Structure-based protein design with deep learning

Since the first revelation of proteins functioning as macromolecular machines through their three dimensional structures, researchers have been intrigued by the marvelous ways the biochemical processes are carried out by proteins. The aspiration to understand protein structures has fueled extensive efforts across different scientific disciplines. In recent years, it has been demonstrated that proteins with new functionality or shapes can be designed via structure-based modeling methods, and the design strategies have combined all available information — but largely piece-by-piece — from sequence derived statistics to the detailed atomic-level modeling of chemical interactions. Despite the significant progress, incorporating data-derived approaches through the use of deep learning methods can be a game changer. In this review, we summarize current progress, compare the arc of developing the deep learning approaches with the conventional methods, and describe the motivation and concepts behind current strategies that may lead to potential future opportunities.     

Multiomic Metabolic Enrichment Network Analysis Reveals Metabolite–Protein Physical Interaction Subnetworks Altered in Cancer

Metabolism is recognized as an important driver of cancer progression and other complex diseases, but global metabolite profiling remains a challenge. Protein expression profiling is often a poor proxy since existing pathway enrichment models provide an incomplete mapping between the proteome and metabolism. To overcome these gaps, we introduce multiomic metabolic enrichment network analysis (MOMENTA), an integrative multiomic data analysis framework for more accurately deducing metabolic pathway changes from proteomics data alone in a gene set analysis context by leveraging protein interaction networks to extend annotated metabolic models. We apply MOMENTA to proteomic data from diverse cancer cell lines and human tumors to demonstrate its utility at revealing variation in metabolic pathway activity across cancer types, which we verify using independent metabolomics measurements. The novel metabolic networks we uncover in breast cancer and other tumors are linked to clinical outcomes, underscoring the pathophysiological relevance of the findings.     

The role of epiphytism in architecture and evolutionary constraint within mycorrhizal networks of tropical orchids

Characterizing the architecture of bipartite networks is increasingly used as a framework to study biotic interactions within their ecological context and to assess the extent to which evolutionary constraint shape them. Orchid mycorrhizal symbioses are particularly interesting as they are viewed as more beneficial for plants than for fungi, a situation expected to result in an asymmetry of biological constraint. This study addressed the architecture and phylogenetic constraint in these associations in tropical context. We identified a bipartite network including 73 orchid species and 95 taxonomic units of mycorrhizal fungi across the natural habitats of Reunion Island. Unlike some recent evidence for nestedness in mycorrhizal symbioses, we found a highly modular architecture that largely reflected an ecological barrier between epiphytic and terrestrial subnetworks. By testing for phylogenetic signal, the overall signal was stronger for both partners in the epiphytic subnetwork. Moreover, in the subnetwork of epiphytic angraecoid orchids, the signal in orchid phylogeny was stronger than the signal in fungal phylogeny. Epiphytic associations are therefore more conservative and may co‐evolve more than terrestrial ones. We suggest that such tighter phylogenetic specialization may have been driven by stressful life conditions in the epiphytic niches. In addition to paralleling recent insights into mycorrhizal networks, this study furthermore provides support for epiphytism as a major factor affecting ecological assemblage and evolutionary constraint in tropical mycorrhizal symbioses.     

Maximum Likelihood Neural Network Prediction Models

Neural networks have received much attention in recent years mostly by non-statisticians. The purpose of this paper is to incorporate neural networks in a non-linear regression model and obtain maximum likelihood estimates of the network parameters using a standard Newton-Raphson algorithm. We use maximum likelihood estimators instead of the usual back-propagation technique and compare the neural network predictions with predictions of quadratic regression models and with non-parametric nearest neighbor predictions. These comparisons are made using data generated from a variety of functions. Because of the number of parameters involved, neural network models can easily over-fit the data, hence validation of results is crucial.     

Protected areas in the world’s ecoregions: How well connected are they?

Protected areas (PAs) are the main instrument for biodiversity conservation, which has triggered the development of numerous indicators and assessments on their coverage, performance and efficiency. The connectivity of the PA networks at a global scale has however been much less explored; previous studies have either focused on particular regions of the world or have only considered some types of PAs.Here we present, and globally assess, ProtConn, an indicator of PA connectivity that (i) quantifies the percentage of a study region covered by protected connected lands, (ii) can be partitioned in several components depicting different categories of land (unprotected, protected or transboundary) through which movement between protected locations may occur, (iii) is easy to communicate, to compare with PA coverage and to use in the assessment of global targets for PA systems.We apply ProtConn to evaluate the connectivity of the PA networks in all terrestrial ecoregions of the world as of June 2016, considering a range of median dispersal distances (1–100 km) encompassing the dispersal abilities of the large majority of terrestrial vertebrates.We found that 9.3% of the world is covered by protected connected lands (average for all the world’s ecoregions) for a reference dispersal distance of 10 km, increasing up to 11.7% for the largest dispersal distance considered of 100 km. These percentages are considerably smaller than the global PA coverage of 14.7%, indicating that the spatial arrangement of PAs is only partially successful in ensuring connectivity of protected lands. The connectivity of PAs largely differed across ecoregions. Only about a third of the world’s ecoregions currently meet the Aichi Target of having 17% of the terrestrial realm covered by well-connected systems of PAs. Finally, our findings suggest that PAs with less strict management objectives (allowing the sustainable use of resources) may play a fundamental role in upholding the connectivity of the PA systems.Our analyses and indicator make it possible to identify where on the globe additional efforts are most needed in expanding or reinforcing the connectivity of PA systems, and can be also used to assess whether newly designated sites provide effective connectivity gains in the PA system by acting as corridors or stepping stones between other PAs. The results of the ProtConn indicator are available, together with a suite of other global PA indicators, in the Digital Observatory for Protected Areas of the Joint Research Centre of the European Commission.