=Paper=
{{Paper
|id=Vol-2796/presentation4
|storemode=property
|title=How to Teach a Computer to Learn About Microbes: KG-COVID-19 and Microbial Graph Learning - Abstract
|pdfUrl=https://ceur-ws.org/Vol-2796/xi-ml-2020-invited_talk_abstract_joachimiak.pdf
|volume=Vol-2796
|authors=Marcin P. Joachimiak
|dblpUrl=https://dblp.org/rec/conf/ki/Joachimiak20
}}
==How to Teach a Computer to Learn About Microbes: KG-COVID-19 and Microbial Graph Learning - Abstract==
<pdf width="1500px">https://ceur-ws.org/Vol-2796/xi-ml-2020-invited_talk_abstract_joachimiak.pdf</pdf>
<pre>
How to teach a computer to learn about microbes: KG-COVID-19 and microbial
graph learning - Abstract

Marcin P. Joachimiak, PhD
Environmental Genomics and Systems Biology Division
Lawrence Berkeley National Laboratory


Abstract

After many decades of research, we have become fairly skilled at modeling and
predicting some aspects of simpler model organisms, such as HIV, E. coli, or S.
cerevisiae. However, the COVID-19 pandemic has shown that this is far from true for
novel species. In fact, even for model organisms the data and tools required to build
highly collaborative and integrated models are still being developed. Although we
cannot make reliable predictions for most biological problems, in many cases we
have large collections of data with different modalities that have not been interlinked
or fully exploited. These data are often siloed by namespaces, APIs, and formats.
They are also affected by data modeling and analysis choices. Based on a core set
of design principles, we have developed a framework for constructing knowledge
graphs which allows to harmonize biological entities and their relationships across
many disparate data sources. We applied this framework to COVID-19 as well as
environmental genomics projects. Using new graph learning methods we can
leverage complex data to compute similarities between different biological entities,
something that has been difficult thus far. We are also applying graph embeddings to
perform link prediction tasks, focusing on target identification and drug repurposing
for COVID-19. While large knowledge graphs and associated methods are exciting
developments, their complexity requires new tools including for data search,
introspection, and visualization. Advances in these areas will be critical to achieving
explainability for both more traditional and new learning methods.




Copyright © 2020 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0
International (CC BY 4.0).

</pre>