Shane

comp bio postdoc @ Mount Sinai, lover of the long shot

[Shane @shane__tweets]

great to see this work published friday in bp-cnni! read the full paper with added bells and whistles here https://t.co/2bc0raDFFt

[Shane @shane__tweets]

very happy to share our latest work which is online today! we carried out bi-directional mendelian randomization analyses of multiple brain imaging phenotypes and bipolar disorder. a short thread on some of our main findings (much more in main text) 1/ https://t.co/vsMqexB6KD

[Pilib Ó Broin @pilib_]

I'm recruiting two research associates/postdocs in #cancer #bioinformatics. Closing date is Jan 2nd. RTs appreciated! https://t.co/84l3Mav0pE

[Shane @shane__tweets]

and of course please feel free to provide feedback! inspre details here: https://t.co/ksX06w18rS #Genomics #MendelianRandomization /14.

[Shane @shane__tweets]

very happy to share our latest work which is online today! we carried out bi-directional mendelian randomization analyses of multiple brain imaging phenotypes and bipolar disorder. a short thread on some of our main findings (much more in main text) 1/ https://t.co/vsMqexB6KD

[Shane @shane__tweets]

this work was made possible by the efforts of a great many people, particularly my wonderful PI @NiamhAMullins. read the full manuscript for more details and recreate all figures using our github repo notebook https://t.co/UR29ipOULV (interactive networks also available!) 13/

[Shane @shane__tweets]

It’s a proof at least of the prediction/inference dichotomy between deep learning and statistics in practice - solving by definition in the deep learning world means an accurate output, whereas a solution looks very different from a classical statistical perspective (both are ok)

[Shane @shane__tweets]

Can these AI approaches tell you anything meaningful about the data-generating distribution? most deep learning approaches require a large degree of reverse engineering to even approximate the parameters of the underlying distribution. very subjective definition of a solution

[Alexia Jolicoeur-Martineau @jm_alexia]

Generating new samples from an unknown distribution given a finite set of samples (training dataset) is a fundamental statistical problem.

Yet Statisticians haven't touched the problem, while AI researchers have solved it (diffusion, GANs, VAE, LLMs). 🤷‍♀️

[Shane @shane__tweets]

Very fun talking about some recent results at #WCPG2024 yesterday (beer for scale)

[Shane @shane__tweets]

@Shlingeen I’ll take one please 🙏