Literature-Based Interactomics Tool

Question

[Anima Mundi]

Hello,

I would like you to suggest me a literature-based interactomics tool. Thanks in advance.

[Khader Shameer]

What do you mean by "literature-based" here ? Text-mining or literature curation ?

[Anima Mundi]

As I would like to see only well supported interaction, a literature manual curation-based system would be better; anyway, I do not discard text-mining strategies.

Answer 1

[Casey Bergman]

iHop is a mature system that uses gene name recognition text-mining on MEDLINE abstracts and allows you to build a gene/protein network model by clicking and adding sentences with multiple gene names. From the iHop help pages:

Nodes in this graph represent genes. Edges correspond to sentences that associate two genes with each other...The model graph makes it possible to analyze your collection of sentences in an interactive manner and to get familiar with the newly aquainted knowledge. By clicking on a gene, all its synonyms will be highlighted in red and the corresponding sentences will be ranked to the top. By clicking on an edge, those sentences containing the connected genes are shown at the top. Furthermore, a line separates these highlighted sentences from all others...By clicking on a gene in a sentence, the corresponding information page for the gene will be opened and additional information can be added to the Gene Model.

Another option that has been around for some time is PubGene:

PubGene mines the abstract texts of 25 Million PubMed articles for co-citation of multiple genes or proteins and displays them as "Literature Networks", where nodes represent each gene or protein and the connecting lines represents the number of articles, in which each gene or protein pair is co-cited....When a gene or protein is studied, there is a good chance its name (or a synonym for that name) will appear in articles together with other gene or protein names. One can visualize how most genes that have been studied will be connected either directly or indirectly to each other in a Literature Network. Connections in the literature are a strong indicator of biological interaction.

[Anima Mundi]

iHop seems to be a really valid aid, PubGene is giving me some problem with javascripts. Thank you.

Answer 2

[Khader Shameer]

Protein-protein interactions reported in PPI databases are be broadly sourced using two approaches:

• Automated text-mining (biological / biomedical text mining; see this article for a background)
• Manual, literature curation (see the article on bigdata in biology and biocuration)

If you are looking at automated / biomedical text-mining based interactions:

I would start with iHOP, but I would cross-refer the interactions with another database (for example STRING or literature curated protein-protein interaction database Human Protein Reference Database (HPRD)) to get a better perspective on the interactions.

STRING aggregates protein-protein interactions from a variety of methods and the interactions can be scored based on different methods, in your case if you are only interested in interactions from text-mining method you can filter the interactions.

Please note that there could be potential false positives in these automated / text-mining based interactions, you may need to read about the literature context to see if they really make biological sense to your problem.

If you are looking for literature curated interactions (curation as in biocuration):

I would recommend HPRD (if you are interested in human proteins) and Experimentally derived interactions in generic protein-protein interaction databases like STRING-DB or BioGrid.

I am adding links to interactions recorded in the 4 databases discussed in this answere. I hope this will give you a cursory view into depth of interactions available in various databases discussed in this answer.

• Interactions from HPRD for CDK2
• Interactions from STRING-DB for CDK2
• Interactions from iHOP for CDK2
• Interactions from BioGrid for CDK2

[Anima Mundi]

Clear and helpful, thanks.

Answer 3

[Stephen]

Check out GRAIL: Gene Relationships Across Implicated Loci

GRAIL is a tool to examine relationships between genes in different disease associated loci. Given several genomic regions or SNPs associated with a particular phenotype or disease, GRAIL looks for similarities in the published scientific text among the associated genes.

Based on textual relationships between genes, GRAIL assigns a p-value to each region suggesting its degree of functional connectivity, and picks the best candidate gene.

I've never used it before myself but seems worth a try.

http://www.broadinstitute.org/mpg/grail/

[Anima Mundi]

It seems to be focused on pathology, but I will examine it anyway, thanks.