Functional motif detection via in silico ablation using AlphaGenome

Area of Science:

• Computational Biology
• Genomics
• Molecular Biology

Background:

• Determining the functional role of transcription factor (TF) binding sites in enhancer activity traditionally requires extensive experimental methods like ChIP assays or genetic perturbations.
• These experimental approaches are often time-consuming and resource-intensive, limiting large-scale functional genomic analyses.

Purpose of the Study:

• To introduce a novel computational framework for inferring the regulatory contribution of individual TF motif instances directly from DNA sequence.
• To validate the accuracy of this in silico approach by comparing its predictions with experimental data from TF knockout experiments.

Main Methods:

• Utilized AlphaGenome, a large sequence-to-function foundation model, to create a virtual motif-perturbation framework.
• In silico ablation of candidate C/EBPβ motifs within active chromatin regions to predict changes in regulatory activity.
• Statistical assessment of predicted changes against a null distribution and comparison with H3K27ac CUT&RUN data following CEBPB knockout in multiple myeloma cells.

Main Results:

• The virtual motif-perturbation framework successfully predicted the regulatory contribution of TF motifs from DNA sequence alone.
• In silico predictions showed strong concordance with experimental measurements of H3K27ac changes after CEBPB knockout.
• Discrepancies between in silico and in vitro results highlighted biologically interpretable mechanisms, further validating the model.

Conclusions:

• Large sequence-based foundation models can effectively approximate the functional consequences of TF perturbations at the motif level.
• Sequence-only virtual ablation offers a scalable and efficient strategy for hypothesis generation and regulatory annotation of TF binding sites.
• This approach can be extended to analyze diverse TFs, cell types, and chromatin states, advancing functional genomics.