Computational scoring and experimental evaluation of enzymes generated by neural networks.

In recent years, generative protein sequence models have been developed to sample novel sequences. However, predicting whether generated proteins will fold and function remains challenging. We evaluate a set of 20 diverse computational metrics to assess the quality of enzyme sequences produced by three contrasting generative models: ancestral sequence reconstruction, a generative adversarial network and a protein language model. Focusing on two enzyme families, we expressed and purified over 500 natural and generated sequences with 70-90% identity to the most similar natural sequences to benchmark computational metrics for predicting in vitro enzyme activity. Over three rounds of experiments, we developed a computational filter that improved the rate of experimental success by 50-150%. The proposed metrics and models will drive protein engineering research by serving as a benchmark for generative protein sequence models and helping to select active variants for experimental testing.

The Impact of Acute Nutritional Interventions on the Plasma Proteome.

CONTEXT: Humans respond profoundly to changes in diet, while nutrition and environment have a great impact on population health. It is therefore important to deeply characterize the human nutritional responses. OBJECTIVE: Endocrine parameters and the metabolome of human plasma are rapidly responding to acute nutritional interventions such as caloric restriction or a glucose challenge. It is less well understood whether the plasma proteome would be equally dynamic, and whether it could be a source of corresponding biomarkers. METHODS: We used high-throughput mass spectrometry to determine changes in the plasma proteome of i) 10 healthy, young, male individuals in response to 2 days of acute caloric restriction followed by refeeding; ii) 200 individuals of the Ely epidemiological study before and after a glucose tolerance test at 4 time points (0, 30, 60, 120 minutes); and iii) 200 random individuals from the Generation Scotland study. We compared the proteomic changes detected with metabolome data and endocrine parameters. RESULTS: Both caloric restriction and the glucose challenge substantially impacted the plasma proteome. Proteins responded across individuals or in an individual-specific manner. We identified nutrient-responsive plasma proteins that correlate with changes in the metabolome, as well as with endocrine parameters. In particular, our study highlights the role of apolipoprotein C1 (APOC1), a small, understudied apolipoprotein that was affected by caloric restriction and dominated the response to glucose consumption and differed in abundance between individuals with and without type 2 diabetes. CONCLUSION: Our study identifies APOC1 as a dominant nutritional responder in humans and highlights the interdependency of acute nutritional response proteins and the endocrine system.

Cell-cell metabolite exchange creates a pro-survival metabolic environment that extends lifespan.

Metabolism is deeply intertwined with aging. Effects of metabolic interventions on aging have been explained with intracellular metabolism, growth control, and signaling. Studying chronological aging in yeast, we reveal a so far overlooked metabolic property that influences aging via the exchange of metabolites. We observed that metabolites exported by young cells are re-imported by chronologically aging cells, resulting in cross-generational metabolic interactions. Then, we used self-establishing metabolically cooperating communities (SeMeCo) as a tool to increase metabolite exchange and observed significant lifespan extensions. The longevity of the SeMeCo was attributable to metabolic reconfigurations in methionine consumer cells. These obtained a more glycolytic metabolism and increased the export of protective metabolites that in turn extended the lifespan of cells that supplied them with methionine. Our results establish metabolite exchange interactions as a determinant of cellular aging and show that metabolically cooperating cells can shape the metabolic environment to extend their lifespan.

Learning deep representations of enzyme thermal adaptation.

Temperature is a fundamental environmental factor that shapes the evolution of organisms. Learning thermal determinants of protein sequences in evolution thus has profound significance for basic biology, drug discovery, and protein engineering. Here, we use a data set of over 3 million BRENDA enzymes labeled with optimal growth temperatures (OGTs) of their source organisms to train a deep neural network model (DeepET). The protein-temperature representations learned by DeepET provide a temperature-related statistical summary of protein sequences and capture structural properties that affect thermal stability. For prediction of enzyme optimal catalytic temperatures and protein melting temperatures via a transfer learning approach, our DeepET model outperforms classical regression models trained on rationally designed features and other deep-learning-based representations. DeepET thus holds promise for understanding enzyme thermal adaptation and guiding the engineering of thermostable enzymes.

Toward learning the principles of plant gene regulation.

Advanced machine learning (ML) algorithms produce highly accurate models of gene expression, uncovering novel regulatory features in nucleotide sequences involving multiple cis-regulatory regions across whole genes and structural properties. These broaden our understanding of gene regulation and point to new principles to test and adopt in the field of plant science.

Controlling gene expression with deep generative design of regulatory DNA.

Design of de novo synthetic regulatory DNA is a promising avenue to control gene expression in biotechnology and medicine. Using mutagenesis typically requires screening sizable random DNA libraries, which limits the designs to span merely a short section of the promoter and restricts their control of gene expression. Here, we prototype a deep learning strategy based on generative adversarial networks (GAN) by learning directly from genomic and transcriptomic data. Our ExpressionGAN can traverse the entire regulatory sequence-expression landscape in a gene-specific manner, generating regulatory DNA with prespecified target mRNA levels spanning the whole gene regulatory structure including coding and adjacent non-coding regions. Despite high sequence divergence from natural DNA, in vivo measurements show that 57% of the highly-expressed synthetic sequences surpass the expression levels of highly-expressed natural controls. This demonstrates the applicability and relevance of deep generative design to expand our knowledge and control of gene expression regulation in any desired organism, condition or tissue.

Enhanced metabolism and negative regulation of ER stress support higher erythropoietin production in HEK293 cells.

Recombinant protein production can cause severe stress on cellular metabolism, resulting in limited titer and product quality. To investigate cellular and metabolic characteristics associated with these limitations, we compare HEK293 clones producing either erythropoietin (EPO) (secretory) or GFP (non-secretory) protein at different rates. Transcriptomic and functional analyses indicate significantly higher metabolism and oxidative phosphorylation in EPO producers compared with parental and GFP cells. In addition, ribosomal genes exhibit specific expression patterns depending on the recombinant protein and the production rate. In a clone displaying a dramatically increased EPO secretion, we detect higher gene expression related to negative regulation of endoplasmic reticulum (ER) stress, including upregulation of ATF6B, which aids EPO production in a subset of clones by overexpression or small interfering RNA (siRNA) knockdown. Our results offer potential target pathways and genes for further development of the secretory power in mammalian cell factories.

Data mining of Saccharomyces cerevisiae mutants engineered for increased tolerance towards inhibitors in lignocellulosic hydrolysates.

The use of renewable plant biomass, lignocellulose, to produce biofuels and biochemicals using microbial cell factories plays a fundamental role in the future bioeconomy. The development of cell factories capable of efficiently fermenting complex biomass streams will improve the cost-effectiveness of microbial conversion processes. At present, inhibitory compounds found in hydrolysates of lignocellulosic biomass substantially influence the performance of a cell factory and the economic feasibility of lignocellulosic biofuels and chemicals. Here, we present and statistically analyze data on Saccharomyces cerevisiae mutants engineered for altered tolerance towards the most common inhibitors found in lignocellulosic hydrolysates: acetic acid, formic acid, furans, and phenolic compounds. We collected data from 7971 experiments including single overexpression or deletion of 3955 unique genes. The mutants included in the analysis had been shown to display increased or decreased tolerance to individual inhibitors or combinations of inhibitors found in lignocellulosic hydrolysates. Moreover, the data included mutants grown on synthetic hydrolysates, in which inhibitors were added at concentrations that mimicked those of lignocellulosic hydrolysates. Genetic engineering aimed at improving inhibitor or hydrolysate tolerance was shown to alter the specific growth rate or length of the lag phase, cell viability, and vitality, block fermentation, and decrease product yield. Different aspects of strain engineering aimed at improving hydrolysate tolerance, such as choice of strain and experimental set-up are discussed and put in relation to their biological relevance. While successful genetic engineering is often strain and condition dependent, we highlight the conserved role of regulators, transporters, and detoxifying enzymes in inhibitor tolerance. The compiled meta-analysis can guide future engineering attempts and aid the development of more efficient cell factories for the conversion of lignocellulosic biomass.

A proteomic survival predictor for COVID-19 patients in intensive care.

Global healthcare systems are challenged by the COVID-19 pandemic. There is a need to optimize allocation of treatment and resources in intensive care, as clinically established risk assessments such as SOFA and APACHE II scores show only limited performance for predicting the survival of severely ill COVID-19 patients. Additional tools are also needed to monitor treatment, including experimental therapies in clinical trials. Comprehensively capturing human physiology, we speculated that proteomics in combination with new data-driven analysis strategies could produce a new generation of prognostic discriminators. We studied two independent cohorts of patients with severe COVID-19 who required intensive care and invasive mechanical ventilation. SOFA score, Charlson comorbidity index, and APACHE II score showed limited performance in predicting the COVID-19 outcome. Instead, the quantification of 321 plasma protein groups at 349 timepoints in 50 critically ill patients receiving invasive mechanical ventilation revealed 14 proteins that showed trajectories different between survivors and non-survivors. A predictor trained on proteomic measurements obtained at the first time point at maximum treatment level (i.e. WHO grade 7), which was weeks before the outcome, achieved accurate classification of survivors (AUROC 0.81). We tested the established predictor on an independent validation cohort (AUROC 1.0). The majority of proteins with high relevance in the prediction model belong to the coagulation system and complement cascade. Our study demonstrates that plasma proteomics can give rise to prognostic predictors substantially outperforming current prognostic markers in intensive care.

metaGEM: reconstruction of genome scale metabolic models directly from metagenomes.

Metagenomic analyses of microbial communities have revealed a large degree of interspecies and intraspecies genetic diversity through the reconstruction of metagenome assembled genomes (MAGs). Yet, metabolic modeling efforts mainly rely on reference genomes as the starting point for reconstruction and simulation of genome scale metabolic models (GEMs), neglecting the immense intra- and inter-species diversity present in microbial communities. Here, we present metaGEM (https://github.com/franciscozorrilla/metaGEM), an end-to-end pipeline enabling metabolic modeling of multi-species communities directly from metagenomes. The pipeline automates all steps from the extraction of context-specific prokaryotic GEMs from MAGs to community level flux balance analysis (FBA) simulations. To demonstrate the capabilities of metaGEM, we analyzed 483 samples spanning lab culture, human gut, plant-associated, soil, and ocean metagenomes, reconstructing over 14,000 GEMs. We show that GEMs reconstructed from metagenomes have fully represented metabolism comparable to isolated genomes. We demonstrate that metagenomic GEMs capture intraspecies metabolic diversity and identify potential differences in the progression of type 2 diabetes at the level of gut bacterial metabolic exchanges. Overall, metaGEM enables FBA-ready metabolic model reconstruction directly from metagenomes, provides a resource of metabolic models, and showcases community-level modeling of microbiomes associated with disease conditions allowing generation of mechanistic hypotheses.

Plastic-Degrading Potential across the Global Microbiome Correlates with Recent Pollution Trends.

Biodegradation is a plausible route toward sustainable management of the millions of tons of plastic waste that have accumulated in terrestrial and marine environments. However, the global diversity of plastic-degrading enzymes remains poorly understood. Taking advantage of global environmental DNA sampling projects, here we constructed hidden Markov models from experimentally verified enzymes and mined ocean and soil metagenomes to assess the global potential of microorganisms to degrade plastics. By controlling for false positives using gut microbiome data, we compiled a catalogue of over 30,000 nonredundant enzyme homologues with the potential to degrade 10 different plastic types. While differences between the ocean and soil microbiomes likely reflect the base compositions of these environments, we find that ocean enzyme abundance increases with depth as a response to plastic pollution and not merely taxonomic composition. By obtaining further pollution measurements, we observed that the abundance of the uncovered enzymes in both ocean and soil habitats significantly correlates with marine and country-specific plastic pollution trends. Our study thus uncovers the earth microbiome's potential to degrade plastics, providing evidence of a measurable effect of plastic pollution on the global microbial ecology as well as a useful resource for further applied research. IMPORTANCE Utilization of synthetic biology approaches to enhance current plastic degradation processes is of crucial importance, as natural plastic degradation processes are very slow. For instance, the predicted lifetime of a polyethylene terephthalate (PET) bottle under ambient conditions ranges from 16 to 48 years. Moreover, although there is still unexplored diversity in microbial communities, synergistic degradation of plastics by microorganisms holds great potential to revolutionize the management of global plastic waste. To this end, the methods and data on novel plastic-degrading enzymes presented here can help researchers by (i) providing further information about the taxonomic diversity of such enzymes as well as understanding of the mechanisms and steps involved in the biological breakdown of plastics, (ii) pointing toward the areas with increased availability of novel enzymes, and (iii) giving a basis for further application in industrial plastic waste biodegradation. Importantly, our findings provide evidence of a measurable effect of plastic pollution on the global microbial ecology.

Potential for improved retention rate by personalized antiseizure medication selection: A register-based analysis.

OBJECTIVE: The first antiseizure medication (ASM) is ineffective or intolerable in 50% of epilepsy cases. Selection between more than 25 available ASMs is guided by epilepsy factors, but also age and comorbidities. Randomized evidence for particular patient subgroups is seldom available. We asked whether register data could be used for retention rate calculations based on demographics, comorbidities, and ASM history, and quantified the potential improvement in retention rates of the first ASM in several large epilepsy cohorts. We also describe retention rates in patients with epilepsy after traumatic brain injury and dementia, patient groups with little available evidence. METHODS: We used medical, demographic, and drug prescription data from epilepsy cohorts from comprehensive Swedish registers, containing 6380 observations. By analyzing 381 840 prescriptions, we studied retention rates of first- and second-line ASMs for patients with epilepsy in multiple sclerosis (MS), brain infection, dementia, traumatic brain injury, or stroke. The rank of retention rates of ASMs was validated by comparison to published randomized control trials. We identified the optimal stratification for each brain disease, and quantified the potential improvement if all patients had received the optimal ASM. RESULTS: Using optimal stratification for each brain disease, the potential improvement in retention rate (percentage points) was MS, 20%; brain infection, 21%; dementia, 14%; trauma, 21%; and stroke, 14%. In epilepsy after trauma, levetiracetam had the highest retention rate at 80% (95% confidence interval [CI] = 65-89), exceeding that of the most commonly prescribed ASM, carbamazepine (p = .04). In epilepsy after dementia, lamotrigine (77%, 95% CI = 68-84) and levetiracetam (74%, 95% CI = 68-79) had higher retention rates than carbamazepine (p = .006 and p = .01, respectively). SIGNIFICANCE: We conclude that personalized ASM selection could improve retention rates and that national registers have potential as big data sources for personalized medicine in epilepsy.

Performance of Regression Models as a Function of Experiment Noise.

BACKGROUND: A challenge in developing machine learning regression models is that it is difficult to know whether maximal performance has been reached on the test dataset, or whether further model improvement is possible. In biology, this problem is particularly pronounced as sample labels (response variables) are typically obtained through experiments and therefore have experiment noise associated with them. Such label noise puts a fundamental limit to the metrics of performance attainable by regression models on the test dataset. RESULTS: We address this challenge by deriving an expected upper bound for the coefficient of determination (R 2) for regression models when tested on the holdout dataset. This upper bound depends only on the noise associated with the response variable in a dataset as well as its variance. The upper bound estimate was validated via Monte Carlo simulations and then used as a tool to bootstrap performance of regression models trained on biological datasets, including protein sequence data, transcriptomic data, and genomic data. CONCLUSIONS: The new method for estimating upper bounds for model performance on test data should aid researchers in developing ML regression models that reach their maximum potential. Although we study biological datasets in this work, the new upper bound estimates will hold true for regression models from any research field or application area where response variables have associated noise.

Learning the Regulatory Code of Gene Expression.

Data-driven machine learning is the method of choice for predicting molecular phenotypes from nucleotide sequence, modeling gene expression events including protein-DNA binding, chromatin states as well as mRNA and protein levels. Deep neural networks automatically learn informative sequence representations and interpreting them enables us to improve our understanding of the regulatory code governing gene expression. Here, we review the latest developments that apply shallow or deep learning to quantify molecular phenotypes and decode the cis-regulatory grammar from prokaryotic and eukaryotic sequencing data. Our approach is to build from the ground up, first focusing on the initiating protein-DNA interactions, then specific coding and non-coding regions, and finally on advances that combine multiple parts of the gene and mRNA regulatory structures, achieving unprecedented performance. We thus provide a quantitative view of gene expression regulation from nucleotide sequence, concluding with an information-centric overview of the central dogma of molecular biology.

A time-resolved proteomic and prognostic map of COVID-19.

COVID-19 is highly variable in its clinical presentation, ranging from asymptomatic infection to severe organ damage and death. We characterized the time-dependent progression of the disease in 139 COVID-19 inpatients by measuring 86 accredited diagnostic parameters, such as blood cell counts and enzyme activities, as well as untargeted plasma proteomes at 687 sampling points. We report an initial spike in a systemic inflammatory response, which is gradually alleviated and followed by a protein signature indicative of tissue repair, metabolic reconstitution, and immunomodulation. We identify prognostic marker signatures for devising risk-adapted treatment strategies and use machine learning to classify therapeutic needs. We show that the machine learning models based on the proteome are transferable to an independent cohort. Our study presents a map linking routinely used clinical diagnostic parameters to plasma proteomes and their dynamics in an infectious disease.

Ultra-fast proteomics with Scanning SWATH.

Accurate quantification of the proteome remains challenging for large sample series and longitudinal experiments. We report a data-independent acquisition method, Scanning SWATH, that accelerates mass spectrometric (MS) duty cycles, yielding quantitative proteomes in combination with short gradients and high-flow (800 µl min-1) chromatography. Exploiting a continuous movement of the precursor isolation window to assign precursor masses to tandem mass spectrometry (MS/MS) fragment traces, Scanning SWATH increases precursor identifications by ~70% compared to conventional data-independent acquisition (DIA) methods on 0.5-5-min chromatographic gradients. We demonstrate the application of ultra-fast proteomics in drug mode-of-action screening and plasma proteomics. Scanning SWATH proteomes capture the mode of action of fungistatic azoles and statins. Moreover, we confirm 43 and identify 11 new plasma proteome biomarkers of COVID-19 severity, advancing patient classification and biomarker discovery. Thus, our results demonstrate a substantial acceleration and increased depth in fast proteomic experiments that facilitate proteomic drug screens and clinical studies.

Benchmarking accuracy and precision of intensity-based absolute quantification of protein abundances in Saccharomyces cerevisiae.

Protein quantification via label-free mass spectrometry (MS) has become an increasingly popular method for predicting genome-wide absolute protein abundances. A known caveat of this approach, however, is the poor technical reproducibility, that is, how consistent predictions are when the same sample is measured repeatedly. Here, we measured proteomics data for Saccharomyces cerevisiae with both biological and inter-batch technical triplicates, to analyze both accuracy and precision of protein quantification via MS. Moreover, we analyzed how these metrics vary when applying different methods for converting MS intensities to absolute protein abundances. We demonstrate that our simple normalization and rescaling approach can perform as accurately, yet more precisely, than methods which rely on external standards. Additionally, we show that inter-batch reproducibility is worse than biological reproducibility for all evaluated methods. These results offer a new benchmark for assessing MS data quality for protein quantification, while also underscoring current limitations in this approach.

Bayesian genome scale modelling identifies thermal determinants of yeast metabolism.

The molecular basis of how temperature affects cell metabolism has been a long-standing question in biology, where the main obstacles are the lack of high-quality data and methods to associate temperature effects on the function of individual proteins as well as to combine them at a systems level. Here we develop and apply a Bayesian modeling approach to resolve the temperature effects in genome scale metabolic models (GEM). The approach minimizes uncertainties in enzymatic thermal parameters and greatly improves the predictive strength of the GEMs. The resulting temperature constrained yeast GEM uncovers enzymes that limit growth at superoptimal temperatures, and squalene epoxidase (ERG1) is predicted to be the most rate limiting. By replacing this single key enzyme with an ortholog from a thermotolerant yeast strain, we obtain a thermotolerant strain that outgrows the wild type, demonstrating the critical role of sterol metabolism in yeast thermosensitivity. Therefore, apart from identifying thermal determinants of cell metabolism and enabling the design of thermotolerant strains, our Bayesian GEM approach facilitates modelling of complex biological systems in the absence of high-quality data and therefore shows promise for becoming a standard tool for genome scale modeling.