Predicting growth rate from gene expression

Microbial reaction rate estimation using proteins and proteomes

Microbes transform their environments using diverse enzymatic reactions. However, it remains challenging to measure microbial reaction rates in natural environments. Despite advances in global quantification of enzyme abundances, the individual relationships between enzyme abundances and their reaction rates have not been systematically examined. Using matched proteomic and reaction rate data from microbial cultures, we show that enzyme abundance is often insufficient to predict its corresponding reaction rate. However, we discovered that global proteomic measurements can be used to make accurate rate predictions of individual reaction rates (median R 2 = 0.78). Accurate rate predictions required only a small number of proteins and they did not need explicit prior mechanistic knowledge or environmental context. These results indicate that proteomes are encoders of cellular reaction rates, potentially enabling proteomic measurements in situ to estimate the rates of microbially mediated reactions in natural systems.

Significance

One of the most basic phenotypes of a microbe is its set of associated reaction rates, but quantifying these rates in situ remains extremely challenging, especially in natural systems. We used molecular data and statistical models to estimate microbial rates in steady state cultures. We found that many reaction rates are highly predictable using proteomic data, though single proteins are typically not informative for their associated reaction rates. This result suggests that gene expression data from complex microbial communities could be used to estimate in situ reaction rates, providing new clues into the lives and environmental function of microbes.

Cell reprogramming design by transfer learning of functional transcriptional networks

Recent developments in synthetic biology, next-generation sequencing, and machine learning provide an unprecedented opportunity to rationally design new disease treatments based on measured responses to gene perturbations and drugs to reprogram cells. The main challenges to seizing this opportunity are the incomplete knowledge of the cellular network and the combinatorial explosion of possible interventions, both of which are insurmountable by experiments. To address these challenges, we develop a transfer learning approach to control cell behavior that is pre-trained on transcriptomic data associated with human cell fates, thereby generating a model of the network dynamics that can be transferred to specific reprogramming goals. The approach combines transcriptional responses to gene perturbations to minimize the difference between a given pair of initial and target transcriptional states. We demonstrate our approach's versatility by applying it to a microarray dataset comprising >9,000 microarrays across 54 cell types and 227 unique perturbations, and an RNASeq dataset consisting of >10,000 sequencing runs across 36 cell types and 138 perturbations. Our approach reproduces known reprogramming protocols with an AUROC of 0.91 while innovating over existing methods by pre-training an adaptable model that can be tailored to specific reprogramming transitions. We show that the number of gene perturbations required to steer from one fate to another increases with decreasing developmental relatedness and that fewer genes are needed to progress along developmental paths than to regress. These findings establish a proof-of-concept for our approach to computationally design control strategies and provide insights into how gene regulatory networks govern phenotype.

Cell reprogramming design by transfer learning of functional transcriptional networks

Recent developments in synthetic biology, next-generation sequencing, and machine learning provide an unprecedented opportunity to rationally design new disease treatments based on measured responses to gene perturbations and drugs to reprogram cell behavior. The main challenges to seizing this opportunity are the incomplete knowledge of the cellular network and the combinatorial explosion of possible interventions, both of which are insurmountable by experiments. To address these challenges, we develop a transfer learning approach to control cell behavior that is pre-trained on transcriptomic data associated with human cell fates to generate a model of the functional network dynamics that can be transferred to specific reprogramming goals. The approach additively combines transcriptional responses to gene perturbations (single-gene knockdowns and overexpressions) to minimize the transcriptional difference between a given pair of initial and target states. We demonstrate the flexibility of our approach by applying it to a microarray dataset comprising over 9,000 microarrays across 54 cell types and 227 unique perturbations, and an RNASeq dataset consisting of over 10,000 sequencing runs across 36 cell types and 138 perturbations. Our approach reproduces known reprogramming protocols with an average AUROC of 0.91 while innovating over existing methods by pre-training an adaptable model that can be tailored to specific reprogramming transitions. We show that the number of gene perturbations required to steer from one fate to another increases as the developmental relatedness decreases. We also show that fewer genes are needed to progress along developmental paths than to regress. Together, these findings establish a proof-of-concept for our approach to computationally design control strategies and demonstrate their ability to provide insights into the dynamics of gene regulatory networks.

Predicting metabolic fluxes from omics data via machine learning: Moving from knowledge-driven towards data-driven approaches

The accurate prediction of phenotypes in microorganisms is a main challenge for systems biology. Genome-scale models (GEMs) are a widely used mathematical formalism for predicting metabolic fluxes using constraint-based modeling methods such as flux balance analysis (FBA). However, they require prior knowledge of the metabolic network of an organism and appropriate objective functions, often hampering the prediction of metabolic fluxes under different conditions. Moreover, the integration of omics data to improve the accuracy of phenotype predictions in different physiological states is still in its infancy. Here, we present a novel approach for predicting fluxes under various conditions. We explore the use of supervised machine learning (ML) models using transcriptomics and/or proteomics data and compare their performance against the standard parsimonious FBA (pFBA) approach using case studies of Escherichia coli organism as an example. Our results show that the proposed omics-based ML approach is promising to predict both internal and external metabolic fluxes with smaller prediction errors in comparison to the pFBA approach. The code, data, and detailed results are available at the project's repository[1].

Reforestation of high elevation pines: Direct seeding success depends on seed source and sowing environment

Forest persistence in regions impacted by increasing water and temperature stress will depend upon species' ability to either rapidly adjust to novel conditions or migrate to track ecological niches. Predicted, rapid climate change is likely to outpace the adaptive and migratory capacity of long-lived isolated tree species, and reforestation may be critical to species' persistence. Facilitating persistence both within and beyond a species' range requires identification of seed lots best adapted to the current and future conditions predicted with rapid climate change. We evaluate variation in emergent seedling performance that leads to differential survival among species and populations for three high elevation five-needle pines. We paired a fully reciprocal field common garden experiment with a greenhouse common garden study to (1) quantify variation in seedling emergence and functional traits, (2) ask how functional traits affect performance under different establishment conditions, and (3) evaluate whether trait and performance variation demonstrates local adaptation and plasticity. Among study species-limber, Great Basin bristlecone, and whitebark pines-we found divergence in emergence and functional traits, though soil moisture was the strongest driver of seedling emergence and abundance across all species. Generalist limber pine had a clear emergence advantage as well as traits associated with drought adaptation, while edaphic specialist bristlecone pine was characterized by low emergence yet high early survival once established. Despite evidence for edaphic specialization, soil characteristics alone did not explain bristlecone success. Across species, trait-environment relationships provided some evidence for local adaptation in drought-adapted traits, but we found no evidence of local adaptation in emergence or survival at this early life stage. For managers looking to promote persistence, sourcing seed from drier environments is likely to impart greater drought resistance into reforestation efforts through strategies such as greater root investment, increasing the probability of early seedling survival. This research demonstrates, through a rigorous reciprocal transplant experimental design, that it may be possible to select climate- and soil-appropriate seed sources for reforestation. However, planting success will ultimately rely on a suitable establishment environment, requiring careful consideration of interannual climate variability for management interventions in these climate and disturbance-impacted tree species.

The proteomic landscape of genome-wide genetic perturbations

Functional genomic strategies have become fundamental for annotating gene function and regulatory networks. Here, we combined functional genomics with proteomics by quantifying protein abundances in a genome-scale knockout library in Saccharomyces cerevisiae, using data-independent acquisition mass spectrometry. We find that global protein expression is driven by a complex interplay of (1) general biological properties, including translation rate, protein turnover, the formation of protein complexes, growth rate, and genome architecture, followed by (2) functional properties, such as the connectivity of a protein in genetic, metabolic, and physical interaction networks. Moreover, we show that functional proteomics complements current gene annotation strategies through the assessment of proteome profile similarity, protein covariation, and reverse proteome profiling. Thus, our study reveals principles that govern protein expression and provides a genome-spanning resource for functional annotation.

Growth rate-associated transcriptome reorganization in response to genomic, environmental, and evolutionary interruptions

The genomic, environmental, and evolutionary interruptions caused the changes in bacterial growth, which were stringently associated with changes in gene expression. The growth and gene expression changes remained unclear in response to these interruptions that occurred combinative. As a pilot study, whether and how bacterial growth was affected by the individual and dual interruptions of genome reduction, environmental stress, and adaptive evolution were investigated. Growth assay showed that the presence of the environmental stressors, i.e., threonine and chloramphenicol, significantly decreased the growth rate of the wild-type Escherichia coli, whereas not that of the reduced genome. It indicated a canceling effect in bacterial growth due to the dual interruption of the genomic and environmental changes. Experimental evolution of the reduced genome released the canceling effect by improving growth fitness. Intriguingly, the transcriptome architecture maintained a homeostatic chromosomal periodicity regardless of the genomic, environmental, and evolutionary interruptions. Negative epistasis in transcriptome reorganization was commonly observed in response to the dual interruptions, which might contribute to the canceling effect. It was supported by the changes in the numbers of differentially expressed genes (DEGs) and the enriched regulons and functions. Gene network analysis newly constructed 11 gene modules, one out of which was correlated to the growth rate. Enrichment of DEGs in these modules successfully categorized them into three types, i.e., conserved, responsive, and epistatic. Taken together, homeostasis in transcriptome architecture was essential to being alive, and it might be attributed to the negative epistasis in transcriptome reorganization and the functional differentiation in gene modules. The present study directly connected bacterial growth fitness with transcriptome reorganization and provided a global view of how microorganisms responded to genomic, environmental, and evolutionary interruptions for survival from wild nature.

Epigenetic regulation of fetal brain development in pig

How fetal brain development is regulated at the molecular level is not well understood. Due to ethical challenges associated with research on the human fetus, large animals particularly pigs are increasingly used to study development and disorders of fetal brain. The pig fetal brain grows rapidly during the last ∼ 50 days before birth which is around day 60 (d60) of pig gestation. But what regulates the onset of accelerated growth of the brain is unknown. The current study tests the hypothesis that epigenetic alteration around d60 is involved in the onset of rapid growth of fetal brain of pig. To test this hypothesis, DNA methylation changes of fetal brain was assessed in a genome-wide manner by Enzymatic Methyl-seq (EM-seq) during two gestational periods (GP): d45 vs. d60 (GP1) and d60 vs. d90 (GP2). The cytosine-guanine (CpG) methylation data was analyzed in an integrative manner with the RNA-seq data generated from the same brain samples from our earlier study. A neural network based modeling approach was implemented to learn changes in methylation patterns of the differentially expressed genes, and then predict methylations of the brain in a genome-wide manner during rapid growth. This approach identified specific methylations that changed in a mutually informative manner during rapid growth of the fetal brain. These methylations were significantly overrepresented in specific genic as well as intergenic features including CpG islands, introns, and untranslated regions. In addition, sex-bias methylations of known single nucleotide polymorphic sites were also identified in the fetal brain ide during rapid growth.

Discovery of positive and purifying selection in metagenomic time series of hypermutator microbial populations

A general method to infer both positive and purifying selection during the real-time evolution of hypermutator pathogens would be broadly useful. To this end, we introduce a Simple Test to Infer Mode of Selection (STIMS) from metagenomic time series of evolving microbial populations. We test STIMS on metagenomic data generated by simulations of bacterial evolution, and on metagenomic data spanning 62,750 generations of Lenski's long-term evolution experiment with Escherichia coli (LTEE). This benchmarking shows that STIMS detects positive selection in both nonmutator and hypermutator populations, and purifying selection in hypermutator populations. Using STIMS, we find strong evidence of ongoing positive selection on key regulators of the E. coli gene regulatory network, even in some hypermutator populations. STIMS also detects positive selection on regulatory genes in hypermutator populations of Pseudomonas aeruginosa that adapted to subinhibitory concentrations of colistin-an antibiotic of last resort-for just twenty-six days of laboratory evolution. Our results show that the fine-tuning of gene regulatory networks is a general mechanism for rapid and ongoing adaptation. The simplicity of STIMS, together with its intuitive visual interpretation, make it a useful test for positive and purifying selection in metagenomic data sets that track microbial evolution in real-time.

A conserved expression signature predicts growth rate and reveals cell & lineage-specific differences

Isogenic cells cultured together show heterogeneity in their proliferation rate. To determine the differences between fast and slow-proliferating cells, we developed a method to sort cells by proliferation rate, and performed RNA-seq on slow and fast proliferating subpopulations of pluripotent mouse embryonic stem cells (mESCs) and mouse fibroblasts. We found that slowly proliferating mESCs have a more naïve pluripotent character. We identified an evolutionarily conserved proliferation-correlated transcriptomic signature that is common to all eukaryotes: fast cells have higher expression of genes for protein synthesis and protein degradation. This signature accurately predicted growth rate in yeast and cancer cells, and identified lineage-specific proliferation dynamics during development, using C. elegans scRNA-seq data. In contrast, sorting by mitochondria membrane potential revealed a highly cell-type specific mitochondria-state related transcriptome. mESCs with hyperpolarized mitochondria are fast proliferating, while the opposite is true for fibroblasts. The mitochondrial electron transport chain inhibitor antimycin affected slow and fast subpopulations differently. While a major transcriptional-signature associated with cell-to-cell heterogeneity in proliferation is conserved, the metabolic and energetic dependency of cell proliferation is cell-type specific.

By performing RNA sequencing on cells sorted by their proliferation rate, this study identifies a gene expression signature capable of predicting proliferation rates in diverse eukaryotic cell types and species. This signature, applied to single-cell RNA sequencing data from embryos of the roundworm C. elegans, reveals lineage-specific proliferation differences during development. In contrast to the universality of the proliferation signature, mitochondria and metabolism related genes show a high degree of cell-type specificity; mouse pluripotent stem cells (mESCs) and differentiated cells (fibroblasts) exhibit opposite relations between mitochondria state and proliferation. Furthermore, we identified a slow proliferating subpopulation of mESCs with higher expression of pluripotency genes. Finally, we show that fast and slow proliferating subpopulations are differentially sensitive to mitochondria inhibitory drugs in different cell types.

Highlights:

A FACS-based method to determine the transcriptomes of fast and slow proliferating subpopulations.

A universal proliferation-correlated transcriptional signature indicates high protein synthesis and degradation in fast proliferating cells across cell types and species.

Applied to scRNA-seq, the expression signature predicts the global proliferation slowdown during C. elegans development.

Mitochondria membrane potential predicts proliferation rate in a cell-type specific manner, with ETC complex III inhibitor having distinct effects on fibroblasts vs mESCs.

Metabolic engineering of microorganisms for the production of multifunctional non-protein amino acids: γ-aminobutyric acid and δ-aminolevulinic acid

Gamma-aminobutyric acid (GABA) and delta-aminolevulinic acid (ALA), playing important roles in agriculture, medicine and other fields, are multifunctional non-protein amino acids with similar and comparable properties and biosynthesis pathways. Recently, microbial synthesis has become an inevitable trend to produce GABA and ALA due to its green and sustainable characteristics. In addition, the development of metabolic engineering and synthetic biology has continuously accelerated and increased the GABA and ALA yield in microorganisms. Here, focusing on the current trends in metabolic engineering strategies for microbial synthesis of GABA and ALA, we analysed and compared the efficiency of various metabolic strategies in detail. Moreover, we provide the insights to meet challenges of realizing industrially competitive strains and highlight the future perspectives of GABA and ALA production.

Beyond the chemical master equation: Stochastic chemical kinetics coupled with auxiliary processes

The chemical master equation and its continuum approximations are indispensable tools in the modeling of chemical reaction networks. These are routinely used to capture complex nonlinear phenomena such as multimodality as well as transient events such as first-passage times, that accurately characterise a plethora of biological and chemical processes. However, some mechanisms, such as heterogeneous cellular growth or phenotypic selection at the population level, cannot be represented by the master equation and thus have been tackled separately. In this work, we propose a unifying framework that augments the chemical master equation to capture such auxiliary dynamics, and we develop and analyse a numerical solver that accurately simulates the system dynamics. We showcase these contributions by casting a diverse array of examples from the literature within this framework and applying the solver to both match and extend previous studies. Analytical calculations performed for each example validate our numerical results and benchmark the solver implementation.

Populations of genetically identical cells tend to exhibit remarkable variability. This seemingly counter-intuitive observation has broad and fascinating implications, and has thus been a focal point of biological modeling. Many important processes act on this cellular heterogeneity at the population level, leading to an intricate coupling between the single-cell and the population-level dynamics. For example, selection pressures or growth rates may depend crucially on the expression of a particular gene (or gene family). Classical single-cell modeling approaches, such as the chemical master equation, can accurately describe the mechanisms driving cellular noise, however, they cannot encapsulate how the aforementioned auxiliary processes affect the population composition. In this work, we propose a unifying framework that extends the classical chemical master equation to faithfully capture the single-cell variability alongside the population-level evolution. We develop, analyse, and showcase an open-source numerical tool to simulate the dynamics of such populations in time. The tool is designed for straightforward use by a non-technical audience: a high-level description of the underlying chemical and population-level processes suffices to simulate complex system dynamics. Simultaneously, we retain high customisability of the underlying mathematical representation for the more advanced user. Ultimately, the unifying framework and the associated computational tool open new horizons in the study of how fundamental microscopic dynamics give rise to complex macroscopic phenomena.

Multimodal regularised linear models with flux balance analysis for mechanistic integration of omics data

MOTIVATION

High-throughput biological data, thanks to technological advances, have become cheaper to collect, leading to the availability of vast amounts of omic data of different types. In parallel, the in silico reconstruction and modelling of metabolic systems is now acknowledged as a key tool to complement experimental data on a large scale. The integration of these model- and data-driven information is therefore emerging as a new challenge in systems biology, with no clear guidance on how to better take advantage of the inherent multi-source and multi-omic nature of these data types while preserving mechanistic interpretation.

RESULTS

Here we investigate different regularisation techniques for high-dimensional data derived from the integration of gene expression profiles with metabolic flux data, extracted from strain-specific metabolic models, to improve cellular growth rate predictions. To this end, we propose ad-hoc extensions of previous regularisation frameworks including group, view-specific and principal component regularisation, and experimentally compare them using data from 1,143 Saccharomyces cerevisiae strains. We observe a divergence between methods in terms of regression accuracy and integration effectiveness based on the type of regularisation employed. In multi-omic regression tasks, when learning from experimental and model-generated omic data, our results demonstrate the competitiveness and ease of interpretation of multimodal regularised linear models compared to data-hungry methods based on neural networks.

AVAILABILITY

All data, models, and code produced in this work are available on GitHub at https://github.com/Angione-Lab/HybridGroupIPFLasso_pc2Lasso.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Biosystems Design by Machine Learning

Biosystems such as enzymes, pathways, and whole cells have been increasingly explored for biotechnological applications. However, the intricate connectivity and resulting complexity of biosystems poses a major hurdle in designing biosystems with desirable features. As -omics and other high throughput technologies have been rapidly developed, the promise of applying machine learning (ML) techniques in biosystems design has started to become a reality. ML models enable the identification of patterns within complicated biological data across multiple scales of analysis and can augment biosystems design applications by predicting new candidates for optimized performance. ML is being used at every stage of biosystems design to help find nonobvious engineering solutions with fewer design iterations. In this review, we first describe commonly used models and modeling paradigms within ML. We then discuss some applications of these models that have already shown success in biotechnological applications. Moreover, we discuss successful applications at all scales of biosystems design, including nucleic acids, genetic circuits, proteins, pathways, genomes, and bioprocesses. Finally, we discuss some limitations of these methods and potential solutions as well as prospects of the combination of ML and biosystems design.

A mechanism-aware and multiomic machine-learning pipeline characterizes yeast cell growth

Metabolic modeling and machine learning are key components in the emerging next generation of systems and synthetic biology tools, targeting the genotype-phenotype-environment relationship. Rather than being used in isolation, it is becoming clear that their value is maximized when they are combined. However, the potential of integrating these two frameworks for omic data augmentation and integration is largely unexplored. We propose, rigorously assess, and compare machine-learning-based data integration techniques, combining gene expression profiles with computationally generated metabolic flux data to predict yeast cell growth. To this end, we create strain-specific metabolic models for 1,143 Saccharomyces cerevisiae mutants and we test 27 machine-learning methods, incorporating state-of-the-art feature selection and multiview learning approaches. We propose a multiview neural network using fluxomic and transcriptomic data, showing that the former increases the predictive accuracy of the latter and reveals functional patterns that are not directly deducible from gene expression alone. We test the proposed neural network on a further 86 strains generated in a different experiment, therefore verifying its robustness to an additional independent dataset. Finally, we show that introducing mechanistic flux features improves the predictions also for knockout strains whose genes were not modeled in the metabolic reconstruction. Our results thus demonstrate that fusing experimental cues with in silico models, based on known biochemistry, can contribute with disjoint information toward biologically informed and interpretable machine learning. Overall, this study provides tools for understanding and manipulating complex phenotypes, increasing both the prediction accuracy and the extent of discernible mechanistic biological insights.

Genetic interactions derived from high-throughput phenotyping of 6589 yeast cell cycle mutants

Over the last 30 years, computational biologists have developed increasingly realistic mathematical models of the regulatory networks controlling the division of eukaryotic cells. These models capture data resulting from two complementary experimental approaches: low-throughput experiments aimed at extensively characterizing the functions of small numbers of genes, and large-scale genetic interaction screens that provide a systems-level perspective on the cell division process. The former is insufficient to capture the interconnectivity of the genetic control network, while the latter is fraught with irreproducibility issues. Here, we describe a hybrid approach in which the 630 genetic interactions between 36 cell-cycle genes are quantitatively estimated by high-throughput phenotyping with an unprecedented number of biological replicates. Using this approach, we identify a subset of high-confidence genetic interactions, which we use to refine a previously published mathematical model of the cell cycle. We also present a quantitative dataset of the growth rate of these mutants under six different media conditions in order to inform future cell cycle models.

Distinguishing cell phenotype using cell epigenotype

The relationship between microscopic observations and macroscopic behavior is a fundamental open question in biophysical systems. Here, we develop a unified approach that-in contrast with existing methods-predicts cell type from macromolecular data even when accounting for the scale of human tissue diversity and limitations in the available data. We achieve these benefits by applying a k-nearest-neighbors algorithm after projecting our data onto the eigenvectors of the correlation matrix inferred from many observations of gene expression or chromatin conformation. Our approach identifies variations in epigenotype that affect cell type, thereby supporting the cell-type attractor hypothesis and representing the first step toward model-independent control strategies in biological systems.

Knowledge-guided analysis of "omics" data using the KnowEnG cloud platform

We present Knowledge Engine for Genomics (KnowEnG), a free-to-use computational system for analysis of genomics data sets, designed to accelerate biomedical discovery. It includes tools for popular bioinformatics tasks such as gene prioritization, sample clustering, gene set analysis, and expression signature analysis. The system specializes in "knowledge-guided" data mining and machine learning algorithms, in which user-provided data are analyzed in light of prior information about genes, aggregated from numerous knowledge bases and encoded in a massive "Knowledge Network." KnowEnG adheres to "FAIR" principles (findable, accessible, interoperable, and reuseable): its tools are easily portable to diverse computing environments, run on the cloud for scalable and cost-effective execution, and are interoperable with other computing platforms. The analysis tools are made available through multiple access modes, including a web portal with specialized visualization modules. We demonstrate the KnowEnG system's potential value in democratization of advanced tools for the modern genomics era through several case studies that use its tools to recreate and expand upon the published analysis of cancer data sets.