Shane Kuei-Hsien Chu

cs phd • interpretability for sequence to function models • cuda, sparse codes

[Shane Kuei-Hsien Chu @guixien]

@fulop_dan Ain't gonna joke about it. Columbia postdocs were out on the street rallying over pay a few days ago.

[Shane Kuei-Hsien Chu @guixien]

@shae_mcl And not necessarily a good thing imo. The main roadblock is that lab work needs institutions. Don’t you think it’s wonderful that 4 dudes from undergrad can start Cursor?

[Shane Kuei-Hsien Chu @guixien]

@jxmnop This makes it sound like learning cuda is a big overhead — it really is not, especially if you’re using language like Julia (ie CUDA.jl). I don’t write cuda kernel code everyday, but it’s extremely handy when I need it

[Shane Kuei-Hsien Chu @guixien]

@halfacrocodile Cool. Will try to post something in the next couple of days.

[WashU Medicine Genetics @WashUGenetics]

The Inaugural Gary D. Stormo Computational & Systems Biology Lectureship was hosted at the Department of Genetics on October 3rd, 2024 with Dr. Michael Brent delivering a talk. Congrats to the 2024 Stormo Fellow, Reilly Sample @ReillyAlexS!!
https://t.co/hBN87OTzci

[Shane Kuei-Hsien Chu @guixien]

@PETERJIN999 @Q_StatGen @Jiacheng_Miao @NatureGenet Congrats @Jiacheng_Miao !

[Gary Stormo @gdstormo]

Today is my last day as a Professor. Tomorrow I become Professor Emeritus. The 50+ years I've been in the field has seen revolutionary advances in both biology and computer science and it has been a joy to in the midst of that.

[Shane Kuei-Hsien Chu @guixien]

@halfacrocodile @dmitrypenzar @WWenya @DariaNogina @ZinkevichA Great to know! Thanks for sharing. Appreciated.

[Shane Kuei-Hsien Chu @guixien]

@jmschreiber91 For nucleotide motifs one can use sparse representations. Interpretable and far less parameters needed than NN that use attention. Reveals cooperative binding patterns of TFs. https://t.co/CZDEdX1Jwn

[Shane Kuei-Hsien Chu @guixien]

@jmschreiber91 I think so. It can capture the gapped motifs, eg. MYOD1 co-occurs with MEF2A that characterized by a large number of spacers. This image is a result from https://t.co/8sEybjhp6K. Usually, there are lots of patterns like this in a typical ChIP-Seq dataset.

[Shane Kuei-Hsien Chu @guixien]

@bio2data Omg you made it. Cool!!!

[Shane Kuei-Hsien Chu @guixien]

@OfficialLoganK @eliascarvdev @JuliaLanguage Lots of disciplines require the ease of creating compiled executables (eg bioinformatics). Julia is much less mature in that regard.

[Shane Kuei-Hsien Chu @guixien]

@trishankkarthik @AlexZevelev Dude Julia is awesome

[Shane Kuei-Hsien Chu @guixien]

@TuXinming @lal_avantika This sounds like a separate problem to motif discovery to me.

[Shane Kuei-Hsien Chu @guixien]

@trishankkarthik @AlexZevelev @HarryDCrane @JuliaLanguage @dzviovich Writing GPU kernel code in Julia for scientific computing tasks is incredibly easy.

[Universal Curiosity @UniverCurious]

You can change the train direction just by thinking about it.