FCNGRU

FCNGRU locates transcription factor binding sites within DNA sequences and estimates sequence-level binding intensity using a fully convolutional neural network combined with gated recurrent units for TF–DNA interaction prediction.

Key Features:

• Integration of Neural Networks: Combines a fully convolutional neural network (FCN) with a gated recurrent unit (GRU) to integrate convolutional feature extraction and recurrent sequence modeling for TFBS prediction.
• Two-task framework (FCNGRU-double): Implements a nucleotide-level classification task that predicts per-nucleotide probabilities of being part of a TFBS and a sequence-level regression task that estimates binding intensity for each DNA sequence.
• Performance on UniPROBE uPBMs: Demonstrated superior performance on 45 in vitro datasets from the UniPROBE database derived from universal protein binding microarrays (uPBMs).
• Advantage over single-task model: Provides a significant advantage over FCNGRU-single, which only locates TFBSs without estimating sequence intensity.
• In vivo analysis: Has been further analyzed using in vivo datasets to assess performance in biological contexts.

Scientific Applications:

• TF–DNA interaction prediction: Produces nucleotide- and sequence-level outputs to support computational inference of transcription factor–DNA interactions.
• Gene regulation studies: Aids elucidation of gene regulation mechanisms and identification of regulatory elements.
• Genetic research, disease studies, and therapeutics: Informs analyses relevant to genetic research, disease-related regulatory changes, and therapeutic development.
• Gene expression control analysis: Supports comprehensive analyses of gene expression control by providing precise localization and binding intensity estimates.

Methodology:

Combines a fully convolutional neural network with a gated recurrent unit in a two-task framework (nucleotide-level classification and sequence-level regression) and was evaluated on 45 in vitro UniPROBE uPBM datasets and further analyzed on in vivo datasets.