Graphs to be added as automatic retrieval

[LucaCappelletti94]

We would like to add some more graphs to the automatic retrieval mechanism.

Currently, we support only StringPPI (human version), CompleteStringPPI (cross-species) and KG-COVID-19.

Which graphs should we add to the list? The requirements for the graph are:

1. Must be publicly available behind an URL that can be resolved with a wget.
2. Must be a TSV/CSV/text file with separators.
3. The server where it is hosted must be reasonably fast.
4. Can be a zip, gzip, tar.gz or plain file.

[LucaCappelletti94]

Karate

[LucaCappelletti94]

Finish adding graphs from Network Repository,

Most of these are now available, we still need to add support for timestamp graphs and graphs with multi-labelled nodes.

[LucaCappelletti94]

Added graphs from kghub with this pull request.

[callahantiff]

@LucaCappelletti94 - what about including any of the resources in this (also copied below) table? If you want, we could select a few to start with. I'd be happy to write some simple code that converts them from PyKeen format into the spec you provide above. Let me know what you think!

I put a ➡️ next to the ones I think are worth starting with and a ⭐ next others worth considering for future incorporation.

Name	Reference	Description
⭐ckg	pykeen.datasets.CKG	The Clinical Knowledge Graph (CKG) dataset from [santos2020]_.
➡️ codexlarge	pykeen.datasets.CoDExLarge	The CoDEx large dataset.
codexmedium	pykeen.datasets.CoDExMedium	The CoDEx medium dataset.
codexsmall	pykeen.datasets.CoDExSmall	The CoDEx small dataset.
➡️ conceptnet	pykeen.datasets.ConceptNet	The ConceptNet dataset from [speer2017]_.
⭐cskg	pykeen.datasets.CSKG	The CSKG dataset.
⭐drkg	pykeen.datasets.DRKG	The DRKG dataset.
fb15k	pykeen.datasets.FB15k	The FB15k dataset.
fb15k237	pykeen.datasets.FB15k237	The FB15k-237 dataset.
⭐ hetionet	pykeen.datasets.Hetionet	The Hetionet dataset is a large biological network.
➡️ kinships	pykeen.datasets.Kinships	The Kinships dataset.
nations	pykeen.datasets.Nations	The Nations dataset.
ogbbiokg	pykeen.datasets.OGBBioKG	The OGB BioKG dataset.
ogbwikikg	pykeen.datasets.OGBWikiKG	The OGB WikiKG dataset.
➡️ openbiolink	pykeen.datasets.OpenBioLink	The OpenBioLink dataset.
openbiolinkf1	pykeen.datasets.OpenBioLinkF1	The PyKEEN First Filtered OpenBioLink 2020 Dataset.
openbiolinkf2	pykeen.datasets.OpenBioLinkF2	The PyKEEN Second Filtered OpenBioLink 2020 Dataset.
openbiolinklq	pykeen.datasets.OpenBioLinkLQ	The low-quality variant of the OpenBioLink dataset.
umls	pykeen.datasets.UMLS	The UMLS dataset.
wn18	pykeen.datasets.WN18	The WN18 dataset.
wn18rr	pykeen.datasets.WN18RR	The WN18-RR dataset.
yago310	pykeen.datasets.YAGO310	The YAGO3-10 dataset is a subset of YAGO3 that only contains entities with at least 10 relations.

[callahantiff]

Also, see the datasets listed in this KG Embedding Review on the bottom of page 22. These are datasets that are most frequently used by people developing new KG embedding methods:

[LucaCappelletti94]

Thank you, @callahantiff! We still need to add support for multi-class support for the nodes (that is, nodes with multiple classes such as a node that is both of class mammal and class cat). Even though we plan to add support for these and other node and edge features, we will surely work on them after finishing Grape. Do you know if these graphs have multiple classes per nodes or just nodes of multiple classes, with each node of a single class? [UPDATE 2021/04/19] We have support for multi-labeled nodes in graphs now!

If it's the second option, then we can surely support now all the considered graphs. How hard is it to convert them into a CSV-like format? And more importantly, where could we host these? Maybe on kg-hub? Would that be an option @justaddcoffee?

[justaddcoffee]

My general feeling is that we can and should allow easy ingest of remote graphs as we are discussing here.

But, I think we should avoid hosting other people's graphs on KG-hub unless they are transformed versions that we are incorporating into our own knowledge graphs (like our ChEMBL transform that we include in KG-COVID-19).

Glad to discuss though

[LucaCappelletti94]

Import graphs, after adding support for time intervals, from http://www.sociopatterns.org/

[hrshdhgd]

Hey @LucaCappelletti94 , as discussed earlier here's the link for kg-microbe graphs: https://kg-hub.berkeleybop.io/kg-microbe/20210422/kg-microbe.tar.gz

[LucaCappelletti94]

Thank you @hrshdhgd!

[LucaCappelletti94]

Hi @hrshdhgd, sorry for the long wait, now all versions of KG-Microbe and KG-COVID are integrated in the automatic retrieval.

[hrshdhgd]

No problem @LucaCappelletti94 , thank you very much!

[LucaCappelletti94]

I am now iterating once more on the graphs from the automatic graph retrieval (we are now at over 80K graphs downloadable). Do you have more suggestions?