PaperlayerView exact revision on arXiv

Source-linked AI summary

Detection of gene communities in multi-networks reveals cancer drivers

Laura Cantini, Enzo Medico, Santo Fortunato, Michele Caselle

arXiv:1507.08415v2q-bio.MNphysics.bio-phphysics.data-anphysics.soc-phq-bio.GN

TL;DR

Single-network analyses do not capture the coordinated genomic layers relevant to identifying cancer driver genes. This paper combines four gene-network layers and analyzes their communities; the authors report that the combined networks provide richer information than expression networks alone, including candidate genes and regulatory and functional context.

Abstract

from arXiv · show

We propose a new multi-network-based strategy to integrate different layers of genomic information and use them in a coordinate way to identify driving cancer genes. The multi-networks that we consider combine transcription factor co-targeting, microRNA co-targeting, protein-protein interaction and gene co-expression networks. The rationale behind this choice is that gene co-expression and protein-protein interactions require a tight coregulation of the partners and that such a fine tuned regulation can be obtained only combining both the transcriptional and post-transcriptional layers of regulation. To extract the relevant biological information from the multi-network we studied its partition into communities. To this end we applied a consensus clustering algorithm based on state of art community detection methods. Even if our procedure is valid in principle for any pathology in this work we concentrate on gastric, lung, pancreas and colorectal cancer and identified from the enrichment analysis of the multi-network communities a set of candidate driver cancer genes. Some of them were already known oncogenes while a few are new. The combination of the different layers of information allowed us to extract from the multi-network indications on the regulatory pattern and functional role of both the already known and the new candidate driver genes.

RESULTS

The study builds tissue-specific multi-networks from four genomic layers, filters dense layers for community detection, and compares multi-network communities with single-layer communities. Across the tested cancers, the communities combine information from multiple layers and show greater functional enrichment than expression-network communities.

RESULTS

  • The authors present the method as adaptable to expression datasets beyond the four cancers studied, while reporting tests supporting integration across genomic layers.The section distinguishes general construction and community detection from analyses performed on gastric, lung, pancreatic, and colon datasets.Source: RESULTS · p. 4

structure.

  • The network construction combined co-expression, TF co-targeting, microRNA co-targeting, and protein-protein interaction layers for normal and tumor tissues across four cancers.Co-expression networks were reconstructed from expression data using mutual information; TF co-targeting links counted shared TF regulators, while microRNA co-targeting used a similar reconstruction procedure.Source: Construction of the multi-network · pp. 4–5

Layers Filtering

  • Dense co-targeting layers and the complete co-expression graph motivated filtering before community detection; the disparity filter was more effective than global thresholding for these networks.The disparity filter's significance threshold α was selected using output density, balance of links across layers, and overlap with validated or predicted interactions.Source: Layers Filtering · pp. 5–6
  • Doubling or halving the selected α produced partitions with 99% overlap, indicating robustness to small changes in the filtering parameter.The authors evaluated parameter sensitivity by comparing community partitions obtained at those altered α values with the partition at the selected value.Source: Layers Filtering · p. 6

Community detection in the multi-network

  • The pipeline computes consensus partitions across 100 runs of five community-detection algorithms per layer, then combines the best layer partitions into a multi-network partition.The authors used OSLOM for subsequent analyses; community counts and size distributions differed by algorithm, including OSLOM's majority of communities in the 20–50 element range.Source: Community detection in the multi-network · pp. 6–7

significance.

  • Multi-network communities had NMI values well below 1, supporting the conclusion that their organization combines information from all layers rather than being driven by one layer.Expression networks had fewer communities than multi-networks in each tissue: 127 vs 158 in gastric, 118 vs 162 in lung, 137 vs 173 in pancreas, and 79 vs 178 in colon.Source: significance · p. 8
  • Tumor multi-network communities had more functionally homogeneous communities than expression communities in all four tissues, and performed better in grouping differentially expressed genes.The functional-homogeneity percentages were gastric 57% vs 63%, lung 71% vs 77%, pancreas 54% vs 67%, and colon 86% vs 88%; gastric and pancreas showed better performance in all differential-expression tests.Source: significance · p. 8
  • Enriched categories with p-values lower than 10^-5 were mostly tissue-specific, while chromosomal-location categories were enriched in more than one cancer type.The analysis selected tumor-associated functions by subtracting functions significant in normal communities, then identified functions not present in expression-network communities.Source: significance · p. 9

DISCUSSION

The discussion argues that multi-network communities provide richer information than expression networks alone for identifying candidate cancer drivers. Pancreatic examples connect chromosomal loci, cancer signatures, and miRNA regulons to candidate genes and their regulatory or functional roles.

DISCUSSION

  • The pancreatic 1q21 community contains five genes, including three known oncogenes: F11R, HDGF, and PRCC [32-34].The authors propose that multi-network loci can narrow hundreds of genes to those sharing a coregulatory scheme.Source: DISCUSSION · p. 10
  • In the pancreatic 1q32 community, enrichment of the DREAM pathway and JUN/FOS-family genes suggests a cooperative role in apoptosis and cell survival.Five community genes belong to 1q32; ATF3, BTG2, and CD46 are known pancreatic cancer markers [35-36].Source: DISCUSSION · p. 11
  • The pancreatic exocrine-like signature is enriched in community 2, which is enriched for HNF1 targets, suggesting a link to the HNF1 regulome [37-42].The analysis suggests a way to enlarge the signature and examine its regulatory pathway.Source: DISCUSSION · p. 11
  • Enriched miRNAs include miR-337 and miR-153, whose reported associations with PDAC outcomes support investigating other community-enriched miRNAs as potential prognostic markers [44-47].The analysis identifies enriched miRNAs across each cancer type; some are known and some are new.Source: DISCUSSION · p. 12
  • Multi-networks combining TF targeting, miRNA targeting, protein interactions, and co-expression provide significantly richer information than single expression networks.The authors say their algorithm could accommodate other genomic data, such as epigenetic markers.Source: DISCUSSION · p. 12

Contributions

L. C. S. F. E. M and M. C designed the research and wrote the paper; L. C. performed the research.

Contributions

  • L. C. S. F. E. M and M. C designed the research and wrote the paper; L. C. performed the research.Source: Contributions · p. 15

Figure Legends

The legends describe a multi-network diagram, a comparison of multi-network and expression-network communities, and enrichment radar plots across four tissues. The supplied text also describes a table of pancreatic community intersections with a cancer signature.

Figure Legends

  • The example multi-network has three layers and ten nodes, with the same nodes in every layer and dashed inter-layer links connecting each node to itself across layers.Intra-layer links are solid.Source: Figure 1 · p. 15, Figure Legends · p. 16
  • Figure 2 compares multi-network and expression-network communities using three differential-expression criteria: mean absolute log2 fold change, Student’s t-test p-value, and standard deviation of log2 fold change.Dots represent communities; darker dots mark communities that are also functionally homogeneous.Source: Figure Legends · pp. 16–17, Figure 2 · p. 16
  • Figure 3 radar plots show reciprocal enrichment p-values for chromosomes, pathways, TF/miRNA motifs, and GO across gastric, lung, pancreas, and colon cancers.Only functions with enrichment p-values below 10^-5 are shown; colors identify tissues.Source: Figure 3 · p. 17
  • The pancreatic signature-intersection table reports gene sets, significant community IDs, corrected hypergeometric p-values, set and community sizes, intersection sizes, and intersection genes.Source: Figure Legends · p. 18

Related papers

Loading 1507.08415v2…