BindSpace

BindSpace embeds DNA sequences and transcription factor labels into a shared representational space to enable multiclass prediction of TF–DNA binding.

Key Features:

• Joint sequence–label embedding: Trains embeddings that place DNA sequences and TF labels in the same representational space using binding data from hundreds of transcription factors and over one million DNA sequences.
• Multiclass binding prediction and validation: Produces multiclass predictions of TF binding with high performance that has been validated using both in vitro and in vivo data.
• Resolution of closely related TFs: Differentiates binding signals from closely related transcription factors to resolve similar binding motifs.

Scientific Applications:

• Transcription factor binding prediction: Predicts TF–DNA interactions for many TFs simultaneously through multiclass classification in the embedding space.
• Regulatory element identification: Aids identification of candidate regulatory elements by mapping sequence binding preferences to TF labels.
• Gene regulation studies: Supports analysis of gene regulation and discovery of regulatory mechanisms by linking sequences to TFs.
• Investigation of transcriptional dysregulation: Facilitates studies of genetic diseases linked to transcriptional dysregulation by predicting altered TF–DNA interactions.

Methodology:

Trains machine-learning embeddings on comprehensive TF binding datasets (hundreds of TFs, >1 million DNA sequences) to embed DNA sequences and TF labels in a shared space for multiclass binding prediction.