Table 1.

Computational models of protein-DNA binding specificity and high-throughput assays for generating the data used to train and test specificity models.

(A) Computational models of protein-DNA binding specificity
Model type	Model Description	Refs.
Position weight matrices (PWMs)	Simple probabilistic models that assume independence between positions in TF binding sites (TFBSs)	[5]
Dinucleotide weight matrices (DWMs)	Generalization of PWM models that incorporates frequencies of dinucleotides	[73, 230]
Bayesian networks	Flexible probabilistic models that can incorporate dependencies between positions in TFBSs	[63]
Hidden Markov models	Probabilistic models that can incorporate dependencies between neighboring positions in TFBSs	[70, 231]
High-order Markov models	Flexible probabilistic models that can incorporate high order dependencies between neighboring positions in TFBSs	[232]
k-mer based regression models	Probabilistic models that predict the level of TF binding based on the frequencies of mono-, di-, and tri-nucleotides	[93, 233]
Markov networks	Flexible probabilistic models that can incorporate high-order dependencies within TFBSs	[72]
Neural networks	Flexible probabilistic models that represent TF binding specificities using a system of interconnected, artificial “neurons”	[75]
Random forest models	Flexible probabilistic models that represent TF specificity using a collection of decision trees	[92]
Support vector models	Probabilistic models that can incorporate complex patterns of similarities between TFBSs	[2, 31]
Variable-order Bayesian networks	Flexible probabilistic models that can incorporate high-order dependencies within TFBSs	[234]
Thermodynamic/Energy-based models	Models that infer DNA binding affinities by fitting thermodynamic equations to experimental data	[73, 81–83, 235–237]
Atomistic/Structure-based models	Models based on known structures of TFs bound to target DNA sites	[86, 90]
Probabilistic models that incorporate structural features	Models that incorporate structural features such as groove geometries and helical parameters	[2, 79, 91, 92]
Probabilistic models that incorporate in vivo data	Models that incorporate in vivo data such as DNA accessibility, histone modifications	[238, 239]

(B) In vivo high-throughput DNA binding assays
Assay name	Assay description	Refs
ChIP-chip	Chromatin immunoprecipitation followed by microarray hybridization	[240]
ChIP-seq	Chromatin immunoprecipitation followed by high-throughput sequencing	[241]
ChIP-exo	Chromatin immunoprecipitation with exonuclease digestion followed by high-throughput sequencing	[242]
DamID	DNA adenine methyltransferase identification	[243]
DNase-seq	DNase I cleavage followed by high-throughput sequencing	[151, 244]
FAIRE-seq	Formaldehyde-assisted isolation of regulatory elements, followed by high-throughput sequencing	[149]
ATAC-seq	Assay for transposase-accessible chromatin using high-throughput	[152]

(C) In vitro high-throughput DNA binding assays
Assay name	Assay description	Refs
B1H	Bacterial one-hybrid	[102, 245]
PBM	Protein binding microarray	[94, 246]
CSI	Cognate site identifier	[247]
MITOMI	Mechanically induced trapping of molecular interactions	[101, 248]
MEGAshift	Microarray evaluation of genomic aptamers by shift	[249]
TIRF-PBM	Total internal reflectance fluorescence protein-binding microarray	[103]
Bind-n-Seq	Analysis of in vitro protein-DNA interactions using massively parallel sequencing	[250]
SELEX-seq/HT-SELEX	Systematic evolution of ligands by exponential enrichment, followed by high-throughput sequencing	[1, 82, 110]
EMSA-seq	Electrophoretic mobility shift assay followed by deep sequencing	[95]
HiTS-FLIP	High-throughput sequencing - fluorescent ligand interaction profiling	[108]
gcPBM	Genomic-context protein binding microarray	[2]