Happy to share our new method F-MoDA to find motifs in genomic attribution maps just published in MLCB 2025 proceedings! F-MoDA has improved specificity, sensitivity, accuracy, conciseness, and runtime compared to TF-MoDISco. Led by Ofir Yaish: proceedings.mlr.press/v311/yaish25...

I'm confused. If you haven't run TF-Modisco, how do you know what motifs the model is learning. Are you testing marginal sensitivity to a known motif?

TF-MoDISco: Transcription Factor Motif Discovery from
Importance Scores (2017) arxiv.org/abs/1811.00416

YouTube: youtube.com/watch?v=fXPGVJg956E
GitHub: github.com/kundajelab/t...

We used model interpretation techniques (DeepLIFT/SHAP & TF-MoDISco) to score the contribution of every nucleotide to accessibility, and discover recurrent patterns of sequences predictive of local chromatin accessibility - and these patterns turned out to primarily resemble TF binding motifs!

What does this pipeline command do?

1. Fit a model (still under development)
2. Predictions
3. Attributions
4. Seqlet Calling + Annotation (new)
5. TF-MoDISco
6. Marginalizations

I haven't considered the quality of the data too much, and I haven't yet run TF-MoDISco, but the trained models does seem to make accurate predictions and also respond strongly to an IRF-like motif.

We've tried this with our TF-MODISCO approach for motif discovery from deep learning models. We've never officially published it in a journal so far. It's been a preprint on arxiv for over 6 years now. Widely used by the community.

We used model interpretation techniques (DeepLIFT/SHAP & TF-MoDISco) to score the contribution of every nucleotide to accessibility, and discover recurrent patterns of sequences predictive of local chromatin accessibility - and these patterns turned out to primarily resemble TF binding motifs!

Does "motif finding" across all the models (TF-MODISCO). Lots of known stuff but also novel but also some spacing grammars - what Anshul calls "hard syntax" when it is a specific base pairs and then "soft syntax" when they are synergisitic nearby but not specific

Trump targets India again over trade and tariffs, calls business ties 'one-sided' yespunjab.com?p=158383

#USIndiaTrade #TrumpRemarks #TradeTensions #IndiaUSRelations #ModiSCO #GlobalTrade #TariffDebate #RussianOil #EconomicNews #InternationalRelations #September2025News

3) "The sequence grammar of cerebellar CREs is
conserved across mammals."

Enhancer grammar is one of those elusive terns that I think we're going to crack with all these "Oracles" in the coming decade. I love the MoDISco plots. @grantonjindal.bsky.social has a nice recent review on grammar here:

4) This technique which they call "Interspecies Point Projection (IPP)" ultimately shows rapid binding-site turnover. Again, I love these MoDISco plots. Really strong evidence for Emma Farley's "Dependency Grammar".

4) By aggregating recurring high importance sequences, we can find the preferences of previously known TFs (as “motifs”), helping identify which TFs are active in regulation, and discover novel, cell type-specific binding modes (TF-MoDISco: doi.org/10.48550/arX...). 16/

[3/4] announced the results of the investigation and stated that the Colombian rebels led by the alias "Ivan Modisco" carried out the attack and offered a reward of 200 million pesos (approximately 385,000 yuan) to find the murderer. On the afternoon of the 25th, a bus traveling on the

EFTA01260538.jpg
#epsteinweb #efta01260538
https://epsteinweb.org
Available in the iOS app store now!
https://apps.apple.com/us/app/epstein-web/id6758880661

excelling in complex genomic data analysis. Furthermore, interpretability investigations using DeepLIFT and Transcription Factor Motif Discovery from the importance scores algorithm (TF-MoDISco) demonstrate that the derived motifs from our model [7/8 of https://arxiv.org/abs/2505.09883v1]

Although there are some challenges with simply mapping seqlets to motif databases, this can be viewed as a fast + dirty alternative to the robust de novo motif discovery of TF-MoDISco. It'll just give you a sense for what your model has learned (if anything at all)!