Proteome-scale inference of transcription factor binding motifs from predicted protein-DNA structures
Transcription factors (TFs) control development and cell identity by binding sequence-specific DNA motifs, yet the binding specificities of most TFs remain unknown because experimental binding assays do not scale. In zebrafish, only one-third of TFs have assigned motifs, nearly all inferred by homology from human or mouse experimental data. Here we present a proteome-scale computational screen that infers DNA-binding specificity from predicted protein-DNA structures. We use AlphaFold3 to model zebrafish TFs with candidate motif sequences derived from accessible chromatin, then use NA-MPNN to infer the DNA motifs from the most confident predictions. Across 210 TFs with experimentally characterized orthologs, predicted motifs match the known motif for 44%. Propagated through sequence clusters, these predictions form ZebraMotif, a motif database for all 2,543 zebrafish TFs. ZebraMotif covers 2.8 times as many TFs as CisBP's zebrafish collection and links accessible regulatory elements to candidate upstream regulators genome-wide. Because the approach requires only a proteome and a set of candidate motifs, it is applicable to any species lacking experimental binding data.
Authors
- Nathan A. Hammond (ORCID: https://orcid.org/0000-0001-9635-2294)
- Joseph L. DeRisi (ORCID: https://orcid.org/0000-0002-4611-9205)
- Löıc A. Royer (ORCID: https://orcid.org/0000-0002-9991-9724)
- John E. Pak (ORCID: https://orcid.org/0000-0002-2998-9735)
- Yang Joon Kim (ORCID: https://orcid.org/0000-0003-1742-5657)
- Rosío Alvarez
- Benjamin Giovanni Iovino (ORCID: https://orcid.org/0009-0007-6512-4214)
- William Law
- Louis Boahene
- Nash Bulaong
- Mark Potts
Publication Details
- Journal
- bioRxiv (Cold Spring Harbor Laboratory)
- Published
- 2026-11-19
- DOI
- https://doi.org/10.64898/2026.11.18.767614
- Primary Topic
- Genomics and Chromatin Dynamics
- Type
- preprint