Kashyap Chhatbar
@kashyapchhatbar.bsky.social
PhD Computational Biology and Artificial Intelligence
Bioinformatics Researcher, Adrian Bird Lab
University of Edinburgh
Bioinformatics Researcher, Adrian Bird Lab
University of Edinburgh
🎯 Why this matters for research: This XAI approach does more than just explain-it predicts.
Genes ranked by our SHAP values are far more likely to be true direct targets than genes ranked by simple ChIP-seq signal. It even helps predict the magnitude of the transcriptional change!
Genes ranked by our SHAP values are far more likely to be true direct targets than genes ranked by simple ChIP-seq signal. It even helps predict the magnitude of the transcriptional change!
October 24, 2025 at 11:37 AM
🎯 Why this matters for research: This XAI approach does more than just explain-it predicts.
Genes ranked by our SHAP values are far more likely to be true direct targets than genes ranked by simple ChIP-seq signal. It even helps predict the magnitude of the transcriptional change!
Genes ranked by our SHAP values are far more likely to be true direct targets than genes ranked by simple ChIP-seq signal. It even helps predict the magnitude of the transcriptional change!
Reality check ✅
ZC3H4 (Restrictor) and INTS11 (Integrator) target many of the same genes-then move them in tandem (r=0.84). 📈
Mimoso et al. report ~50% overlap in Restrictor-sensitive mRNAs.
Together, this argues for deep functional coupling in transcription control 🧬
ZC3H4 (Restrictor) and INTS11 (Integrator) target many of the same genes-then move them in tandem (r=0.84). 📈
Mimoso et al. report ~50% overlap in Restrictor-sensitive mRNAs.
Together, this argues for deep functional coupling in transcription control 🧬
October 24, 2025 at 11:37 AM
Reality check ✅
ZC3H4 (Restrictor) and INTS11 (Integrator) target many of the same genes-then move them in tandem (r=0.84). 📈
Mimoso et al. report ~50% overlap in Restrictor-sensitive mRNAs.
Together, this argues for deep functional coupling in transcription control 🧬
ZC3H4 (Restrictor) and INTS11 (Integrator) target many of the same genes-then move them in tandem (r=0.84). 📈
Mimoso et al. report ~50% overlap in Restrictor-sensitive mRNAs.
Together, this argues for deep functional coupling in transcription control 🧬
🤯 Key Finding 2: We found an unexpected link between the Restrictor (ZC3H4) and Integrator (INTS11) complexes!
Our model showed that INTS11 target genes were also highly dependent on ZC3H4, and vice-versa. This points to major crosstalk! 🤝
Our model showed that INTS11 target genes were also highly dependent on ZC3H4, and vice-versa. This points to major crosstalk! 🤝
October 24, 2025 at 11:37 AM
🤯 Key Finding 2: We found an unexpected link between the Restrictor (ZC3H4) and Integrator (INTS11) complexes!
Our model showed that INTS11 target genes were also highly dependent on ZC3H4, and vice-versa. This points to major crosstalk! 🤝
Our model showed that INTS11 target genes were also highly dependent on ZC3H4, and vice-versa. This points to major crosstalk! 🤝
🔥Key Finding 1: ZC3H4 acts beyond promoters.
While its job at promoters is known, we revealed it also has a significant regulatory contribution within the gene body.
Consistent with ZC3H4 recruiting PNUTS-PP1 (Erickson; Kelly et al.) which functions across gene bodies #Epigenetics #GeneRegulation
While its job at promoters is known, we revealed it also has a significant regulatory contribution within the gene body.
Consistent with ZC3H4 recruiting PNUTS-PP1 (Erickson; Kelly et al.) which functions across gene bodies #Epigenetics #GeneRegulation
October 24, 2025 at 11:37 AM
🔥Key Finding 1: ZC3H4 acts beyond promoters.
While its job at promoters is known, we revealed it also has a significant regulatory contribution within the gene body.
Consistent with ZC3H4 recruiting PNUTS-PP1 (Erickson; Kelly et al.) which functions across gene bodies #Epigenetics #GeneRegulation
While its job at promoters is known, we revealed it also has a significant regulatory contribution within the gene body.
Consistent with ZC3H4 recruiting PNUTS-PP1 (Erickson; Kelly et al.) which functions across gene bodies #Epigenetics #GeneRegulation
But is it just correlation? 🤔 We validated our model's insights using degron experiments (for rapid protein degradation).
💥Genes most affected by the protein's removal were exactly the ones our model (trained on unperturbed data!) had flagged as most important. This shows functional relevance!
💥Genes most affected by the protein's removal were exactly the ones our model (trained on unperturbed data!) had flagged as most important. This shows functional relevance!
October 24, 2025 at 11:37 AM
But is it just correlation? 🤔 We validated our model's insights using degron experiments (for rapid protein degradation).
💥Genes most affected by the protein's removal were exactly the ones our model (trained on unperturbed data!) had flagged as most important. This shows functional relevance!
💥Genes most affected by the protein's removal were exactly the ones our model (trained on unperturbed data!) had flagged as most important. This shows functional relevance!
How does it work? 🤖 We trained an AI on data from unperturbed cells to predict gene transcription (RNA Pol-II occupancy).
The secret sauce? We fed it signals from both promoters and gene bodies, then used SHAP (explainable AI) to tell us exactly which proteins matter most, and where!
The secret sauce? We fed it signals from both promoters and gene bodies, then used SHAP (explainable AI) to tell us exactly which proteins matter most, and where!
October 24, 2025 at 11:37 AM
How does it work? 🤖 We trained an AI on data from unperturbed cells to predict gene transcription (RNA Pol-II occupancy).
The secret sauce? We fed it signals from both promoters and gene bodies, then used SHAP (explainable AI) to tell us exactly which proteins matter most, and where!
The secret sauce? We fed it signals from both promoters and gene bodies, then used SHAP (explainable AI) to tell us exactly which proteins matter most, and where!