/AI4h ago

Transfer Learning Reduces Need for Massive Cell Context Readouts in Genomics

--0--
Comments
Anshul Kundaje#1675
Original post
Anshul Kundaje@anshulkundaje#1675inAI

IMO, we will not need massive quantities of such read outs & they will not be required in every possible cell context. There are effective strategies for learning from a few deeply profiled cell contexts (eg. cell lines) & transferring to all other cell contexts. 4/

Anshul Kundaje@anshulkundaje

This means knocking down / ablating regulatory elements, inserting elements in new contexts, arbitrary rearrangements, swaps, deletions etc in several cell contexts with coupled chromatin / RNA / cellular read outs. 3/

1:48 AM · Jun 4, 2026 · 206 Views
Sentiment
Sentiment building, check back later.
Cluster Engagement
-
Views
-
Comments
-
Reposts
-
Bookmarks
Expand data
Posts from X
Most Activity
Most ActivityTimeline
VIEWS276BOOKMARKS1LIKES4RETWEETS1
Anshul Kundaje@anshulkundaje

Please fund such tech dev and data generation efforts and please support academic labs where all this innovation is truly happening. 10/10

Anshul Kundaje@anshulkundaje

For cis regulation models, the next frontier is long range genome regulation. Big leaps here will also automatically translate to big leaps in discovering causal target genes of the large number of non coding genetic variants influencing risk for polygenic traits & diseases. 9/

4hViews 276Likes 4Bookmarks 1
REPLIES1
Anshul Kundaje@anshulkundaje

Genome perturbation experiments are going to be an extremely important data modality for obtaining truly causal models of gene regulation. 5/

Anshul Kundaje@anshulkundaje

IMO, we will not need massive quantities of such read outs & they will not be required in every possible cell context. There are effective strategies for learning from a few deeply profiled cell contexts (eg. cell lines) & transferring to all other cell contexts. 4/

4hViews 199Likes 3Bookmarks 0