Degui Zhi
@zhizhid
Followers
1K
Following
10K
Statuses
2K
Professor Biomedical informatics. MedAI (Co-creator of Med-BERT), AI imaging genetics (https://t.co/EIdKENaGOD), popgen (PBWT and IBD geek), and other stuff.
Houston, TX
Joined April 2009
Here is the paper! We showed rather than using human-defined enhancers, AI can learn to mark (mask) the potentially functional (causal) part of the genome and improve gene expression prediction. Kudos to Xingyu Su & @ShuiwangJi !
Excited to finally getting my first AI conference paper! Even though I am just a middle author, but still very excited about the potential of this work.
1
0
11
Congratulations to Pramesh Shakya for leading our latest PBWT paper accepted at RECOMB: Dynamic mu-PBWT: Dynamic Run-length Compressed PBWT for Biobank Scale Data, with Ahsan Sanaullah and @ShaojieZhang_
🎉Congratulations to the authors!🎉 The list of accepted papers for #RECOMB2025 is now live ➡️ What are your plans for Seoul? Let us know! #seoul
0
1
14
You know NIH is under stress when your proposal is assigned to the panel that are invited to serve as reviewer
0
0
3
Excited to finally getting my first AI conference paper! Even though I am just a middle author, but still very excited about the potential of this work.
0
1
21
🤗
In Top10 @ScienceMagazine 2024 breakthroughs of the year: Reconstructing ancient pedigrees + IBD segment sharing in aDNA! 😮😊 🎉
0
0
4
RT @ClementDelangue: Just 10 days after o1's public debut, we’re thrilled to unveil the open-source version of the groundbreaking technique…
0
636
0
@random_state_0 @han_chen_phd Good question! the z vectors are not assumed to be independent. That's why we have the R matrix in the middle.
0
0
1
@ShihchengGuo @han_chen_phd Thank you for your kind words. Fighting AD and brain diseases are a community efforts! We are doing our share as method developers. Much more work is needed to lead to drug or MOA.
0
0
0
Compared to other multi-phenotype statistical methods, JAGWAS are much faster and light weighted. Plus it can output p-values < 5*10^-324, breaking the numerical limit of double float, useful when you have 128 dimensions!
0
0
2
@karpathy @DBahdanau The rest are all history! I have odd feeling teaching my deep learning course. It is now like a AI history course as all the great models, RNNs, LSTMs, even CNN to an extend, are becoming obsolete.
0
0
1
Looking at NIH uses of genAI for grant review: While local institutions hosted LLM is prohibited, how about the LLM hosted on my laptop with no access to internet? I am not reviewing any grants right now, just ask for a friend
0
0
3
1
0
2