificity (Sp), and Precision or Positive Predictive Value (PPV) of target gene selection. C. The layers of motif2TF increase the performance. Recovery for ENCODE signatures and their control sets using different motif2TF parameters: 1) Motif collection effect (J, T, A barcharts), 2) Homology effect using a threshold on Identity% for all motifs (A+O barcharts), 3) Motif similarity effect using a threshold on the p-value (A+S barcharts), and combinations (A+O+S). Only Jaspar motifs (J); Only Transfac Pro (T); All motifs from Jaspar and Transfac pro, and others databases (A); All motifs+Orthology (A+O); and All motifs+Orthology+Similarity (A+O+S); blue indicates the analysis done on ENCODE sets and grey indicates on the control sets. D. Tool comparison using a benchmark of 30 gene sets constructed as the top 200 target genes based on ChIP peak occurrences in the 20 kb regulatory region for 30 TFs (these TFs were selected from FactorBook having their canonical motif as top enriched in the actual ChIP peaks). The number of times the queried TF was identified in the top1 (red) and top5 (yellow) is recorded. The dashed boxes represent top 5 recoveries if similar motifs are manually re-associated to the query TF. Default parameters were used, but when possible, they were adjusted to use the tss-centered-20 kb regions. See also Figures S2–S3 and Table S1.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4109854&req=5

pcbi-1003731-g002: Evaluation of iRegulon and comparison to other methods.The TF recovery (y-axis) corresponds to the fraction of TFs correctly detected among all TFs for which a motif from our library can be associated. A. Positive sets consist of the top 200 genes ranked by the maximum signal value of the ChIP-Seq peak in the corresponding search space. Control sets are negatives from ENCODE (genes without a ChIP-Seq peak); TF neighborhoods (TFNB; all TFs within 5 Mb around a query TF); and random signatures (RND). The color (from red to yellow) and order of stacked bars indicate the number of times the queried TF was identified in the 1st rank (top1), 2nd rank (top2), 3rd rank (top3), 4th rank (top4), 5th rank (top5) and 6th to 10th rank (top10). White color indicates the number of detected TFs (motif enrichment ≥3) but with rank >10. B. Positives are mixed with negative genes (noise) from 0% to 100% of noise. The lines represent the sensitivity (Sn), Specificity (Sp), and Precision or Positive Predictive Value (PPV) of target gene selection. C. The layers of motif2TF increase the performance. Recovery for ENCODE signatures and their control sets using different motif2TF parameters: 1) Motif collection effect (J, T, A barcharts), 2) Homology effect using a threshold on Identity% for all motifs (A+O barcharts), 3) Motif similarity effect using a threshold on the p-value (A+S barcharts), and combinations (A+O+S). Only Jaspar motifs (J); Only Transfac Pro (T); All motifs from Jaspar and Transfac pro, and others databases (A); All motifs+Orthology (A+O); and All motifs+Orthology+Similarity (A+O+S); blue indicates the analysis done on ENCODE sets and grey indicates on the control sets. D. Tool comparison using a benchmark of 30 gene sets constructed as the top 200 target genes based on ChIP peak occurrences in the 20 kb regulatory region for 30 TFs (these TFs were selected from FactorBook having their canonical motif as top enriched in the actual ChIP peaks). The number of times the queried TF was identified in the top1 (red) and top5 (yellow) is recorded. The dashed boxes represent top 5 recoveries if similar motifs are manually re-associated to the query TF. Default parameters were used, but when possible, they were adjusted to use the tss-centered-20 kb regions. See also Figures S2–S3 and Table S1.

Mentions: To evaluate the performance of iRegulon, we derived direct target gene sets for 115 sequence-specific TFs from the ENCODE ChIP-Seq data [46], and for each target set we investigate whether the ChIP'ped TF can be correctly recovered (see Materials and Methods). Out of 115 tested TFs, iRegulon correctly identifies up to 94 TFs (82.6%) with Normalized Enrichment Scores (NES) above 3 (Fig. 2A, and Materials and Methods). We found iRegulon to be robust to noisy gene sets by adding increasing levels of noise (negative genes) to each set of targets (Fig. 2B). The motif2TF step is crucial to link an enriched motif to a candidate TF; and including motifs from other species and unknown motifs allow detecting many more correct regulators compared to using only known human motifs from TRANSFAC or JASPAR (Fig. 2C). After optimizing the parameters of iRegulon and motif2TF (see Materials and Methods and Fig. S2), we compared iRegulon with eight other motif discovery methods that use a similar input (a set of co-expressed genes) and generate a similar output (candidate regulators) using a non-ambiguous subset from Factorbook [46] (Materials and Methods). iRegulon identifies the correct TF at the first position in 17/30 cases while the other tools on average detect only 5.1/30 TFs at the first position (Fig. 2D, Table S1). Interestingly, the improved performances of iRegulon are not only due to the large PWM collection and the motif2TF mapping. Indeed, iRegulon still outperforms the other methods when using only the JASPAR collection and disabling the motif2TF step (Fig. S2C) or vice versa, when manually promoting similar motifs in the other tools to the correct TF (dashed bars in Fig. 2D). As expected, the true positive target gene recovery is significantly higher when iRegulon uses a 20 kb search space around TSS compared to using only the proximal promoter (Wilcoxon rank-sum paired test, p-value = 0.004) (Fig. S2D). We conclude that the core motif discovery framework of iRegulon is better than other tools, and that the large motif collection and the motif2TF step deliver a marked step forward in TF identification performance.


iRegulon: from a gene list to a gene regulatory network using large motif and track collections.

Janky R, Verfaillie A, Imrichová H, Van de Sande B, Standaert L, Christiaens V, Hulselmans G, Herten K, Naval Sanchez M, Potier D, Svetlichnyy D, Kalender Atak Z, Fiers M, Marine JC, Aerts S - PLoS Comput. Biol. (2014)

Evaluation of iRegulon and comparison to other methods.The TF recovery (y-axis) corresponds to the fraction of TFs correctly detected among all TFs for which a motif from our library can be associated. A. Positive sets consist of the top 200 genes ranked by the maximum signal value of the ChIP-Seq peak in the corresponding search space. Control sets are negatives from ENCODE (genes without a ChIP-Seq peak); TF neighborhoods (TFNB; all TFs within 5 Mb around a query TF); and random signatures (RND). The color (from red to yellow) and order of stacked bars indicate the number of times the queried TF was identified in the 1st rank (top1), 2nd rank (top2), 3rd rank (top3), 4th rank (top4), 5th rank (top5) and 6th to 10th rank (top10). White color indicates the number of detected TFs (motif enrichment ≥3) but with rank >10. B. Positives are mixed with negative genes (noise) from 0% to 100% of noise. The lines represent the sensitivity (Sn), Specificity (Sp), and Precision or Positive Predictive Value (PPV) of target gene selection. C. The layers of motif2TF increase the performance. Recovery for ENCODE signatures and their control sets using different motif2TF parameters: 1) Motif collection effect (J, T, A barcharts), 2) Homology effect using a threshold on Identity% for all motifs (A+O barcharts), 3) Motif similarity effect using a threshold on the p-value (A+S barcharts), and combinations (A+O+S). Only Jaspar motifs (J); Only Transfac Pro (T); All motifs from Jaspar and Transfac pro, and others databases (A); All motifs+Orthology (A+O); and All motifs+Orthology+Similarity (A+O+S); blue indicates the analysis done on ENCODE sets and grey indicates on the control sets. D. Tool comparison using a benchmark of 30 gene sets constructed as the top 200 target genes based on ChIP peak occurrences in the 20 kb regulatory region for 30 TFs (these TFs were selected from FactorBook having their canonical motif as top enriched in the actual ChIP peaks). The number of times the queried TF was identified in the top1 (red) and top5 (yellow) is recorded. The dashed boxes represent top 5 recoveries if similar motifs are manually re-associated to the query TF. Default parameters were used, but when possible, they were adjusted to use the tss-centered-20 kb regions. See also Figures S2–S3 and Table S1.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4109854&req=5

pcbi-1003731-g002: Evaluation of iRegulon and comparison to other methods.The TF recovery (y-axis) corresponds to the fraction of TFs correctly detected among all TFs for which a motif from our library can be associated. A. Positive sets consist of the top 200 genes ranked by the maximum signal value of the ChIP-Seq peak in the corresponding search space. Control sets are negatives from ENCODE (genes without a ChIP-Seq peak); TF neighborhoods (TFNB; all TFs within 5 Mb around a query TF); and random signatures (RND). The color (from red to yellow) and order of stacked bars indicate the number of times the queried TF was identified in the 1st rank (top1), 2nd rank (top2), 3rd rank (top3), 4th rank (top4), 5th rank (top5) and 6th to 10th rank (top10). White color indicates the number of detected TFs (motif enrichment ≥3) but with rank >10. B. Positives are mixed with negative genes (noise) from 0% to 100% of noise. The lines represent the sensitivity (Sn), Specificity (Sp), and Precision or Positive Predictive Value (PPV) of target gene selection. C. The layers of motif2TF increase the performance. Recovery for ENCODE signatures and their control sets using different motif2TF parameters: 1) Motif collection effect (J, T, A barcharts), 2) Homology effect using a threshold on Identity% for all motifs (A+O barcharts), 3) Motif similarity effect using a threshold on the p-value (A+S barcharts), and combinations (A+O+S). Only Jaspar motifs (J); Only Transfac Pro (T); All motifs from Jaspar and Transfac pro, and others databases (A); All motifs+Orthology (A+O); and All motifs+Orthology+Similarity (A+O+S); blue indicates the analysis done on ENCODE sets and grey indicates on the control sets. D. Tool comparison using a benchmark of 30 gene sets constructed as the top 200 target genes based on ChIP peak occurrences in the 20 kb regulatory region for 30 TFs (these TFs were selected from FactorBook having their canonical motif as top enriched in the actual ChIP peaks). The number of times the queried TF was identified in the top1 (red) and top5 (yellow) is recorded. The dashed boxes represent top 5 recoveries if similar motifs are manually re-associated to the query TF. Default parameters were used, but when possible, they were adjusted to use the tss-centered-20 kb regions. See also Figures S2–S3 and Table S1.
Mentions: To evaluate the performance of iRegulon, we derived direct target gene sets for 115 sequence-specific TFs from the ENCODE ChIP-Seq data [46], and for each target set we investigate whether the ChIP'ped TF can be correctly recovered (see Materials and Methods). Out of 115 tested TFs, iRegulon correctly identifies up to 94 TFs (82.6%) with Normalized Enrichment Scores (NES) above 3 (Fig. 2A, and Materials and Methods). We found iRegulon to be robust to noisy gene sets by adding increasing levels of noise (negative genes) to each set of targets (Fig. 2B). The motif2TF step is crucial to link an enriched motif to a candidate TF; and including motifs from other species and unknown motifs allow detecting many more correct regulators compared to using only known human motifs from TRANSFAC or JASPAR (Fig. 2C). After optimizing the parameters of iRegulon and motif2TF (see Materials and Methods and Fig. S2), we compared iRegulon with eight other motif discovery methods that use a similar input (a set of co-expressed genes) and generate a similar output (candidate regulators) using a non-ambiguous subset from Factorbook [46] (Materials and Methods). iRegulon identifies the correct TF at the first position in 17/30 cases while the other tools on average detect only 5.1/30 TFs at the first position (Fig. 2D, Table S1). Interestingly, the improved performances of iRegulon are not only due to the large PWM collection and the motif2TF mapping. Indeed, iRegulon still outperforms the other methods when using only the JASPAR collection and disabling the motif2TF step (Fig. S2C) or vice versa, when manually promoting similar motifs in the other tools to the correct TF (dashed bars in Fig. 2D). As expected, the true positive target gene recovery is significantly higher when iRegulon uses a 20 kb search space around TSS compared to using only the proximal promoter (Wilcoxon rank-sum paired test, p-value = 0.004) (Fig. S2D). We conclude that the core motif discovery framework of iRegulon is better than other tools, and that the large motif collection and the motif2TF step deliver a marked step forward in TF identification performance.

Bottom Line: Identifying master regulators of biological processes and mapping their downstream gene networks are key challenges in systems biology.Next, we use iRegulon on more challenging types of gene lists, including microRNA target sets, protein-protein interaction networks, and genetic perturbation data.Similarly we map a repressive network with no indication of direct p53 regulation but rather an indirect effect via E2F and NFY.

View Article: PubMed Central - PubMed

Affiliation: Laboratory of Computational Biology, KU Leuven Center for Human Genetics, Leuven, Belgium.

ABSTRACT
Identifying master regulators of biological processes and mapping their downstream gene networks are key challenges in systems biology. We developed a computational method, called iRegulon, to reverse-engineer the transcriptional regulatory network underlying a co-expressed gene set using cis-regulatory sequence analysis. iRegulon implements a genome-wide ranking-and-recovery approach to detect enriched transcription factor motifs and their optimal sets of direct targets. We increase the accuracy of network inference by using very large motif collections of up to ten thousand position weight matrices collected from various species, and linking these to candidate human TFs via a motif2TF procedure. We validate iRegulon on gene sets derived from ENCODE ChIP-seq data with increasing levels of noise, and we compare iRegulon with existing motif discovery methods. Next, we use iRegulon on more challenging types of gene lists, including microRNA target sets, protein-protein interaction networks, and genetic perturbation data. In particular, we over-activate p53 in breast cancer cells, followed by RNA-seq and ChIP-seq, and could identify an extensive up-regulated network controlled directly by p53. Similarly we map a repressive network with no indication of direct p53 regulation but rather an indirect effect via E2F and NFY. Finally, we generalize our computational framework to include regulatory tracks such as ChIP-seq data and show how motif and track discovery can be combined to map functional regulatory interactions among co-expressed genes. iRegulon is available as a Cytoscape plugin from http://iregulon.aertslab.org.

Show MeSH
Related in: MedlinePlus