Co-evolutionary model of
TF-DNA interaction DCA algorithm Hamiltonian score References

Welcome to the Coevolved-TF-DNA web service! The Coevolved-TF-DNA web application offers comprehensive analyses of interactions between Transcription factors and DNA. The webserver contains 184 TFs with 147 human TFs and 37 mouse TFs. For each TFs, a global pairwise modeling DCA-Scapes was trained with the experimental HT-SELEX data to uncover the high-resolution TF recognition specificity patterns, predicted the binding of DNA sequences.

Co-evolutionary model of TF-DNA interaction

For each TF protein, the coevolutionary model was trained with the TF-bound DNA sequences in round 4 of HT-SELEX experiment, and the initial non-selected sequence pool in round 0 as the background set. With the DNA sequences that selected from HT-SELEX, this model first estimated a joint probability distribution of these sequences with two types of recognition parameters: pairwise couplings eij and local biases hi. For any given DNA sequence, these recognition parameters can be collectively interpreted as the Hamiltonian score (see below sections for details), which quantitatively predicts how likely this DNA sequence binds to the Transcription factor. A random sequence distribution null model that included one million random sequences with similar genomic nucleotide bias was generated to further quantify the probability of interaction. This global comprehensive model was first applied and validated the prediction of exploring the TF-DNA interaction with in vivo binding site of ChIP-seq experiments. Based on the accuracy in Chip-seq data prediction, TFs are separated into two groups: Validated TFs models (validated using ChIP-seq data with high accuracy: average AUC>0.7, N=18) and not-validated TFs models (TFs don’t have ChIP-seq data to compare, N=81 or get low AUC score, N=48).

DCA algorithm

For the training data (FR) and background data (BG), This model separately infer the joint probability distribution of 20mer DNA sequences with two types of parameters: pairwise couplings eij and local biases (fields) hi :

Furthermore, these recognition parameters can elucidate high-resolution maps of specificity to further understand the nucleotide contribution of couplings in TF recognitions.

Protein-binding recognition landscape

Hamiltonian score

For a given DNA sequence x, the Hamiltonian score could quantitatively predict how likely this DNA sequence be a target for the Transcription factors. The more negative Hamiltonian score suggests more favorable TF recognition.

For the given DNA sequence of interested or the entire genome sequence, the Hamiltonian scores were calculated for each 20mer sequences, with a sliding window size of 20, from the first nucleotide to the end of the sequence. For users input DNA sequences, all the Hamiltonian score of every 20mers will be reported with p-value calculated based on the null model. For genome sequences, only the most favorable potential binding sequences are reported.

References

Qin Zhou, Alex David Stanowick, Faruck Morcos. Characterizing DNA recognition preferences of transcription factors using global couplings and high throughput sequencing.
Zhou, Qin, et al. "Global pairwise RNA interaction landscapes reveal core features of protein recognition." Nature communications 9.1 (2018): 1-10.
Morcos, Faruck, et al. "Direct coupling analysis for protein contact prediction." Protein structure prediction. Humana Press, New York, NY, 2014. 55-70.
Morcos, Faruck, et al. "Direct-coupling analysis of residue coevolution captures native contacts across many protein families." Proceedings of the National Academy of Sciences 108.49 (2011): E1293-E1301.