Functional motif detection via in silico ablation using AlphaGenome

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Identifying functional transcription factor (TF) binding sites for enhancer activity typically requires extensive ChIP-based assays or perturbation experiments.
• Large sequence-to-function foundation models offer potential for predicting regulatory roles directly from DNA sequence.

Purpose of the Study:

• To introduce and validate a virtual motif-perturbation framework for inferring TF motif-level regulatory contribution from DNA sequence.
• To assess the concordance between in silico predictions and experimental measurements of TF-dependent regulatory activity.

Main Methods:

• Utilized AlphaGenome, a large sequence-to-function foundation model, to perform in silico motif ablation.
• Identified candidate C/EBPβ motifs in chromatin-active regions and computationally perturbed them.
• Quantified changes in predicted regulatory activity and compared in silico predictions with H3K27ac CUT&RUN data following CEBPB knockout.

Main Results:

• The virtual motif-perturbation framework successfully inferred the regulatory contribution of C/EBPβ motifs.
• In silico predictions showed strong concordance with experimental H3K27ac changes, validating the model's accuracy.
• Discrepancies between predictions and experiments highlighted biologically interpretable mechanisms.

Conclusions:

• Large sequence-based models can effectively approximate the functional consequences of TF perturbations.
• Sequence-only virtual ablation provides a scalable framework for TF motif analysis, hypothesis generation, and regulatory annotation.
• This approach can be extended to various TFs, cell types, and chromatin states.