Judith Kribelbauer
@jk-swietek.bsky.social
Gabilan Assistant Professor of Biological Sciences @ USC - Molecular and Computational Biology Section
same here. Have an 18-month-old and the only time I get anything done is at night when he sleeps 🫠 You are not alone, hang in there!
June 23, 2025 at 6:37 PM
same here. Have an 18-month-old and the only time I get anything done is at night when he sleeps 🫠 You are not alone, hang in there!
Agree, there are definitely cases where syntax is important, the question is more how to model it well when it isn't. I would think the latter is important for accuracy, which is where models tend to start underperforming (e.g. predicting activity for personalised genomes). I like your suggestion!
May 28, 2025 at 5:52 PM
Agree, there are definitely cases where syntax is important, the question is more how to model it well when it isn't. I would think the latter is important for accuracy, which is where models tend to start underperforming (e.g. predicting activity for personalised genomes). I like your suggestion!
might be time to reconsider how we train machine learning models. Almost every model I come across uses convolutional filters that attempt to learn syntax, when perhaps none is present...
May 27, 2025 at 7:34 PM
might be time to reconsider how we train machine learning models. Almost every model I come across uses convolutional filters that attempt to learn syntax, when perhaps none is present...
Thank you! Very complementary to the synthetic platform your lab built. Still challenging to go beyond 5kb for us, so tackling distal enhancer regulation will take a lot more engineering! Looking forward to future work from your lab!
December 18, 2024 at 9:05 PM
Thank you! Very complementary to the synthetic platform your lab built. Still challenging to go beyond 5kb for us, so tackling distal enhancer regulation will take a lot more engineering! Looking forward to future work from your lab!
Scalability is definitely the next big step in the journey. Also in the works a universal replacement cassette. With the current throughput it can still be useful for benchmarking/providing insights into potential shortcoming of "AI" tools.
December 17, 2024 at 9:39 PM
Scalability is definitely the next big step in the journey. Also in the works a universal replacement cassette. With the current throughput it can still be useful for benchmarking/providing insights into potential shortcoming of "AI" tools.
Thank you! Been a long time coming...
December 16, 2024 at 8:59 PM
Thank you! Been a long time coming...
Looking forward to feedback from the community. We want to move into new applications (SynBio, genAI, ...), reach out if interested!
December 16, 2024 at 2:39 PM
Looking forward to feedback from the community. We want to move into new applications (SynBio, genAI, ...), reach out if interested!
Stay tuned, we have some exciting collaborations in the works (@lucapinello.bsky.social lab), applying EXTRA-seq to test cell type specificity & the dynamic range of synthetically created sequences.
December 16, 2024 at 2:39 PM
Stay tuned, we have some exciting collaborations in the works (@lucapinello.bsky.social lab), applying EXTRA-seq to test cell type specificity & the dynamic range of synthetically created sequences.
We evaluated combinatorial E-P modifications, finding a new role for the TATA box, which acts as a 'noise suppressor' to prevent expression from weak (1 TFBS) "enhancers". With increasing E strength it switches to activation thus expanding the dynamic range.
December 16, 2024 at 2:39 PM
We evaluated combinatorial E-P modifications, finding a new role for the TATA box, which acts as a 'noise suppressor' to prevent expression from weak (1 TFBS) "enhancers". With increasing E strength it switches to activation thus expanding the dynamic range.
It also allows testing "out-of context" enhancer activities and what enhancer-native molecular features correlate with expression.
December 16, 2024 at 2:39 PM
It also allows testing "out-of context" enhancer activities and what enhancer-native molecular features correlate with expression.
We sys compare it to plasmid-based MPRAs and deep learning models (Enformer @DeepMind), finding new bio insights & suggesting ways to improve predictions.
We resolve expression differences down to distinct TF configurations on DNA, highlighting shortcomings of MPRAs and DL tools.
We resolve expression differences down to distinct TF configurations on DNA, highlighting shortcomings of MPRAs and DL tools.
December 16, 2024 at 2:39 PM
We sys compare it to plasmid-based MPRAs and deep learning models (Enformer @DeepMind), finding new bio insights & suggesting ways to improve predictions.
We resolve expression differences down to distinct TF configurations on DNA, highlighting shortcomings of MPRAs and DL tools.
We resolve expression differences down to distinct TF configurations on DNA, highlighting shortcomings of MPRAs and DL tools.
EXTRA-seq can be targeted to endogenous loci, allowing us to measure "near-native" gene expression as a function of dozens of newly introduced E-P genotypes (over 2kb apart).
December 16, 2024 at 2:39 PM
EXTRA-seq can be targeted to endogenous loci, allowing us to measure "near-native" gene expression as a function of dozens of newly introduced E-P genotypes (over 2kb apart).