Transcription factor (TF) binding to DNA is critical to transcription regulation. Although the binding properties of numerous individual TFs are well-documented, a more detailed comprehension of how TFs interact cooperatively with DNA is required. We present COBIND, a novel method based on non-negative matrix factorization (NMF) to identify TF co-binding patterns automatically. COBIND applies NMF to one-hot encoded regions flanking known TF binding sites (TFBSs) to pinpoint enriched DNA patterns at fixed distances. We applied COBIND to 5699 TFBS datasets from UniBind for 401 TFs in seven species. The method uncovered already established co-binding patterns and new co-binding configurations not yet reported in the literature and inferred through motif similarity and protein-protein interaction knowledge. Our extensive analyses across species revealed that 67% of the TFs shared a co-binding motif with other TFs from the same structural family. The co-binding patterns captured by COBIND are likely functionally relevant as they harbor higher evolutionarily conservation than isolated TFBSs. Open chromatin data from matching human cell lines further supported the co-binding predictions. Finally, we used single-molecule footprinting data from mouse embryonic stem cells to confirm that the COBIND-predicted co-binding events associated with some TFs likely occurred on the same DNA molecules.

中文翻译：

---

使用非阴性基质分解鉴定转录因子共结合模式


转录因子 （TF） 与 DNA 的结合对转录调控至关重要。尽管许多单个 TF 的结合特性有据可查，但需要更详细地了解 TF 如何与 DNA 协同相互作用。我们提出了 COBIND，这是一种基于非负基质分解 （NMF） 的新方法，可自动识别 TF 共结合模式。COBIND 将 NMF 应用于已知 TF 结合位点 （TFBS） 侧翼的 onehot 编码区域，以在固定距离处精确定位富集的 DNA 模式。我们将 COBIND 应用于 UniBind 的 5699 个 TFBS 数据集，用于 7 个物种的 401 个 TFs。该方法揭示了文献中尚未报道的已经建立的共结合模式和新的共结合构型，并通过基序相似性和蛋白质-蛋白质相互作用知识进行推断。我们对物种的广泛分析表明，67% 的 TFs 与来自同一结构家族的其他 TFs 共享一个共结合基序。COBIND 捕获的共结合模式可能具有功能相关性，因为它们比分离的 TFBS 具有更高的进化保守性。来自匹配人类细胞系的开放染色质数据进一步支持了共结合预测。最后，我们使用来自小鼠胚胎干细胞的单分子足迹数据来确认与某些 TFs 相关的 COBIND 预测的共结合事件可能发生在相同的 DNA 分子上。