Unsupervised protein embeddings outperform hand-crafted sequence and structure features at predicting molecular function.

MOTIVATION: Protein function prediction is a difficult bioinformatics problem. Many recent methods use deep neural networks to learn complex sequence representations and predict function from these. Deep supervised models require a lot of labeled training data which are not available for this task. However, a very large amount of protein sequences without functional labels is available. RESULTS: We applied an existing deep sequence model that had been pretrained in an unsupervised setting on the supervised task of protein molecular function prediction. We found that this complex feature representation is effective for this task, outperforming hand-crafted features such as one-hot encoding of amino acids, k-mer counts, secondary structure and backbone angles. Also, it partly negates the need for complex prediction models, as a two-layer perceptron was enough to achieve competitive performance in the third Critical Assessment of Functional Annotation benchmark. We also show that combining this sequence representation with protein 3D structure information does not lead to performance improvement, hinting that 3D structure is also potentially learned during the unsupervised pretraining. AVAILABILITY AND IMPLEMENTATION: Implementations of all used models can be found at https://github.com/stamakro/GCN-for-Structure-and-Function. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Automatic Gene Function Prediction in the 2020's.

The current rate at which new DNA and protein sequences are being generated is too fast to experimentally discover the functions of those sequences, emphasizing the need for accurate Automatic Function Prediction (AFP) methods. AFP has been an active and growing research field for decades and has made considerable progress in that time. However, it is certainly not solved. In this paper, we describe challenges that the AFP field still has to overcome in the future to increase its applicability. The challenges we consider are how to: (1) include condition-specific functional annotation, (2) predict functions for non-model species, (3) include new informative data sources, (4) deal with the biases of Gene Ontology (GO) annotations, and (5) maximally exploit the GO to obtain performance gains. We also provide recommendations for addressing those challenges, by adapting (1) the way we represent proteins and genes, (2) the way we represent gene functions, and (3) the algorithms that perform the prediction from gene to function. Together, we show that AFP is still a vibrant research area that can benefit from continuing advances in machine learning with which AFP in the 2020s can again take a large step forward reinforcing the power of computational biology.

Metric learning on expression data for gene function prediction.

MOTIVATION: Co-expression of two genes across different conditions is indicative of their involvement in the same biological process. However, when using RNA-Seq datasets with many experimental conditions from diverse sources, only a subset of the experimental conditions is expected to be relevant for finding genes related to a particular Gene Ontology (GO) term. Therefore, we hypothesize that when the purpose is to find similarly functioning genes, the co-expression of genes should not be determined on all samples but only on those samples informative for the GO term of interest. RESULTS: To address this, we developed Metric Learning for Co-expression (MLC), a fast algorithm that assigns a GO-term-specific weight to each expression sample. The goal is to obtain a weighted co-expression measure that is more suitable than the unweighted Pearson correlation for applying Guilt-By-Association-based function predictions. More specifically, if two genes are annotated with a given GO term, MLC tries to maximize their weighted co-expression and, in addition, if one of them is not annotated with that term, the weighted co-expression is minimized. Our experiments on publicly available Arabidopsis thaliana RNA-Seq data demonstrate that MLC outperforms standard Pearson correlation in term-centric performance. Moreover, our method is particularly good at more specific terms, which are the most interesting. Finally, by observing the sample weights for a particular GO term, one can identify which experiments are important for learning that term and potentially identify novel conditions that are relevant, as demonstrated by experiments in both A. thaliana and Pseudomonas Aeruginosa. AVAILABILITY AND IMPLEMENTATION: MLC is available as a Python package at www.github.com/stamakro/MLC. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Improving protein function prediction using protein sequence and GO-term similarities.

MOTIVATION: Most automatic functional annotation methods assign Gene Ontology (GO) terms to proteins based on annotations of highly similar proteins. We advocate that proteins that are less similar are still informative. Also, despite their simplicity and structure, GO terms seem to be hard for computers to learn, in particular the Biological Process ontology, which has the most terms (>29 000). We propose to use Label-Space Dimensionality Reduction (LSDR) techniques to exploit the redundancy of GO terms and transform them into a more compact latent representation that is easier to predict. RESULTS: We compare proteins using a sequence similarity profile (SSP) to a set of annotated training proteins. We introduce two new LSDR methods, one based on the structure of the GO, and one based on semantic similarity of terms. We show that these LSDR methods, as well as three existing ones, improve the Critical Assessment of Functional Annotation performance of several function prediction algorithms. Cross-validation experiments on Arabidopsis thaliana proteins pinpoint the superiority of our GO-aware LSDR over generic LSDR. Our experiments on A.thaliana proteins show that the SSP representation in combination with a kNN classifier outperforms state-of-the-art and baseline methods in terms of cross-validated F-measure. AVAILABILITY AND IMPLEMENTATION: Source code for the experiments is available at https://github.com/stamakro/SSP-LSDR. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Homologues of potato chromosome 5 show variable collinearity in the euchromatin, but dramatic absence of sequence similarity in the pericentromeric heterochromatin.

BACKGROUND: In flowering plants it has been shown that de novo genome assemblies of different species and genera show a significant drop in the proportion of alignable sequence. Within a plant species, however, it is assumed that different haplotypes of the same chromosome align well. In this paper we have compared three de novo assemblies of potato chromosome 5 and report on the sequence variation and the proportion of sequence that can be aligned. RESULTS: For the diploid potato clone RH89-039-16 (RH) we produced two linkage phase controlled and haplotype-specific assemblies of chromosome 5 based on BAC-by-BAC sequencing, which were aligned to each other and compared to the 52 Mb chromosome 5 reference sequence of the doubled monoploid clone DM 1-3 516 R44 (DM). We identified 17.0 Mb of non-redundant sequence scaffolds derived from euchromatic regions of RH and 38.4 Mb from the pericentromeric heterochromatin. For 32.7 Mb of the RH sequences the correct position and order on chromosome 5 was determined, using genetic markers, fluorescence in situ hybridisation and alignment to the DM reference genome. This ordered fraction of the RH sequences is situated in the euchromatic arms and in the heterochromatin borders. In the euchromatic regions, the sequence collinearity between the three chromosomal homologs is good, but interruption of collinearity occurs at nine gene clusters. Towards and into the heterochromatin borders, absence of collinearity due to structural variation was more extensive and was caused by hemizygous and poorly aligning regions of up to 450 kb in length. In the most central heterochromatin, a total of 22.7 Mb sequence from both RH haplotypes remained unordered. These RH sequences have very few syntenic regions and represent a non-alignable region between the RH and DM heterochromatin haplotypes of chromosome 5. CONCLUSIONS: Our results show that among homologous potato chromosomes large regions are present with dramatic loss of sequence collinearity. This stresses the need for more de novo reference assemblies in order to capture genome diversity in this crop. The discovery of three highly diverged pericentric heterochromatin haplotypes within one species is a novelty in plant genome analysis. The possible origin and cytogenetic implication of this heterochromatin haplotype diversity are discussed.

A quantitative and dynamic model of the Arabidopsis flowering time gene regulatory network.

Various environmental signals integrate into a network of floral regulatory genes leading to the final decision on when to flower. Although a wealth of qualitative knowledge is available on how flowering time genes regulate each other, only a few studies incorporated this knowledge into predictive models. Such models are invaluable as they enable to investigate how various types of inputs are combined to give a quantitative readout. To investigate the effect of gene expression disturbances on flowering time, we developed a dynamic model for the regulation of flowering time in Arabidopsis thaliana. Model parameters were estimated based on expression time-courses for relevant genes, and a consistent set of flowering times for plants of various genetic backgrounds. Validation was performed by predicting changes in expression level in mutant backgrounds and comparing these predictions with independent expression data, and by comparison of predicted and experimental flowering times for several double mutants. Remarkably, the model predicts that a disturbance in a particular gene has not necessarily the largest impact on directly connected genes. For example, the model predicts that SUPPRESSOR OF OVEREXPRESSION OF CONSTANS (SOC1) mutation has a larger impact on APETALA1 (AP1), which is not directly regulated by SOC1, compared to its effect on LEAFY (LFY) which is under direct control of SOC1. This was confirmed by expression data. Another model prediction involves the importance of cooperativity in the regulation of APETALA1 (AP1) by LFY, a prediction supported by experimental evidence. Concluding, our model for flowering time gene regulation enables to address how different quantitative inputs are combined into one quantitative output, flowering time.

Exploring genetic variation in the tomato (Solanum section Lycopersicon) clade by whole-genome sequencing.

We explored genetic variation by sequencing a selection of 84 tomato accessions and related wild species representative of the Lycopersicon, Arcanum, Eriopersicon and Neolycopersicon groups, which has yielded a huge amount of precious data on sequence diversity in the tomato clade. Three new reference genomes were reconstructed to support our comparative genome analyses. Comparative sequence alignment revealed group-, species- and accession-specific polymorphisms, explaining characteristic fruit traits and growth habits in the various cultivars. Using gene models from the annotated Heinz 1706 reference genome, we observed differences in the ratio between non-synonymous and synonymous SNPs (dN/dS) in fruit diversification and plant growth genes compared to a random set of genes, indicating positive selection and differences in selection pressure between crop accessions and wild species. In wild species, the number of single-nucleotide polymorphisms (SNPs) exceeds 10 million, i.e. 20-fold higher than found in most of the crop accessions, indicating dramatic genetic erosion of crop and heirloom tomatoes. In addition, the highest levels of heterozygosity were found for allogamous self-incompatible wild species, while facultative and autogamous self-compatible species display a lower heterozygosity level. Using whole-genome SNP information for maximum-likelihood analysis, we achieved complete tree resolution, whereas maximum-likelihood trees based on SNPs from ten fruit and growth genes show incomplete resolution for the crop accessions, partly due to the effect of heterozygous SNPs. Finally, results suggest that phylogenetic relationships are correlated with habitat, indicating the occurrence of geographical races within these groups, which is of practical importance for Solanum genome evolution studies.

The genome of the stress-tolerant wild tomato species Solanum pennellii.

Solanum pennellii is a wild tomato species endemic to Andean regions in South America, where it has evolved to thrive in arid habitats. Because of its extreme stress tolerance and unusual morphology, it is an important donor of germplasm for the cultivated tomato Solanum lycopersicum. Introgression lines (ILs) in which large genomic regions of S. lycopersicum are replaced with the corresponding segments from S. pennellii can show remarkably superior agronomic performance. Here we describe a high-quality genome assembly of the parents of the IL population. By anchoring the S. pennellii genome to the genetic map, we define candidate genes for stress tolerance and provide evidence that transposable elements had a role in the evolution of these traits. Our work paves a path toward further tomato improvement and for deciphering the mechanisms underlying the myriad other agronomic traits that can be improved with S. pennellii germplasm.

The genomes of the fungal plant pathogens Cladosporium fulvum and Dothistroma septosporum reveal adaptation to different hosts and lifestyles but also signatures of common ancestry.

We sequenced and compared the genomes of the Dothideomycete fungal plant pathogens Cladosporium fulvum (Cfu) (syn. Passalora fulva) and Dothistroma septosporum (Dse) that are closely related phylogenetically, but have different lifestyles and hosts. Although both fungi grow extracellularly in close contact with host mesophyll cells, Cfu is a biotroph infecting tomato, while Dse is a hemibiotroph infecting pine. The genomes of these fungi have a similar set of genes (70% of gene content in both genomes are homologs), but differ significantly in size (Cfu >61.1-Mb; Dse 31.2-Mb), which is mainly due to the difference in repeat content (47.2% in Cfu versus 3.2% in Dse). Recent adaptation to different lifestyles and hosts is suggested by diverged sets of genes. Cfu contains an α-tomatinase gene that we predict might be required for detoxification of tomatine, while this gene is absent in Dse. Many genes encoding secreted proteins are unique to each species and the repeat-rich areas in Cfu are enriched for these species-specific genes. In contrast, conserved genes suggest common host ancestry. Homologs of Cfu effector genes, including Ecp2 and Avr4, are present in Dse and induce a Cf-Ecp2- and Cf-4-mediated hypersensitive response, respectively. Strikingly, genes involved in production of the toxin dothistromin, a likely virulence factor for Dse, are conserved in Cfu, but their expression differs markedly with essentially no expression by Cfu in planta. Likewise, Cfu has a carbohydrate-degrading enzyme catalog that is more similar to that of necrotrophs or hemibiotrophs and a larger pectinolytic gene arsenal than Dse, but many of these genes are not expressed in planta or are pseudogenized. Overall, comparison of their genomes suggests that these closely related plant pathogens had a common ancestral host but since adapted to different hosts and lifestyles by a combination of differentiated gene content, pseudogenization, and gene regulation.

Predicting the impact of alternative splicing on plant MADS domain protein function.

Several genome-wide studies demonstrated that alternative splicing (AS) significantly increases the transcriptome complexity in plants. However, the impact of AS on the functional diversity of proteins is difficult to assess using genome-wide approaches. The availability of detailed sequence annotations for specific genes and gene families allows for a more detailed assessment of the potential effect of AS on their function. One example is the plant MADS-domain transcription factor family, members of which interact to form protein complexes that function in transcription regulation. Here, we perform an in silico analysis of the potential impact of AS on the protein-protein interaction capabilities of MIKC-type MADS-domain proteins. We first confirmed the expression of transcript isoforms resulting from predicted AS events. Expressed transcript isoforms were considered functional if they were likely to be translated and if their corresponding AS events either had an effect on predicted dimerisation motifs or occurred in regions known to be involved in multimeric complex formation, or otherwise, if their effect was conserved in different species. Nine out of twelve MIKC MADS-box genes predicted to produce multiple protein isoforms harbored putative functional AS events according to those criteria. AS events with conserved effects were only found at the borders of or within the K-box domain. We illustrate how AS can contribute to the evolution of interaction networks through an example of selective inclusion of a recently evolved interaction motif in the MADS AFFECTING FLOWERING1-3 (MAF1-3) subclade. Furthermore, we demonstrate the potential effect of an AS event in SHORT VEGETATIVE PHASE (SVP), resulting in the deletion of a short sequence stretch including a predicted interaction motif, by overexpression of the fully spliced and the alternatively spliced SVP transcripts. For most of the AS events we were able to formulate hypotheses about the potential impact on the interaction capabilities of the encoded MIKC proteins.

Mutational robustness of gene regulatory networks.

Mutational robustness of gene regulatory networks refers to their ability to generate constant biological output upon mutations that change network structure. Such networks contain regulatory interactions (transcription factor-target gene interactions) but often also protein-protein interactions between transcription factors. Using computational modeling, we study factors that influence robustness and we infer several network properties governing it. These include the type of mutation, i.e. whether a regulatory interaction or a protein-protein interaction is mutated, and in the case of mutation of a regulatory interaction, the sign of the interaction (activating vs. repressive). In addition, we analyze the effect of combinations of mutations and we compare networks containing monomeric with those containing dimeric transcription factors. Our results are consistent with available data on biological networks, for example based on evolutionary conservation of network features. As a novel and remarkable property, we predict that networks are more robust against mutations in monomer than in dimer transcription factors, a prediction for which analysis of conservation of DNA binding residues in monomeric vs. dimeric transcription factors provides indirect evidence.

Correlated mutations via regularized multinomial regression.

BACKGROUND: In addition to sequence conservation, protein multiple sequence alignments contain evolutionary signal in the form of correlated variation among amino acid positions. This signal indicates positions in the sequence that influence each other, and can be applied for the prediction of intra- or intermolecular contacts. Although various approaches exist for the detection of such correlated mutations, in general these methods utilize only pairwise correlations. Hence, they tend to conflate direct and indirect dependencies. RESULTS: We propose RMRCM, a method for Regularized Multinomial Regression in order to obtain Correlated Mutations from protein multiple sequence alignments. Importantly, our method is not restricted to pairwise (column-column) comparisons only, but takes into account the network nature of relationships between protein residues in order to predict residue-residue contacts. The use of regularization ensures that the number of predicted links between columns in the multiple sequence alignment remains limited, preventing overprediction. Using simulated datasets we analyzed the performance of our approach in predicting residue-residue contacts, and studied how it is influenced by various types of noise. For various biological datasets, validation with protein structure data indicates a good performance of the proposed algorithm for the prediction of residue-residue contacts, in comparison to previous results. RMRCM can also be applied to predict interactions (in addition to only predicting interaction sites or contact sites), as demonstrated by predicting PDZ-peptide interactions. CONCLUSIONS: A novel method is presented, which uses regularized multinomial regression in order to obtain correlated mutations from protein multiple sequence alignments. AVAILABILITY: R-code of our implementation is available via http://www.ab.wur.nl/rmrcm.

Finished genome of the fungal wheat pathogen Mycosphaerella graminicola reveals dispensome structure, chromosome plasticity, and stealth pathogenesis.

The plant-pathogenic fungus Mycosphaerella graminicola (asexual stage: Septoria tritici) causes septoria tritici blotch, a disease that greatly reduces the yield and quality of wheat. This disease is economically important in most wheat-growing areas worldwide and threatens global food production. Control of the disease has been hampered by a limited understanding of the genetic and biochemical bases of pathogenicity, including mechanisms of infection and of resistance in the host. Unlike most other plant pathogens, M. graminicola has a long latent period during which it evades host defenses. Although this type of stealth pathogenicity occurs commonly in Mycosphaerella and other Dothideomycetes, the largest class of plant-pathogenic fungi, its genetic basis is not known. To address this problem, the genome of M. graminicola was sequenced completely. The finished genome contains 21 chromosomes, eight of which could be lost with no visible effect on the fungus and thus are dispensable. This eight-chromosome dispensome is dynamic in field and progeny isolates, is different from the core genome in gene and repeat content, and appears to have originated by ancient horizontal transfer from an unknown donor. Synteny plots of the M. graminicola chromosomes versus those of the only other sequenced Dothideomycete, Stagonospora nodorum, revealed conservation of gene content but not order or orientation, suggesting a high rate of intra-chromosomal rearrangement in one or both species. This observed "mesosynteny" is very different from synteny seen between other organisms. A surprising feature of the M. graminicola genome compared to other sequenced plant pathogens was that it contained very few genes for enzymes that break down plant cell walls, which was more similar to endophytes than to pathogens. The stealth pathogenesis of M. graminicola probably involves degradation of proteins rather than carbohydrates to evade host defenses during the biotrophic stage of infection and may have evolved from endophytic ancestors.

PRI-CAT: a web-tool for the analysis, storage and visualization of plant ChIP-seq experiments.

Although several tools for the analysis of ChIP-seq data have been published recently, there is a growing demand, in particular in the plant research community, for computational resources with which such data can be processed, analyzed, stored, visualized and integrated within a single, user-friendly environment. To accommodate this demand, we have developed PRI-CAT (Plant Research International ChIP-seq analysis tool), a web-based workflow tool for the management and analysis of ChIP-seq experiments. PRI-CAT is currently focused on Arabidopsis, but will be extended with other plant species in the near future. Users can directly submit their sequencing data to PRI-CAT for automated analysis. A QuickLoad server compatible with genome browsers is implemented for the storage and visualization of DNA-binding maps. Submitted datasets and results can be made publicly available through PRI-CAT, a feature that will enable community-based integrative analysis and visualization of ChIP-seq experiments. Secondary analysis of data can be performed with the aid of GALAXY, an external framework for tool and data integration. PRI-CAT is freely available at http://www.ab.wur.nl/pricat. No login is required.

Assessing the contribution of alternative splicing to proteome diversity in Arabidopsis thaliana using proteomics data.

BACKGROUND: Large-scale analyses of genomics and transcriptomics data have revealed that alternative splicing (AS) substantially increases the complexity of the transcriptome in higher eukaryotes. However, the extent to which this complexity is reflected at the level of the proteome remains unclear. On the basis of a lack of conservation of AS between species, we previously concluded that AS does not frequently serve as a mechanism that enables the production of multiple functional proteins from a single gene. Following this conclusion, we hypothesized that the extent to which AS events contribute to the proteome diversity in Arabidopsis thaliana would be lower than expected on the basis of transcriptomics data. Here, we test this hypothesis by analyzing two large-scale proteomics datasets from Arabidopsis thaliana. RESULTS: A total of only 60 AS events could be confirmed using the proteomics data. However, for about 60% of the loci that, based on transcriptomics data, were predicted to produce multiple protein isoforms through AS, no isoform-specific peptides were found. We therefore performed in silico AS detection experiments to assess how well AS events were represented in the experimental datasets. The results of these in silico experiments indicated that the low number of confirmed AS events was the consequence of a limited sampling depth rather than in vivo under-representation of AS events in these datasets. CONCLUSION: Although the impact of AS on the functional properties of the proteome remains to be uncovered, the results of this study indicate that AS-induced diversity at the transcriptome level is also expressed at the proteome level.

SLIDER: a generic metaheuristic for the discovery of correlated motifs in protein-protein interaction networks.

Correlated motif mining (cmm) is the problem of finding overrepresented pairs of patterns, called motifs, in sequences of interacting proteins. Algorithmic solutions for cmm thereby provide a computational method for predicting binding sites for protein interaction. In this paper, we adopt a motif-driven approach where the support of candidate motif pairs is evaluated in the network. We experimentally establish the superiority of the Chi-square-based support measure over other support measures. Furthermore, we obtain that cmm is an np-hard problem for a large class of support measures (including Chi-square) and reformulate the search for correlated motifs as a combinatorial optimization problem. We then present the generic metaheuristic slider which uses steepest ascent with a neighborhood function based on sliding motifs and employs the Chi-square-based support measure. We show that slider outperforms existing motif-driven cmm methods and scales to large protein-protein interaction networks. The slider-implementation and the data used in the experiments are available on http://bioinformatics.uhasselt.be.

Genome-wide computational function prediction of Arabidopsis proteins by integration of multiple data sources.

Although Arabidopsis (Arabidopsis thaliana) is the best studied plant species, the biological role of one-third of its proteins is still unknown. We developed a probabilistic protein function prediction method that integrates information from sequences, protein-protein interactions, and gene expression. The method was applied to proteins from Arabidopsis. Evaluation of prediction performance showed that our method has improved performance compared with single source-based prediction approaches and two existing integration approaches. An innovative feature of our method is that it enables transfer of functional information between proteins that are not directly associated with each other. We provide novel function predictions for 5,807 proteins. Recent experimental studies confirmed several of the predictions. We highlight these in detail for proteins predicted to be involved in flowering and floral organ development.

Sequence motifs in MADS transcription factors responsible for specificity and diversification of protein-protein interaction.

Protein sequences encompass tertiary structures and contain information about specific molecular interactions, which in turn determine biological functions of proteins. Knowledge about how protein sequences define interaction specificity is largely missing, in particular for paralogous protein families with high sequence similarity, such as the plant MADS domain transcription factor family. In comparison to the situation in mammalian species, this important family of transcription regulators has expanded enormously in plant species and contains over 100 members in the model plant species Arabidopsis thaliana. Here, we provide insight into the mechanisms that determine protein-protein interaction specificity for the Arabidopsis MADS domain transcription factor family, using an integrated computational and experimental approach. Plant MADS proteins have highly similar amino acid sequences, but their dimerization patterns vary substantially. Our computational analysis uncovered small sequence regions that explain observed differences in dimerization patterns with reasonable accuracy. Furthermore, we show the usefulness of the method for prediction of MADS domain transcription factor interaction networks in other plant species. Introduction of mutations in the predicted interaction motifs demonstrated that single amino acid mutations can have a large effect and lead to loss or gain of specific interactions. In addition, various performed bioinformatics analyses shed light on the way evolution has shaped MADS domain transcription factor interaction specificity. Identified protein-protein interaction motifs appeared to be strongly conserved among orthologs, indicating their evolutionary importance. We also provide evidence that mutations in these motifs can be a source for sub- or neo-functionalization. The analyses presented here take us a step forward in understanding protein-protein interactions and the interplay between protein sequences and network evolution.

Conserved and variable correlated mutations in the plant MADS protein network.

BACKGROUND: Plant MADS domain proteins are involved in a variety of developmental processes for which their ability to form various interactions is a key requisite. However, not much is known about the structure of these proteins or their complexes, whereas such knowledge would be valuable for a better understanding of their function. Here, we analyze those proteins and the complexes they form using a correlated mutation approach in combination with available structural, bioinformatics and experimental data. RESULTS: Correlated mutations are affected by several types of noise, which is difficult to disentangle from the real signal. In our analysis of the MADS domain proteins, we apply for the first time a correlated mutation analysis to a family of interacting proteins. This provides a unique way to investigate the amount of signal that is present in correlated mutations because it allows direct comparison of mutations in various family members and assessing their conservation. We show that correlated mutations in general are conserved within the various family members, and if not, the variability at the respective positions is less in the proteins in which the correlated mutation does not occur. Also, intermolecular correlated mutation signals for interacting pairs of proteins display clear overlap with other bioinformatics data, which is not the case for non-interacting protein pairs, an observation which validates the intermolecular correlated mutations. Having validated the correlated mutation results, we apply them to infer the structural organization of the MADS domain proteins. CONCLUSION: Our analysis enables understanding of the structural organization of the MADS domain proteins, including support for predicted helices based on correlated mutation patterns, and evidence for a specific interaction site in those proteins.