Predicting DNA-binding sites of proteins from amino acid sequence

Background  

Understanding the molecular details of protein-DNA interactions is critical for deciphering the mechanisms of gene regulation. We present a machine learning approach for the identification of amino acid residues involved in protein-DNA interactions.     

Selection of DNA binding sites by regulatory proteins. II. The binding specificity of cyclic AMP receptor protein to recognition sites

The statistics of base-pair usage within known recognition sites for a particular DNA-binding protein can be used to estimate the relative protein binding affinities to these sites, as well as to sites containing any other combinations of base-pairs. As has been described elsewhere, the connection between base-pair statistics and binding free energy is made by an equal probability selection assumption; i.e. that all base-pair sequences that provide appropriate binding strength are equally likely to have been chosen as recognition sites in the course of evolution. This is analogous to a statistical-mechanical system where all configurations with the same energy are equally likely to occur. In this communication, we apply the statistical-mechanical selection theory to analyze the base-pair statistics of the known recognition sequences for the cyclic AMP receptor protein (CRP). The theoretical predictions are found to be in reasonable agreement with binding data for those sequences for which experimental binding information is available, thus lending support to the basic assumptions of the selection theory. On the basis of this agreement, we can predict the affinity for CRP binding to any base-pair sequence, albeit with a large statistical uncertainty. When the known recognition sites for CRP are ranked according to predicted binding affinities, we find that the ranking is consistent with the hypothesis that the level of function of these sites parallels their fractional saturation with CRP-cAMP under in-vivo conditions. When applied to the entire genome, the theory predicts the existence of a large number of randomly occurring "pseudosites" with strong binding affinity for CRP. It appears that most CRP molecules are engaged in non-productive binding at non-specific or pseudospecific sites under in-vivo conditions. In this sense, the specificity of the CRP binding site is very low. Relative specificity requirements for polymerases, repressors and activators are compared in light of the results of this and the first paper in this series.     

Altering the DNA-binding specificity of the yeast Matalpha 2 homeodomain protein.

Homeodomain proteins are a highly conserved class of DNA-binding proteins that are found in virtually every eukaryotic organism. The conserved mechanism that these proteins use to bind DNA suggests that there may be at least a partial DNA recognition code for this class of proteins. To test this idea, we have investigated the sequence-specific requirements for DNA binding and repression by the yeast alpha2 homeodomain protein in association with its cofactors, Mcm1 and Mata1. We have determined the contribution for each residue in the alpha2 homeodomain that contacts the DNA in the co-crystal structures of the protein. We have also engineered mutants in the alpha2 homeodomain to alter the DNA-binding specificity of the protein. Although we were unable to change the specificity of alpha2 by making substitutions at residues 47, 54, and 55, we were able to alter the DNA-binding specificity by making substitutions at residue 50 in the homeodomain. Since other homeodomain proteins show similar changes in specificity with substitutions at residue 50, this suggests that there is at least a partial DNA recognition code at this position.     

Conservation of binding site specificity of three yeast DNA binding proteins.

Sequence specific binding of protein extracts from 13 different yeast species to three oligonucleotide probes and two points mutants derived from Saccharomyces cerevisiae DNA binding proteins were tested using mobility shift assays. The probes were high affinity binding sites for GRF1/RAP1/ABF1 and CP1/CPF1. Most yeasts in the genus Saccharomyces showed specific binding to all three probes and also displayed similar sequence requirements when challenged by molar excesses of mutant probes. The affinities for the probes varied amongst the other yeasts tested, but in general, CPF1 binding activity was the most widespread, while the other two were more limited.     

A new method for the purification of DNA-binding proteins with sequence specificity

We describe a technique for a rapid and efficient isolation and purification of proteins binding to defined DNA sequences. Cloned double-stranded DNA was covalently coupled to m-aminobenzyloximethylcellulose in order to purify proteins which recognize and bind to specific sequences on the DNA. The purification of two DNA-binding proteins from Drosophila melanogaster is demonstrated using the respective cloned DNA sequences.     

Selection of DNA binding sites by regulatory proteins. Functional specificity and pseudosite competition

The frequency of base-pair occurrence in a set of recognition sequences for a particular DNA-binding protein is strongly related to the contributions to the binding free energy from the individual base pairs. Thus from the statistics of base-pair choice, it is possible to estimate the relative binding strengths of any base-pair sequences and to predict the effect of point mutations in specific sites. On the same basis, one can describe the binding properties of random DNA sequences and thereby the expected competitive effects from all the nonspecific DNA sites in the genome of a living cell. The statistical selection theory [Berg & von Hippel.J. Mol. Biol. 193 (1987) 723-750] describing these relations is extended and tested with computer simulations. The theory is shown to hold up well also in the case when base pairs contribute cooperatively to the binding interaction. The simulations also demonstrate the effects of the statistical small-sample uncertainty that appears due to the limited size of all sets of recognition sites identified.     

Selection of DNA binding sites by regulatory proteins: the LexA protein and the arginine repressor use different strategies for functional specificity.

The DNA sequences in the operator sites of the arginine regulon and of the SOS regulon have been subject to a statistical analysis. A quantitative correlation is found between the statistics of sequence choice and the activity at individual operator sites in both systems, as expected from theoretical considerations [Berg & von Hippel, J.Mol.Biol. (1987) 193, 723-750]. Based on these correlations it is possible to predict the effect of various sequence mutations. There is a significant difference in the slopes of the correlation lines between sequence and activity for the two systems. From this difference it can be expected that individual point mutations in the ARG boxes will have a much smaller effect on activity than similar changes in the SOS boxes. This difference may be related to a strong cooperative activity at tandem ARG boxes while the binding at SOS boxes appears to be mostly noncooperative.     

Prediction of RNA binding sites in proteins from amino acid sequence

RNA-protein interactions are vitally important in a wide range of biological processes, including regulation of gene expression, protein synthesis, and replication and assembly of many viruses. We have developed a computational tool for predicting which amino acids of an RNA binding protein participate in RNA-protein interactions, using only the protein sequence as input. RNABindR was developed using machine learning on a validated nonredundant data set of interfaces from known RNA-protein complexes in the Protein Data Bank. It generates a classifier that captures primary sequence signals sufficient for predicting which amino acids in a given protein are located in the RNA-protein interface. In leave-one-out cross-validation experiments, RNABindR identifies interface residues with >85% overall accuracy. It can be calibrated by the user to obtain either high specificity or high sensitivity for interface residues. RNABindR, implementing a Naive Bayes classifier, performs as well as a more complex neural network classifier (to our knowledge, the only previously published sequence-based method for RNA binding site prediction) and offers the advantages of speed, simplicity and interpretability of results. RNABindR predictions on the human telomerase protein hTERT are in good agreement with experimental data. The availability of computational tools for predicting which residues in an RNA binding protein are likely to contact RNA should facilitate design of experiments to directly test RNA binding function and contribute to our understanding of the diversity, mechanisms, and regulation of RNA-protein complexes in biological systems. (RNABindR is available as a Web tool from http://bindr.gdcb.iastate.edu.).