Thread by @jamietmorton

I’m very excited to share our latest paper on infer microbe-metabolite relationships.
nature.com/articles/s4159…

with @LF_NOTHIAS @Quinn_Labs @michellebadri @marcwvg @yoshikivb @mingxunwang @Pdorrestein1 @KnightLabNews and many others (thread 👇)

@LF_NOTHIAS @Quinn_Labs @michellebadri @marcwvg @yoshikivb @mingxunwang @Pdorrestein1 @KnightLabNews 1/ Nailing down microbe-metabolite interactions from sequencing / MS data is key for understanding microbial ecosystem structure. Unfortunately, the vast majority statistical techniques available to do this perform WORSE than random, because of the nature of compositional data.

@LF_NOTHIAS @Quinn_Labs @michellebadri @marcwvg @yoshikivb @mingxunwang @Pdorrestein1 @KnightLabNews 2/ So if you are planning on using Pearson or Spearman correlation to infer microbe-metabolite interactions, don’t bother even collecting paired sequencing / MS – you are better of randomly rolling die and reporting those results.

@LF_NOTHIAS @Quinn_Labs @michellebadri @marcwvg @yoshikivb @mingxunwang @Pdorrestein1 @KnightLabNews 3/ To beat random chance, we developed an NN approach, we call mmvec to infer co-occurrence probabilities instead of correlation. With simulated / real benchmarks, we show some sanity is restored.

@LF_NOTHIAS @Quinn_Labs @michellebadri @marcwvg @yoshikivb @mingxunwang @Pdorrestein1 @KnightLabNews 4/ The code is available at github.com/biocore/mmvec with a qiime2 plugin.

@LF_NOTHIAS @Quinn_Labs @michellebadri @marcwvg @yoshikivb @mingxunwang @Pdorrestein1 @KnightLabNews 5/ Binder tutorials can also be found here : github.com/mortonjt/q2sta…

@LF_NOTHIAS @Quinn_Labs @michellebadri @marcwvg @yoshikivb @mingxunwang @Pdorrestein1 @KnightLabNews 6/ Finally, a blog post is available below
protocolsmethods.springernature.com/users/328822-j…