Search

Accurate preharvest yield estimation is an important issue for agricultural planning purposes and precision farming. Machine learning (ML) based on readily obtained information on the cropping system, typically including spectral reflectance measurements, is an essential approach for achieving practical solutions. We tested in a 9-year soil compaction experiment the accuracy of ML-based yield predictions made up to 2 months before harvest from a Ratio Vegetation Index (RVI) and recordings of precipitation and reference evapotranspiration. The applied data set comprises 224 combinations of plots and years with measured grain yields in the range of 4.22–9.34 Mg/ha. The best ML model [i.e., with the smallest mean absolute error (MAE)] was selected automatically by the AutoML interface included in the R program package H2O. Its cross-validated predictions made on June 30 more than 1 month before harvest showed an MAE of 0.38 Mg/ha when trained on all data from all years except the one under consideration. MAE increased to about 0.68 Mg/ha when determined 3 weeks earlier on June 10. MAE values in the range of 0.32–0.42 Mg/ha were obtained for predictions made on June 30 when based on data from at least six consecutive years; however, MAE showed no generally decreasing trend with the number of years. Yield estimations were robust towards a considerable soil variation observed within the experimental area due in part to the experimental treatments. The results show a potential of making yield predictions in barley 1–2 months before harvest, which, however, is not sufficiently early to support decisions on top-dress N fertilization.

Through advanced mechanistic modeling and the generation of large high-quality datasets, machine learning is becoming an integral part of understanding and engineering living systems. Here we show that mechanistic and machine learning models can be combined to enable accurate genotype-to-phenotype predictions. We use a genome-scale model to pinpoint engineering targets, efficient library construction of metabolic pathway designs, and high-throughput biosensor-enabled screening for training diverse machine learning algorithms. From a single data-generation cycle, this enables successful forward engineering of complex aromatic amino acid metabolism in yeast, with the best machine learning-guided design recommendations improving tryptophan titer and productivity by up to 74 and 43%, respectively, compared to the best designs used for algorithm training. Thus, this study highlights the power of combining mechanistic and machine learning models to effectively direct metabolic engineering efforts., In metabolic engineering, mechanistic models require prior metabolism knowledge of the chassis strain, whereas machine learning models need ample training data. Here, the authors combine the mechanistic and machine learning models to improve prediction performance of tryptophan metabolism in baker’s yeast.

Protein quality control (PQC) degrons are short protein segments that target misfolded proteins for proteasomal degradation, and thus protect cells against the accumulation of potentially toxic non-native proteins. Studies have shown that PQC degrons are hydrophobic and rarely contain negatively charged residues, features which are shared with chaperone-binding regions. Here we explore the notion that chaperone-binding regions may function as PQC degrons. When directly tested, we found that a canonical Hsp70-binding motif (the APPY peptide) functioned as a dose-dependent PQC degron both in yeast and in human cells. In yeast, Hsp70, Hsp110, Fes1, and the E3 Ubr1 target the APPY degron. Screening revealed that the sequence space within the chaperone-binding region of APPY that is compatible with degron function is vast. We find that the number of exposed Hsp70-binding sites in the yeast proteome correlates with a reduced protein abundance and half-life. Our results suggest that when protein folding fails, chaperone-binding sites may operate as PQC degrons, and that the sequence properties leading to PQC-linked degradation therefore overlap with those of chaperone binding.

Protein quality control (PQC) degrons are short protein segments that target misfolded proteins for proteasomal degradation, and thus protect cells against the accumulation of potentially toxic non-native proteins. Studies have shown that PQC degrons are hydrophobic and rarely contain negatively charged residues, features which are shared with chaperone-binding regions. Here we explore the notion that chaperone-binding regions may function as PQC degrons. When directly tested, we found that a canonical Hsp70-binding motif (the APPY peptide) functioned as a dose-dependent PQC degron both in yeast and in human cells. In yeast, Hsp70, Hsp110, Fes1, and the E3 Ubr1 target the APPY degron. Screening revealed that the sequence space within the chaperone-binding region of APPY that is compatible with degron function is vast. We find that the number of exposed Hsp70-binding sites in the yeast proteome correlates with a reduced protein abundance and half-life. Our results suggest that when protein folding fails, chaperone-binding sites may operate as PQC degrons, and PQC-linked degradation therefore overlaps in specificity with chaperone binding. This sheds new light on how the PQC system has evolved to exploit the intrinsic capacity of chaperones to recognize misfolded proteins, thereby placing them at the nexus of protein folding and degradation.Significance StatementIt is broadly accepted that misfolded proteins are often rapidly degraded by the ubiquitin-proteasome system, but how cells specifically recognize this immensely diverse group of proteins is largely unknown. Here we show that upon uncoupling of protein folding from protein degradation, a canonical chaperone-binding motif doubles as a degradation signal (degron), and that within the context of a Hsp70-binding region, many sequences are compatible with degron function. We find that degradation is correlated with the number of Hsp70-binding sites within a protein, and that the number of exposed Hsp70-binding sites in the yeast proteome correlates with more rapid degradation.

With the introduction of powerful mass spectrometry equipment into the field of snake venom proteomics, a large body of venomics data is accumulating. To allow for better comparison between venom compositions from different snake species and to provide an online database containing this data, we devised the Snake Venomics Display toolbox for visualization of snake venomics data on linear scales. This toolbox is freely available to be used online at https://tropicalpharmacology.com/tools/snake-venomics-display/ and allows researchers to visualize venomics data in a Relative Abundance (%) visualization mode and in an Absolute Abundance (mg) visualization mode, the latter taking venom yields into account. The curated venomics data for all snake species included in this database is also made available in a downloadable Excel file format. The Snake Venomics Display toolbox represents a simple way of handling snake venomics data, which is better suited for large data sets of venom compositions from multiple snake species.

SUMMARYIn combination with advanced mechanistic modeling and the generation of high-quality multi-dimensional data sets, machine learning is becoming an integral part of understanding and engineering living systems. Here we show that mechanistic and machine learning models can complement each other and be used in a combined approach to enable accurate genotype-to-phenotype predictions. We use a genome-scale model to pinpoint engineering targets and produce a large combinatorial library of metabolic pathway designs with different promoters which, once phenotyped, provide the basis for machine learning algorithms to be trained and used for new design recommendations. The approach enables successful forward engineering of aromatic amino acid metabolism in yeast, with the new recommended designs improving tryptophan production by up to 17% compared to the best designs used for algorithm training, and ultimately producing a total increase of 106% in tryptophan accumulation compared to optimized reference designs. Based on a single high-throughput data-generation iteration, this study highlights the power of combining mechanistic and machine learning models to enhance their predictive power and effectively direct metabolic engineering efforts.

Only 25% of bacterial membrane transporters have functional annotation owing to the difficulty of experimental study and of accurate prediction of their function. Here we report a sequence-independent method for high-throughput mining of novel transporters. The method is based on ligand-responsive biosensor systems that enable selective growth of cells only if they encode a ligand-specific importer. We developed such a synthetic selection system for thiamine pyrophosphate and mined soil and gut metagenomes for thiamine-uptake functions. We identified several members of a novel class of thiamine transporters, PnuT, which is widely distributed across multiple bacterial phyla. We demonstrate that with modular replacement of the biosensor, we could expand our method to xanthine and identify xanthine permeases from gut and soil metagenomes. Our results demonstrate how synthetic-biology approaches can effectively be deployed to functionally mine metagenomes and elucidate sequence-function relationships of small-molecule transport systems in bacteria.

Functional characterization of regulatory DNA elements in broad genetic contexts is a prerequisite for forward engineering of biological systems. Translation initiation site (TIS) sequences are attractive to use for regulating gene activity and metabolic pathway fluxes because the genetic changes are minimal. However, limited knowledge is available on tuning gene outputs by varying TISs in different genetic and environmental contexts. Here, we created TIS hexamer libraries in baker’s yeast Saccharomyces cerevisiae directly 5′ end of a reporter gene in various promoter contexts and measured gene activity distributions for each library. Next, selected TIS sequences, resulted in almost 10-fold changes in reporter outputs, were experimentally characterized in various environmental and genetic contexts in both yeast and mammalian cells. From our analyses, we observed strong linear correlations (R2 = 0.75–0.98) between all pairwise combinations of TIS order and gene activity. Finally, our analysis enabled the identification of a TIS with almost 50% stronger output than a commonly used TIS for protein expression in mammalian cells, and selected TISs were also used to tune gene activities in yeast at a metabolic branch point in order to prototype fitness and carotenoid production landscapes. Taken together, the characterized TISs support reliable context-independent forward engineering of translation initiation in eukaryotes.