Local function conservation in sequence and structure space

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.

Intelligent Design of 14-3-3 Docking Proteins Utilizing Synthetic Evolution Artificial Intelligence (SYN-AI)

The ability to write DNA code from scratch will allow for the discovery of new and interesting chemistries as well as allowing the rewiring of cell signal pathways. Herein, we have utilized synthetic evolution artificial intelligence (SYN-AI) to intelligently design a set of 14-3-3 docking genes. SYN-AI engineers synthetic genes utilizing a parental gene as an evolution template. Wherein, evolution is fast-forwarded by transforming template gene sequences to DNA secondary and tertiary codes based upon gene hierarchical structural levels. The DNA secondary code allows identification of genomic building blocks across an orthologous sequence space comprising multiple genomes. Where, the DNA tertiary code allows engineering of supersecondary structures. SYN-AI constructed a library of 10 million genes that was reduced to three structurally functional 14-3-3 docking genes by applying natural selection protocols. Synthetic protein identity was verified utilizing Clustal Omega sequence alignments and Phylogeny.fr phylogenetic analysis. Wherein, we were able to confirm the three-dimensional structure utilizing I-TASSER and protein-ligand interactions utilizing COACH and Cofactor. The conservation of allosteric communications was confirmed utilizing elastic and anisotropic network models. Whereby, we utilized elNemo and ANM2.1 to confirm conservation of the 14-3-3 ζ amphipathic groove. Notably, to the best of our knowledge, we report the first 14-3-3 docking genes to be written from scratch.

The impact of structural diversity and parameterization on maps of the protein universe

BACKGROUND

Low dimensional maps of protein structure space (MPSS) provide a powerful global representation of all proteins. In such mappings structural relationships are depicted through spatial adjacency of points, each of which represents a molecule. MPSS can help in understanding the local and global topological characteristics of the structure space, as well as elucidate structure-function relationships within and between sets of proteins. A number of meta- and method-dependent parameters are involved in creating MPSS. However, at the state-of-the-art, a systematic investigation of the influence of these parameters on MPSS construction has yet to be carried out. Further, while specific cases in which MPSS out-perform pairwise distances for prediction of functional annotations have been noted, no general explanation for this phenomenon has yet been advanced.

METHODS

We address the above questions within the technical context of creating MPSS by utilizing multidimensional scaling (MDS) for obtaining low-dimensional projections of structure alignment distances.

RESULTS AND CONCLUSION

MDS is demonstrated as an effective method for construction of MPSS where related structures are co-located, even when their functional and evolutionary proximity cannot be deduced from distributions of pairwise comparisons alone. In particular, we show that MPSS exceed pairwise distance distributions in predictive capability for those annotations of shared function or origin which are characterized by a high level of structural diversity. We also determine the impact of the choice of structure alignment and MDS algorithms on the accuracy of such predictions.

Prediction of RNA binding proteins comes of age from low resolution to high resolution

Networks of protein-RNA interactions is likely to be larger than protein-protein and protein-DNA interaction networks because RNA transcripts are encoded tens of times more than proteins (e.g. only 3% of human genome coded for proteins), have diverse function and localization, and are controlled by proteins from birth (transcription) to death (degradation). This massive network is evidenced by several recent experimental discoveries of large numbers of previously unknown RNA-binding proteins (RBPs). Meanwhile, more than 400 non-redundant protein-RNA complex structures (at 25% sequence identity or less) have been deposited into the protein databank. These sequences and structural resources for RBPs provide ample data for the development of computational techniques dedicated to RBP prediction, as experimentally determining RNA-binding functions is time-consuming and expensive. This review compares traditional machine-learning based approaches with emerging template-based methods at several levels of prediction resolution ranging from two-state binding/non-binding prediction, to binding residue prediction and protein-RNA complex structure prediction. The analysis indicates that the two approaches are complementary and their combinations may lead to further improvements.

Protein Homology Analysis for Function Prediction with Parallel Sub-Graph Isomorphism

To annotate the biological function of a protein molecule, it is essential to have information on its 3D structure. Many successful methods for function prediction are based on determining structurally conserved regions because the functional residues are proved to be more conservative than others in protein evolution. Since the 3D conformation of a protein can be represented by a contact map graph, graph matching, algorithms are often employed to identify the conserved residues in weakly homologous protein pairs. However, the general graph matching algorithm is computationally expensive because graph similarity searching is essentially a NP-hard problem. Parallel implementations of the graph matching are often exploited to speed up the process. In this chapter,the authors review theoretical and computational approaches of graph theory and the recently developed graph matching algorithms for protein function prediction.

Dissecting protein loops with a statistical scalpel suggests a functional implication of some structural motifs

Background

One of the strategies for protein function annotation is to search particular structural motifs that are known to be shared by proteins with a given function.

Results

Here, we present a systematic extraction of structural motifs of seven residues from protein loops and we explore their correspondence with functional sites. Our approach is based on the structural alphabet HMM-SA (Hidden Markov Model - Structural Alphabet), which allows simplification of protein structures into uni-dimensional sequences, and advanced pattern statistics adapted to short sequences. Structural motifs of interest are selected by looking for structural motifs significantly over-represented in SCOP superfamilies in protein loops. We discovered two types of structural motifs significantly over-represented in SCOP superfamilies: (i) ubiquitous motifs, shared by several superfamilies and (ii) superfamily-specific motifs, over-represented in few superfamilies. A comparison of ubiquitous words with known small structural motifs shows that they contain well-described motifs as turn, niche or nest motifs. A comparison between superfamily-specific motifs and biological annotations of Swiss-Prot reveals that some of them actually correspond to functional sites involved in the binding sites of small ligands, such as ATP/GTP, NAD(P) and SAH/SAM.

Conclusions

Our findings show that statistical over-representation in SCOP superfamilies is linked to functional features. The detection of over-represented motifs within structures simplified by HMM-SA is therefore a promising approach for prediction of functional sites and annotation of uncharacterized proteins.

Discrete-continuous duality of protein structure space

Recently, the nature of protein structure space has been widely discussed in the literature. The traditional discrete view of protein universe as a set of separate folds has been criticized in the light of growing evidence that almost any arrangement of secondary structures is possible and the whole protein space can be traversed through a path of similar structures. Here we argue that the discrete and continuous descriptions are not mutually exclusive, but complementary: the space is largely discrete in evolutionary sense, but continuous geometrically when purely structural similarities are quantified. Evolutionary connections are mainly confined to separate structural prototypes corresponding to folds as islands of structural stability, with few remaining traceable links between the islands. However, for a geometric similarity measure, it is usually possible to find a reasonable cutoff that yields paths connecting any two structures through intermediates.