Ai-driven platform for accelerating the cultivation of novel bacteria

Abstract

Systems, methods, and processes for predicting and validating one or more previously unknowing growth conditions for a microorganism which can allow for the culturing of previously unculturable microorganisms, or can identify optimal grown conditions for a known or unknown microorganism.

Description

[0001] AI-DRIVEN PLATFORM FOR ACCELERATING THE CULTIVATION OF NOVEL BACTERIA

[0002] CROSS-REFERENCE TO RELATED APPLICATIONS

[0003] This International PCT Application claims the benefit of and priority to U.S. Provisional Application No. 63 / 539,392, filed September 20, 2023. The entire specification, claims, and figures of the above-referenced application is hereby incorporated, in its entirety by reference. TECHNICAL FIELD

[0004] The invention described herein related to the field of microbiology and machine-learning systems, and in particular the application of machine-learning systems and processes to identify and predict growth conditions for unknown or unculturable microorganisms.

[0005] BACKGROUND

[0006] Predicting the environmental preferences of organisms is an important goal in ecology. If the conditions under which a given taxon can thrive are known, then scientists can better predict biogeographical distributions, guide ecological restoration efforts, design effective probiotics, and understand taxon-specific responses to global change factors. Unfortunately, the environmental preferences of most microbial taxa and the genomic attributes associated with those preferences often remain undetermined. One reason for this is that most microorganisms, particularly those found in non-host associated environments, can be difficult to cultivate in vitro , making it difficult to measure environmental preferences directly. Even for those taxa that can be cultivated, quantifying how microbial growth rates vary across broad environmental gradients can be time consuming and the environmental gradients created in vitro may not necessarily mimic those found in situ. However, when direct information on environmental preferences can be collected, such information can be very useful for predicting microbial distributions and functions across space and time.

[0007] Even without the direct measurement of environmental preferences, it is feasible to infer some environmental preferences from genomic information. Information contained in curated genomic databases can be leveraged, which can include both cultivated and uncultivated microbial taxa, to infer the specific environmental preferences of uncharacterized microorganisms. For example, genomic information from isolates whose environmental preferences have been measured in vitro has been used to infer the preferences of bacteria across gradients in oxygen and temperature. Using genomic information to determine the environmental preferences of microbial taxa can have important ramifications. For example, it could improve the ability to predict community assembly across different environmental gradients, identify the conditions under which specific taxa can thrive, and better optimize medium conditions to improve the cultivability of fastidious taxa. However, genome-based inferences of environmental preferences can be difficult to validate (especially for uncultivated taxa), and it is not always knowing which genes or other genomic attributes are associated with adaptations to specific environmental conditions of interest.

[0008] Consider microbial preferences for specific pH conditions. Currently, it is not now possible to predict bacterial pH preferences from genomic information alone, although it is known that pH is often a key factor determining the niche space occupied by microorganisms. The distributions of specific bacterial taxa and the overall composition of bacterial communities are often strongly associated with gradients in pH, as has been observed in a wide range of environments including soils, freshwater, and geothermal systems. Despite the importance of this environmental factor, the pH preferences of most bacterial taxa remain undetermined, although it is known that bacterial pH preferences can vary widely. This knowledge gap is particularly evident in nonhost-associated systems that are often dominated by taxa that are difficult to cultivate and study in vitro . Even across cultivated taxa, pH tolerances and pH optima for growth are rarely determined experimentally, and most cultivation media likely select for taxa that grow at near-neutral pH conditions. It should be noted that “pH preference” across a given gradient is akin to the “realized niche” of a population (as opposed to the “fundamental niche”), where pH preference is the pH at which an organism achieves maximal relative abundances in nature. This relative abundance is determined by the metabolically optimal pH for growth as well as other biotic or abiotic constraints. For example, a given taxon may grow optimally at around pH 7 under controlled conditions, but its pH preference could be lower if its abundance in a given environment is maximized at pH 6 due to its interplay with other factors.

[0009] In the absence of whole genome sequence data, one might still be able to predict bacterial pH preferences from taxonomic and phylogenetic information alone. Shifts in bacterial community composition across pH gradients can be evident at broad levels of taxonomic resolution, suggesting some degree of conservatism in the traits related to pH adaptation. However, closely related taxa can have very distinct pH preferences, such as bacteria within the phylum Acidobacteria which tend to be more abundant in lower pH soils, a pattern that does not necessarily hold for subgroups within the phylum. Thus, it is not clear whether taxonomy and phylogenetic information is sufficient to predict bacterial pH preferences. Previous studies have focused on identifying common adaptations to changes in pH conditions across different taxonomic groups and the genes or transcripts associated with those adaptations. This generates the expectation that the presence or absence of specific functional genes can be used to predict bacterial pH preferences. An approach that integrates biogeographic distributions of multiple individual taxa with genomic information across environmental gradients can address whether taxonomic, phylogenetic, or genomic information can be used to predict microbial environmental preferences.

[0010] Applicants set out to determine whether bacterial pH preferences are predictable. In other words, Applicants asked whether taxonomic, phylogenetic, or genomic information could be used to predict where along a pH gradient a taxon is most likely to achieve its highest relative abundance (which Applicants define here as its “pH preference”), searching for patterns that are generalizable across distinct environment types. To do so, Applicants used information on bacterial distributions across five independent datasets that span large pH gradients in soil and freshwater systems to infer the putative pH preferences of the bacterial taxa found in these environments. Applicants then used this information to assess the degree to which bacterial pH preferences are phylogenetically and taxonomically conserved. To determine the genomic features associated with adaptations to pH, Applicants analyzed representative genomes from taxa with varying pH preferences, as inferred from the cultivation-independent analyses, and identified genes that are consistently associated with differences in bacterial pH preference across environments. Last, Applicants developed and validated a machine learning model that enables the accurate identification of bacterial pH preferences from the presence or absence of 56 functional genes, making it feasible to infer pH preferences for both uncultivated and cultivated taxa. More generally, Applicants demonstrate how the workflow shown in Fig. 1 can be used to investigate other bacterial environmental preferences and the genes associated with environmental adaptations while overcoming the limitations of cultivation-based experimental approaches to expand traitbased understanding of microorganisms.

[0011] SUMMARY OF THE INVENTION

[0012] In one aspect, the present invention is directed to systems, methods, and process for implementing an Al model trained on genomes of previously cultured microorganisms. As described below, in one preferred embodiment the invention predicts the ideal growth media and culturing conditions for both known microorganisms with unidentified optimal conditions and those never before successfully cultured. In a preferred embodiment, the invention includes methods and systems for culturing previously uncultured microorganisms as well as refining culturing conditions for scaled microbial cultivation. In other preferred embodiment, using annotated genomic information from microbial samples, refined cultivation parameters, generally referred to herein as growth conditions such pH, temperature, osmotic potential, oxygen requirements can be determined and optimized.

[0013] Additional aspects of the invention will be evident from the specification, figures, tables, and claims provided herein.

[0014] BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Fig. 1A-C. Workflow to predict environmental preferences from genomic and biogeographical information. (A) Marker gene sequencing datasets that include samples spanning a broad range in the environmental factor of interest (pH in this case) are chosen and rare taxa are excluded on the basis of occurrence. The remaining taxa are then used for the estimation of the environmental conditions at which the relative abundances of individual taxa are highest using either bootstrapping or generalized additive model approaches. (B) The taxa for which an environmental preference has been statistically estimated are then queried against curated genome databases. Genomes matching the query sequence at high similarity (e.g., at 99.6% in this study) are considered representative of the taxa. (C) All genes present in those genomes are then tested for their relationship to environmental preference, and those that are significantly associated with environmental preference (e g., sharing the same statistically significant positive or negative relationship with pH across datasets) are incorporated into a machine learning modeling framework. Predictions can be validated using an independent test set of the data, from pH preferences estimated in independent datasets, or from phenotypic assays.

[0016] Fig. 2A-E. Range of pH preferences of those amplicon sequence variants (ASVs) from which pH preferences could effectively be inferred from the biogeographical patterns in the marker gene sequencing data. The taxonomic assignments of the ASVs are shown at the phylum level with more complete taxonomic information included in Fig. 7. (A to E) correspond to the different datasets included in the study (brown labels, soil; turquoise labels, freshwater). Values at the end of the boxplots correspond to the number of ASVs with estimated pH preferences in each phylum. The length and width of the violins depict the entire range of pH preferences and the most frequent pH preferences of taxa within each phylum, respectively. SRS, Savannah River Site; AUS, Australia; PAN, Panama; ROMAINE, La Romaine watershed; CARBBAS, Carbon Biogeochemistry in Boreal Aquatic Systems.

[0017] Fig. 3A-E. Phylogenetic distributions of the most abundant amplicon sequence variants (ASVs) with estimated pH preferences in each dataset. Panels show the phylogenetic trees of the most abundant ASVs found in soil (A to C) and freshwater (D and E) habitats. Clades are colored by the affdiation at the phylum level. The inner ring depicts the estimated pH preference of the ASVs, and the outer ring indicates whether a genome with no more than one nucleotide mismatch to the amplicon sequence is available in the Genome Taxonomy Database (GTDB). The displays are maximum likelihood (ML) phylogenetic trees, and these trees only show the 300 most abundant ASVs per dataset, with the complete set of ASVs (numbers indicated in Fig, 2) used for the calculation of the phylogenetic signal in pH preferences (see “Phylogenetic visualization and analysis” section and Fig. 8). The minimum pH preference is depicted in white, and the maximum is depicted in blue according to the legend. SRS, Savannah River Site; AUS, Australia; PAN, Panama; ROMAINE, La Romaine watershed; CARBBAS, Carbon Biogeochemistry in Boreal Aquatic Systems.

[0018] Fig. 4. Associations between gene categories and estimated pH preferences in bacteria across soils and freshwater systems. Positive trends (blue cells) indicate that the presence of the gene is associated with inferred bacterial preferences for higher pH, and negative trends (orange cells) indicate that presence of the gene is associated with bacterial preferences for lower pH conditions. Applicantsonly included gene types that had consistent relationships to pH preference in at least three datasets and / or in both soil and freshwater. Applicants highlighted categories with well-known involvement in pH tolerance and response in bacteria. For a complete summary of the putative functionality, Pfam identifier, and available references linking these specific gene types to pH adaptations (see Fig. 15). The x axis was sorted left to right in the order of Savannah River Site (SRS), Australia (AUS), Panama (PAN), La Romaine watershed (ROMAINE), and Carbon Biogeochemistry in Boreal Aquatic Systems (CARBBAS) datasets, respectively.

[0019] Fig. 5A-E. Correlations between the estimated and predicted pH preferences of bacterial taxa based on a gradient boosted decision tree model built on the presence / absence of 56 top predictor gene types. (A to E) Each panel corresponds to a distinct dataset from either soil (brown) or freshwater (blue). The regressions on the plot include the predictions using the full dataset (black) and the independent validation using 10% of the genomes (orange) not included in the model development. The gray-shaded area depicts the SE around the fitted line. SRS, Savannah River Site; AUS, Australia; PAN, Panama; ROMAINE, La Romaine watershed; CARBBAS, Carbon Biogeochemistry in Boreal Aquatic Systems

[0020] Fig. 6A-E. Distribution of samples across pH gradients in the datasets used for this study. The datasets contained 16S rRNA gene sequencing data of bacteria across soils (SRS, USA; AUS, Australia, (18); PAN, Panama, (18); depicted in brown), and freshwater (ROMAINE, Canada, (21); CARBBAS; Canada, (22); depicted in turquoise). The number of samples (n) included in each dataset is indicated. Applicants also show Mantel tests with 9999 permutations between sample pH and bacterial community composition (pairwise Bray-Curtis distances). In all cases, the Mantel tests were significant (P<0.001), indicating strong associations between sample pH and overall bacterial community composition for each of the 5 datasets.

[0021] Fig. 7A-B. Relative abundance of the AS Vs with inferred pH preferences classified at the phylum level (A) and range in pH preference of the ASVs across taxonomic levels (B). The panels in (A) correspond to soil (SRS, AUS, and PAN; in brown) and freshwater (ROMAINE and CARBBAS; in turquoise) datasets. Asterisks in panel (B) indicate a statistically significant difference in the range of pH preference between these broad taxonomic categories based on Welch two-sample tests (Holm-Bonferroni-corrected P < 0.05)

[0022] Fig. 8. Phylogenetic correlograms showing the phylogenetic depth at which pH preference is conserved across soil and freshwater systems. The text on the graphs shows the Pagel’s and Blomberg’s k statistics for phylogenetic signal in pH preference with statistical significance set at P < 0.05*, P < 0.01**, or P < 0.001***. The mean pairwise phylogenetic distances between taxa within the same family, order, and class are 0.15, 0.25 and 0.30, respectively. Colored lines on the x-axes depict whether pH preference is phylogenetically conserved (red), phylogenetically overdispersed (blue) or showing no phylogenetic signal (black).

[0023] Fig. 9. Proportion of genomes with inferred pH preferences were obtained from bacterial isolates (“Isolate”) compared to environmental (metagenome) and single cell-assembled genomes (“Assembled”) as indicated in the Genome Taxonomy Database (GTDB).

[0024] Fig. 10 -B. The numbers of ASVs with inferred pH preferences (full bars) and those ASVs which also had representative genomes available (dark grey) (A), and the taxonomic affiliations of the representative genomes at the phylum level (B). Fig. 11. Summary of the impact on bacterial pH preference predictions of the 56 genes used to predict bacterial pH preferences with machine learning. Positive values have a positive effect on pH preference predictions (higher inferred pH preferences), and negative values have a negative effect on pH preference predictions (lower inferred pH preferences). The coloring indicates the impact on model predictions of the presence (red) and absence (blue) of a given gene type in a genome.

[0025] Fig. 12. Correlation between the estimated and predicted pH preferences of bacterial taxa based on a gradient boosted decision tree model based on the presence / absence of 56 top predictor gene types. These data are for the 148 representative genomes obtained for taxa with estimated pH preferences in a study of soils across the UK (38).

[0026] Fig. 13. Low phylogenetic signal of pH preference in bacteria from soil and freshwater habitats. Pearson’s correlation coefficients are shown on the panels that had a statistically significant relationship between the Euclidean distance in pH preference and the phylogenetic distance based on branch lengths between genomes.

[0027] Fig. 14. Features of the ASVs included in the Application. The ubiquity threshold forced ASVs to occur in at least 20 samples for each study.

[0028] Fig. 15. List of genes significantly associated with bacterial pH preference across habitats. When known, the main functions associated with the genes are reported, and the direction of their relationship to pH preference is indicated as either positive (+) or negative (-). When available, we also included the relevant references reporting a putative role of the genes in bacterial pH preference. These 56 genes were included in a regression predictive modelling framework based on gradient boosted decision trees that predicts pH preference of bacterial taxa based on presence / absence information on those genes.

[0029] DETAILED DESCRIPTION OF THE INVENTION

[0030] While the invention has been particularly shown and described with reference to a number of embodiments, it would be understood by those skilled in the art that changes in the form and details may be made to the various embodiments disclosed herein without departing from the spirit and scope of the invention and that the various embodiments disclosed herein are not intended to act as limitations on the scope of the claims. All references cited herein are incorporated in their entirety by reference. The terminology used herein is for describing particular embodiments and is not intended to be limiting. As used herein, the singular forms “a,” “and” and “the” include plural referents unless the content and context clearly dictate otherwise. Thus, for example, a reference to “a” or “the” growth condition may include a combination of two or more such growth conditions. Unless defined otherwise herein, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. Although any methods and materials similar or equivalent to those described herein find use in the practice or testing of the embodiments disclosed herein, some methods and materials are described in detail and represent preferred embodiments of the current inventive technology.

[0031] The present invention includes systems and methods for identifying a growth condition of a microorganism, and preferably an unknown or unculturable bacterium. In a preferred embodiment, the method of the invention includes annotating the genomes of a first training sample of microorganisms to identify genes and / or proteins that are associated with a known target growth condition. In this embodiment, the target growth condition of the microorganisms are known. This annotated genomic training set can be applied to an Al-based system machine learning system to train a neural network to infer a target growth condition. As shown in the workflow of Figure 1, in a preferred embodiment, a reference sample of microorganisms, such as a reference sample of bacteria, and preferably unknown or unculturable bacteria can be established. In this embodiment, the target growth condition of the population making up the reference sample is unknown, or non-optimized. In this embodiment, the genomes of the reference sample are annotated to to identify genes and / or proteins that are associated with a target growth condition. This annotated genomic data can be used, preferably by an Al-based machine learning system to predict a growth condition of one or more microorganisms from the reference set based on the presence of absence of the one or more of the genes and / or proteins associated with a target growth condition. Having predicted a growth condition, one or more microorganisms can be cultured using the predicted conditions, preferably with positive and / or negative growth controls to validated the predicted growth condition in vitro.

[0032] As used herein, the term “microorganism” means cells that are not mammalian (e.g., bacteria, fungi, yeast, archaea). Typical microbe samples may be obtained from the earth (e.g., soil, permafrost, sediments), water (e.g., fresh water, seawater, deep-sea vents), air, materials in the environment (e.g., decaying materials like rotting wood, compost), from the surface (e.g., skin) of animals (e.g., mammals, insects, worms), from inside (e.g., digestive tract, gut) animals (e.g., humans), from plants or plant-associated material (e g., plant roots, plant seeds), and the like. Environmental samples may also be procured from man-made or artificial environments (e.g., wastewater, activated sludge, hospitals, and ventilation systems). In general, the environmental samples may be procured from natural environments, artificial environments, from attempted replications of natural environments, and the like. In one example, the environmental sample is soil. In one example, the environmental sample is water. In another embodiment, a sample can be generated in silico, for example from known or established databases of genomic libraries for various microorganisms.

[0033] A proportion of the microbes in samples from some of these environments may be “unculturable.” Unculturable microbes are so named because, using current technologies, the microbes cannot be cultured (e.g., in the laboratory). The unculturable microbes may be detected in the samples, however, because techniques like DNA sequencing can detect genomes of the unculturable microbes. There are also microbes in some environmental samples whose DNA has not been detected. These microbes may be called “unknown” because they are not currently known to exist. An unknown microbe may be culturable or may be unculturable. An unculturable microbe may be known or may be unknown. In one example, unculturable and / or unknown microbes may be prokaryotic (e.g., archaea, bacteria). In one example, unculturable and / or unknown microbes may be eukaryotic (e.g., fungi, yeast). When a conserved nucleotide sequence (e.g., a 16s rRNA nucleotide sequence) has been obtained from a microbe from a sample, for example, it may be possible to determine if the conserved sequence originates from an unculturable and / or unknown microbe. Various publicly available nucleotide sequence databases contain, for example, 16S rRNA sequences that are generally designated as originating from either culturable or unculturable microbes. A 16S rRNA sequence, for example, can be used to query these databases for identical sequences (e.g., using BLASTN). Sequence matches resulting from these sequence queries (i.e., retrieved sequences in the databases that have identity with the query sequence) will generally be identified as originating from either culturable or unculturable microbes and may help determine whether the query sequence originates from a microbe known to be culturable or unculturable. Lack of a sequence match between a query sequence and sequences in a database may indicate that the 16S rRNA query sequence originates from a microbe that has not previously been described (i.e., an unknown microbe).

[0034] A “sequence match” between a query sequence and a retrieved sequence may be determined based on identity between the query sequence and the retrieved sequence, the identity generally measurable over a given length (e g., 99% of the length) of the query sequence. Sequence identity matches may include identity levels between the query and retrieved sequences, for example, of at least 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, or 100%. Lack of a sequence identity match may include identity levels of less than any of these stated levels. In one example, a sequence identity match of at least 94, 95, or 96% between a 16S rRNA query sequence and a sequence retrieved from a database may indicate that the microbe from which the query sequence originated should be classified within the same taxonomic genus as the microbe from which the retrieved sequence originated. In one example, a sequence identity match of at least 94, 95, or 96% between a 16S rRNA query sequence and a retrieved sequence, where the retrieved sequence is indicated to be from an unculturable microbe, may indicate that the microbe from which the query sequence originated should also be classified as unculturable, at least until such a microbe is cultured using the methods disclosed herein.

[0035] As used herein, the term “bacteria” or “bacterium” means a prokaryotic organism that have peptidoglycan in their cell walls and lipids in their membranes that contain fatty acids.

[0036] As used herein, “culturing,” when referring to microbes, means to grow or proliferate the microbes. “Cultured from” refers to the source from which the growing microbes were obtained. “Cultured in” or “cultured on” refers to where the microbes are cultured. For example, a microbe that is cultured on a medium containing agar, is generally being grown on a medium that is solid or semisolid in consistency.

[0037] As used herein, the term “culturable” means a microorganism, and preferably a bacteria that has been cultured previously, or for which the culture conditions are known.

[0038] As used herein, the term “unculturable,” when referring to a microbe, means unable to be cultured, using current technologies (i.e., technologies prior to this disclosure). A microbe that is considered unculturable may eventually be cultured, for example, when technologies are improved. In one example, a microbe cultured using the methods disclosed herein may not have been cultured previously. In the context of this disclosure, such a microbe would be called unculturable because it was the technological improvement disclosed herein that resulted in the microbe being cultured.

[0039] As used herein, “unknown,” when referring to a microbe, means that the microbe was not previously known to exist or was known to exists but is unculturable. As used herein, the term “genome” refers to the complete set of genes or genetic material present in a cell or organism. The genome includes DNA or RNA in RNA viruses. The genome includes both the genes, (the coding regions), the noncoding DNA and the genomes of the mitochondria and chloroplasts.

[0040] As used herein, the term “gene” means the deoxyribonucleotide sequences comprising the coding region of a structural gene. A “gene” may also include non-translated sequences located adj acent to the coding region on both the 5 ' and 3 ' ends such that the gene corresponds to the length of the full-length mRNA. The sequences which are located 5' of the coding region and which are present on the mRNA are referred to as 5' non-translated sequences. The sequences which are located 3' or downstream of the coding region and which are present on the mRNA are referred to as 3' non-translated sequences. The term “gene” encompasses both cDNA and genomic forms of a gene. A genomic form or clone of a gene contains the coding region interrupted with noncoding sequences termed “introns” or “intervening regions” or “intervening sequences.” Introns are segments of a gene which are transcribed into heterogenous nuclear RNA (hnRNA); introns may contain regulatory elements such as enhancers. Introns are removed or “spliced out” from the nuclear or primary transcript; introns therefore are absent in the messenger RNA (mRNA) transcript. The mRNA functions during translation to specify the sequence or order of amino acids in a nascent polypeptide.

[0041] The term “polypeptide” is equally used herein with the term “ polypeptide.” Proteins (including fragments thereof, preferably biologically active fragments, and peptides, usually having less than 30 amino acids) comprise one or more amino acids coupled to each other via a covalent peptide bond (resulting in a chain of amino acids). The term “polypeptide” as used herein describes a group of molecules, which, for example, consist of more than 30 amino acids. Polypeptides may further form multimers such as dimers, trimers and higher oligomers, i.e., consisting of more than one polypeptide molecule. Polypeptide molecules forming such dimers, trimers etc. may be identical or non-identical. The corresponding higher order structures of such multimers are, consequently, termed homo- or heterodimers, homo- or heterotrimers etc. The terms “polypeptide” and “protein” also refer to naturally modified polypeptides / proteins wherein the modification is affected e.g., by post-translational modifications like glycosylation, acetylation, phosphorylation and the like. As used herein, a “machine learning system” is a trained computational model that takes a training sample of annotated genomic data infers a growth condition as generally described herein. Examples of machine learning models include neural networks, including recurrent neural networks and convolutional neural networks; random forests models, including random forests; restricted Boltzmann machines; recurrent tensor networks; and gradient boosted trees. The term “classifier” (or classification model) is sometimes used to describe all forms of classification model including deep learning models (e.g., neural networks having many layers) as well as random forests models.

[0042] As used herein, a machine learning system may include a deep learning model that may include a function approximation method aiming to develop custom dictionaries configured to achieve a given task, be it classification or dimension reduction. It may be implemented in various forms such as by a neural network (e g., a convolutional neural network), etc. In general, though not necessarily, it includes multiple layers. Each such layer includes multiple processing nodes and the layers process in sequence, with nodes of layers closer to the model input layer processing before nodes of layers closer to the model output. In various embodiments, one-layer feeds to the next, etc. The output layer may include nodes that represent various classifications. In some embodiments, a deep learning model is a model that takes data with very little preprocessing, although it may be segmented data such as cellular artifact, or other features of interest may be extracted from an image, and outputs a classification of the cellular artifact.

[0043] In various embodiments, a deep learning model may have significant depth and can classify a large or heterogeneous array of growth conditions. In some contexts, the term “deep” means that model has a plurality of layers of processing nodes that receive values from preceding layers (or as direct inputs) and that output values to succeeding layers (or the final output). Interior nodes are often “hidden” in the sense that their input and output values are not visible outside the model. In various embodiments, the operation of the hidden nodes may not be monitored or recorded during operation. The nodes and connections of a deep learning model can be trained, for example with a “training sample,” and retrained without redesigning their number, arrangement, interface inputs, etc, and yet classify a large heterogeneous range of growth conditions and the like.

[0044] In various aspects, provided herein are systems and methods for identifying and optionally characterizing a growth condition in an example microorganism such as an uncultured bacterium, by analyzing the growth condition from a test sample, sometimes referred to as a training sample, and thereby generating a test dataset and comparing it to a training dataset generated from a reference sample, and optionally one or more additional samples. A growth condition in this embodiment may include a growth condition selected from: pH, culture media, nutrients, temperature, osmotic potential, water, oxygen, or a combination of the same.

[0045] In certain embodiments, machine learning systems may include artificial neural networks (ANNs) which are a type of computational system that can learn the relationships between an input data set and a target data set. ANN name originates from a desire to develop a simplified mathematical representation of a portion of the human neural system, intended to capture its “learning” and “generalization” abilities. ANNs are a major foundation in the field of artificial intelligence. ANNs are widely applied in research because they can model highly non-linear systems in which the relationship among the variables is unknown or very complex. ANNs are typically trained on empirically observed data sets. The data set may conventionally be divided into a training set, a test set, and a validation set.

[0046] In supervised learning applications, the labeled data is used to form an objective function (e.g., cross-entropy loss, “triplet “loss, “Siamese” loss, or custom loss functions encoding physical information). The network parameters are updated to optimize the specified loss function. In particular, a type of neural network called a feed-forward back-propagation classifier can be trained on an input data set to generate growth condition representations minimizing the cost function over the training samples. Variants of stochastic gradient descent are often used to search parameter space in combination with the back-propagation algorithm to minimize the cost function specified over the training data inputs. After a large number of training iterations, the ANN parameter updates may be stopped; the stopping criteria typically leverages evaluations of the network on the validation data set (the other stopping criteria can be applied).

[0047] Many types of machine learning models may be employed in certain embodiments of inventive technology. In general, such models take as inputs such as annotated genomic data, and with little or no additional preprocessing, they infer growth conditions without further intervention. Two primary embodiments of machine learning models may include “deep” convolutional neural network (ConvNet) models and a randomized Principal Component Analysis (PCA) random forests model. However, other forms machine learning model may be employed in the context of this disclosure. A random forests model is relatively easy to generate from a training dataset and may employ relatively fewer training set members. A convolutional neural network may be more time-consuming and computationally expensive to generate from a training set, but it tends to be better at accurately classifying growth conditions.

[0048] In certain embodiment, the Al-machine learning systems can be operated using machine readable instructions, generally referred to as software. The hardware devices of the invention can be any kind of device that can be programmed including, for example, any kind of computer including smart mobile devices (watches, phones, tablets, and the like), personal computers, powerful servers or supercomputers, or the like. The device includes one or more processors such as an ASIC or any combination processors, for example, one general purpose processor and two FPGAs. The device may be implemented as a combination of hardware and software, such as an ASIC and an FPGA, or at least one microprocessor and at least one memory with software modules located therein. In various embodiments, the system includes at least one hardware component and / or at least one software component. The embodiments described herein could be implemented in pure hardware or partly in hardware and partly in software. In some cases, the disclosed embodiments may be implemented on different hardware devices, for example using a plurality of CPUs equipped with GPUs capable of accelerating scientific computation.

[0049] Each computational element may be implemented as an organized collection of computer data and instructions. In certain embodiments, an image acquisition algorithm and a machine learning model can each be viewed as a form of application software that interfaces with a user and with system software. System software typically interfaces with computer hardware, typically implemented as one or more processors (e g., CPUs or ASICs as mentioned) and associated memory. In certain embodiments, the system software includes operating system software and / or firmware, as well as any middleware and drivers installed in the system. The system software provides basic non-task-specific functions of the computer. In contrast, the modules and other application software are used to accomplish specific tasks. Each native instruction for a module is stored in a memory device and is represented by a numeric value.

[0050] At one level a computational element is implemented as a set of commands prepared by the programmer / developer. However, the module software that can be executed by the computer hardware is executable code committed to memory using “machine codes” selected from the specific machine language instruction set, or “native instructions,” designed into the hardware processor. The machine language instruction set, or native instruction set, is known to, and essentially built into, the hardware processor(s). This is the “language” by which the system and application software communicates with the hardware processors. Each native instruction is a discrete code that is recognized by the processing architecture and that can specify particular registers for arithmetic, addressing, or control functions; particular memory locations or offsets; and particular addressing modes used to interpret operands. More complex operations are built up by combining these simple native instructions, which are executed sequentially, or as otherwise directed by control flow instructions. The inter-relationship between the executable software instructions and the hardware processor may be structural. In other words, the instructions per se may include a series of symbols or numeric values. They do not intrinsically convey any information. It is the processor, which by design was preconfigured to interpret the symbols / numeric values, which imparts meaning to the instructions.

[0051] In certain embodiments, the modules or systems generally used herein may be configured to execute on a single machine at a single location, on multiple machines at a single location, or on multiple machines at multiple locations. When multiple machines are employed, the individual machines may be tailored for their particular tasks. For example, operations requiring large blocks of code and / or significant processing capacity may be implemented on large and / or stationary machines not suitable for mobile or field operations. Such operations may be implemented on hardware remote from the site where the sample is processed, for example on a server or server farm connected by a network to a field device that captures the sample image, or through a cloudbased network. Less computationally intensive operations may be implemented on a portable or mobile device used in the field for image capture.

[0052] Embodiments of the present disclosure may comprise or utilize a special purpose or general-purpose computer including computer hardware, such as, for example, one or more processors and system memory, as discussed in greater detail below. Embodiments within the scope of the present disclosure also include physical and other computer-readable media for carrying or storing computer-executable instructions and / or data structures. In particular, one or more of the processes described herein may be implemented at least in part as instructions embodied in a non-transitory computer-readable medium and executable by one or more computing devices ( e.g., any of the media content access devices described herein). In general, a processor (e.g., a microprocessor) receives instructions, from a non-transitory computer-readable medium, (e.g., a memory, etc.), and executes those instructions, thereby performing one or more processes, including one or more of the processes described herein. Computer-readable media can be any available media that can be accessed by a general purpose or special purpose computer system. Computer-readable media that store computerexecutable instructions are non-transitory computer-readable storage media (devices). Computer- readable media that carry computer-executable instructions are transmission media. Thus, by way of example, and not limitation, embodiments of the disclosure can comprise at least two distinctly different kinds of computer-readable media: non-transitory computer-readable storage media (devices) and transmission media. Non-transitory computer-readable storage media (devices) includes RAM, ROM, EEPROM, CD-ROM, solid state drives (“SSDs”) (e.g., based on ~), Flash memory, phase-change memory (“PCM”), other types of memory, other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store desired program code means in the form of computer-executable instructions or data structures and which can be accessed by a general purpose or special purpose computer.

[0053] A “network” is defined as one or more data links that enable the transport of electronic data between computer systems and / or modules and / or other electronic devices. When information is transferred or provided over a network or another communications connection ( either hardwired, wireless, or a combination of hardwired or wireless) to a computer, the computer properly views the connection as a transmission medium. Transmissions media can include a network and / or data links which can be used to carry desired program code means in the form of computer-executable instructions or data structures and which can be accessed by a general purpose or special purpose computer. Combinations of the above should also be included within the scope of computer readable media.

[0054] Further, upon reaching various computer system components, program code means in the form of computer-executable instructions or data structures can be transferred automatically from transmission media to non-transitory computer-readable storage media (devices) (or vice versa). For example, computer-executable instructions or data structures received over a network or data link can be buffered in RAM within a network interface module (e.g., a “NIC”), and then eventually transferred to computer system RAM and / or to less volatile computer storage media (devices) at a computer system. Thus, it should be understood that non-transitory computer- readable storage media (devices) can be included in computer system components that also (or even primarily) utilize transmission media. Computer-executable instructions comprise, for example, instructions and data which, when executed at a processor, cause a general purpose computer, special purpose computer, or special purpose processing device to perform a certain function or group of functions. In some embodiments, computer-executable instructions are executed on a general-purpose computer to turn the general-purpose computer into a special purpose computer implementing elements of the disclosure. The computer executable instructions may be, for example, binaries, intermediate format instructions such as assembly language, or even source code. Although the subject matter has been described in language specific to structural features and / or methodological acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the described features or acts described above. Rather, the described features and acts are disclosed as example forms of implementing the claims.

[0055] Those skilled in the art will appreciate that the disclosure may be practiced in network computing environments with many types of computer system configurations, including, personal computers, desktop computers, laptop computers, message processors, hand-held devices, multiprocessor systems, microprocessor-based or programmable consumer electronics, network PCs, minicomputers, mainframe computers, mobile telephones, PDAs, tablets, pagers, routers, switches, and the like. The disclosure may also be practiced in distributed system environments where local and remote computer systems, which are linked ( either by hardwired data links, wireless data links, or by a combination of hardwired and wireless data links) through a network, both perform tasks. In a distributed system environment, program modules may be located in both local and remote memory storage devices.

[0056] Embodiments of the present disclosure can also be implemented in cloud computing environments. In this description, “cloud computing” is defined as a model for enabling on- demand network access to a shared pool of configurable computing resources. For example, cloud computing can be employed in the marketplace to offer ubiquitous and convenient on-demand access to the shared pool of configurable computing resources. The shared pool of configurable computing resources can be rapidly provisioned via virtualization and released with low management effort or service provider interaction, and then scaled accordingly. A cloudcomputing model can be composed of various characteristics such as, for example, on-demand self-service, broad network access, resource pooling, rapid elasticity, measured service, and so forth. A cloud-computing model can also expose various service models, such as, for example, Software as a Service (“SaaS”), Platform as a Service (“PaaS”), and Infrastructure as a Service (“laaS”). A cloud-computing model can also be deployed using different deployment models such as private cloud, community cloud, public cloud, hybrid cloud, and so forth. In this description and in the claims, a “cloud-computing environment” is an environment in which cloud computing is employed.

[0057] The invention now being generally described will be more readily understood by reference to the following examples, which are included merely for the purposes of illustration of certain aspects of the embodiments of the present invention. The examples are not intended to limit the invention, as one of skill in the art would recognize from the above teachings and the following examples that other techniques and methods can satisfy the claims and can be employed without departing from the scope of the claimed invention.

[0058] EXAMPLES

[0059] Example 1 : Overview and Inventive Rationale.

[0060] Applicants first inferred bacterial pH preferences using biogeographical information from natural pH gradients found in distinct systems (soil and freshwater environments) (Fig. 1A). Applicants acknowledge that pH is unlikely to be the only factor influencing the distributions of bacteria in these systems and that the pH preferences of many taxa could not be inferred using this approach as there are other biotic or abiotic factors that are of similar, or greater, importance in shaping their distributions. However, Applicants note that this approach is similar to the approach routinely used in plant and animal systems to quantify the relationships between environmental factors and growth optima or tolerances. After inferring pH preferences from the biogeographical information, Applicants then matched the 16S ribosomal RNA (rRNA) gene sequences from those taxa for which pH preferences could be inferred to representative genomes, identifying sets of genes that are consistently associated with pH preference based on their presence or absence in a given genome (Fig. IB). These genes were then incorporated into a machine learning modeling framework (Fig. 1C) to predict pH preferences, with the approach validated using independent test sets.

[0061] The biogeographical analyses included 16S rRNA gene sequence data from a total of 795 soil samples and 675 freshwater samples spanning a pH range from 3 to 10 with a total of 250,275 amplicon sequence variants (ASVs) included in downstream analyses (Fig. 6 and Fig. 14). These data came from five independent datasets, and, in all cases, pH was an important driver of overall bacterial community composition, with Mantel p values ranging from 0.37 [La Romaine watershed (ROMAINE), freshwater] to 0.78 [Panama (PAN), tropical forest soils] (Fig. 6). These patterns are expected considering that pH is often observed to have a strong influence on bacterial community structure in many systems and considering that each of these sample sets was specifically selected to span broad gradients in pH. Applicants also note that a broad diversity of bacterial taxa was found within and across each of the five sample sets (Fig. 7A), which is important as Applicants were trying to identify patterns in pH preferences across a broad array of taxa. Using Applicants conservative approach, Applicants were able to estimate the pH preferences of bacterial taxa in these environmental samples for 0.5 to 4.9% of all ASVs per dataset (Fig. 14). The analyses of the taxonomic and phylogenetic signals in bacterial pH preferences and the search for representative genomes from these taxa were ultimately based on a total of 4568 ASVs (468 to 1614 ASVs per dataset) spanning 38 bacterial phyla.

[0062] Example 2: Clarification of predictors of the pH preference of bacterial taxa.

[0063] Taxonomy was shown to be a poor predictor of bacterial pH preferences (Fig. 2). In almost every phylum, there were ASVs with very distinct pH preferences and there were few cases where a high proportion of ASVs from a particular phylum were found to have a similar pH preference. For example, in the ROMAINE freshwater dataset, many ASVs assigned to the phylum Acidobacteria did exhibit a general preference for acidic pH conditions, but this observation was not consistent across datasets (Fig. 2). Thus phylum-level information provides relatively little insight into bacterial pH preferences, most likely because taxa within a given phylum can have divergent pH preferences [as noted previously for different acidobacterial subdivisions]. Applicants observed similar patterns at finer levels of taxonomic resolution. For example, ASVs within some of the most ubiquitous bacterial families observed across these five datasets (Xanthobacteraceae in soil and Chitinophagaceae in freshwater systems) included ASVs with inferred pH preferences ranging from 4.01 to 8.20 and from 4.63 to 8.35, respectively.

[0064] Consistent with the taxonomy-based results, Applicants also found that bacterial pH preferences were not readily predictable from phylogenetic information. In other words, there was minimal phylogenetic conservation in pH preferences. Although Applicants detected a significant phylogenetic signal in pH preferences for all but one dataset (Fig. 3 and Fig. 8), the signal was relatively weak (Pagel’s X, 0.22 to 0.78) (Fig. 8). This is qualitatively evident from the phylogenetic trees associated with each dataset, which show that even closely related taxa often had very distinct pH preferences. For example, within the Proteobacteria phylum, some clusters of ASVs with similar preference for higher or lower pH conditions were indeed observed in the phylogenies from both Australia (AUS) (soil) and ROMAINE (freshwater) datasets, but these clusters were not evident in the other datasets (Fig. 3). These observations were further confirmed by the phylogenetic correlogram analyses that show that the Moran’s I autocorrelation values were low in all cases [<0.25; (37)], even at shallow phylogenetic depth (Figs. 7B and 8). A significant Pagel’s X indicates that there is a significant phylogenetic signal of the trait by comparison to a random model of evolution, while Moran’s / indicates whether this signal is clustered around a particular phylogenetic depth based on correlation. Because these indices are differently affected by sample size and trait variation, there is no consensus on thresholds that help interpret these parameters together. A statistically significant Pagel’s X verifies that there is phylogenetic signal, and the magnitude of Moran’s I indicates how strong this signal is at a given phylogenetic depth. These results are in line with previous work that found no phylogenetic conservation of pH preferences across both cultured bacteria and uncultured bacteria. Neither taxonomic nor phylogenetic information is generally useful for determining pH preferences without additional information, highlighting the potential value of incorporating functional gene information to better predict bacterial pH preferences.

[0065] Example 3: Associations between bacterial pH preferences and functional genes.

[0066] Applicants next analyzed representative genomes of those ASVs with inferred pH preferences. Given the well-recognized biases in reference genome databases, only a relatively small fraction (6 to 24%) of those ASVs identified in the sample sets had representative genomes in Genome Taxonomy Database (GTDB) (40) and a substantial portion (-35%) of those genomes were metagenome-assembled genomes (MAGs) and single-cell-assembled genomes (SAGs) from uncultured taxa (Fig. 9). The number of genome matches obtained ranged from 57 genomes in the AUS soil dataset to 293 genomes in the Carbon Biogeochemistry in Boreal Aquatic Systems (CARBBAS) freshwater dataset (with a total across all five datasets of 669 ASV-genome matches representing 580 unique genomes with inferred pH preferences; Fig. 14and Fig. 10A). From the taxa with inferred pH preferences, the proportion of ASVs with available genome representatives was 10.6% in soils and 22.4% in freshwater on average (Fig.10). Still, the taxonomic composition of the available genomes spanned 20 different phyla, with a general predominance of Actinobacteria and Proteobacteria among the matching genomes (Fig. 10B), a result that is expected given that these two phyla are relatively well represented in genome databases.

[0067] Applicants identified 332 gene types that had the same significant association with inferred pH preference in at least two datasets, while 56 of those genes had the same association with pH in at least three datasets across soil and freshwater habitats (Fig. 4 and 14). Applicants note that the taxonomic distinctiveness of the datasets and the generally even distribution of pH preferences across phyla support the notion that the identified associations between gene types and inferred pH preferences were general and unlikely to be a product of the weak phylogenetic signal detected (Figs. 8 and 10B; see also Materials and Methods for additional information). No gene types were identified as having a significant association with pH in all datasets, which is likely a result of pH adaptations not being conserved across taxa, these genes having other functionalities besides just pH adaptation, and the habitat-specific nature of some of the observed associations. The 56 gene types identified as having shared pH associations across both soil and freshwater habitats encoded for proteins known for their involvement in pH tolerance such as adenosine triphosphatases (ATPases), anion and cation transporters and antiporters, and alkaline and acidic phosphatases (Fig. 4). Generally, bacteria need to maintain pH homeostasis to maintain enzymatic function and cytoplasmic membrane stability, and they generally use four main mechanisms to cope with acid stress. Across those genes that Applicants identified as being associated with pH preferences (Fig. 4), Applicants see evidence for all four mechanisms. First, proton-consuming reactions, notably decarboxylation and deamination of amino acids, buffer proton concentrations in the cytoplasm by incorporation of H+ to metabolic by-products. Applicants identified genes for decarboxylases (AAL decarboxy, soils), amino acid transporters (AA_permease, freshwater), carboxylate transporters (TctA, soils and freshwater), and amino acid deaminases (Queuosine synth, soils and freshwater) associated with inferred pH preferences. Second, cells will produce basic compounds, such as ammonia released from urea to counter acidity. Applicants found genes assigned to urea membrane transporters (ureide_permeases) overrepresented in taxa with low pH preference in soil and freshwater, as well as a gene for urease (UreE_C, soils) that hydrolyzes urea into ammonia. Third, bacteria can actively efflux protons to maintain intracellular pH levels. Applicants identified genes for a wide range of cation and anion efflux pumps such as the Kdp K+ membrane transporters (KdpACD) that were overrepresented in taxa with low pH preference in all habitats. In contrast, Na+ / H+ antiporters [PhaGF, MnhG, MrpF, and YufB] and anion transporters such as citrate (CitMHS, soils and freshwater) and lactate permeases (freshwater) were overrepresented in taxa with preferences for higher pH. Last, bacteria also modify the permeability of the cytoplasmic membrane and control the maturation and folding of proteins to limit acid stress. Applicants identified genes for multiple hydrogenase quality control proteins (HypCD, Hycl, and HupF) as overrepresented in taxa with low pH preference across soils and freshwater, with these genes known to be involved in acid stress response. Applicants also identified genes for process-specific proteins that act in a pH-dependent manner such as acidic phosphatases [Phosphoesterase and CpsB CapC in soils and freshwater]. Overall, Applicants found that 30 of the 56 gene types that Applicants found are consistently associated with pH across habitats have previously been linked to bacterial pH adaptations in other studies (Fig. 4 and 14).

[0068] While Applicants cannot confirm that the genes that Applicants identified are associated with pH preferences across multiple datasets actually represent specific bacterial adaptations to pH, these results make it possible to evaluate genes not previously involved in bacterial responses to pH. These results expand on previous work by identifying multiple gene functions associated with pH preferences beyond transporter genes while also confirming established associations between specific genes and bacterial pH adaptations (28). In addition, because most of the current knowledge regarding the specific genes involved in pH tolerance is derived from in vitro studies focused on bacterial pathogens, these findings provide a basis to extend the investigation of bacterial adaptations to pH more broadly beyond those selected taxa.

[0069] Example 4: Prediction of bacterial pH preference from genomic information.

[0070] Applicants incorporated the information on the presence / absence of the 56 gene types consistently associated with pH preference across habitats into a machine learning modeling framework to predict bacterial pH preferences. Preselecting gene types with consistent associations across habitats increased the likelihood that the model would maintain its predictive power on independent datasets (i.e., those containing genomes not included in this study). Across datasets, Applicants obtained an average coefficient of determination (R2) value of 0.80 for the linear regression between predicted and observed pH preferences using the training data (Fig. 5). These genes encoded very diverse functions, from well-established ones such as transmembrane anion and cation transport, ATPase, and phosphatase activity, to functions less known to be involved in bacterial pH responses such as nucleases for DNA repair, endolysins, or the type V secretion system (14). With only presence / absence information of these 56 genes, Applicants were able to predict the pH preferences of bacterial taxa that had not been used for model training. The validations conducted on a randomly selected subset of the genomes in each dataset (10% of genomes) had a mean absolute error (MAE) of 0.63 (MAE = 0.43 using the training data; Fig. 5), indicating that, with available presence / absence information of these 56 genes, this machine learning model can predict the pH preference of a given bacterial taxon with an accuracy of 0.63 pH units. Considering bacterial taxa generally have pH optima within 1 pH unit, this error is relatively small. Likewise, the average R2of the linear regression between the observed and predicted pH preferences in the independent validation set was 0.55. Note that, with this model, the absence of a gene is as informative for predictive purposes as gene presence (Fig. 12) and that, due to the lack of data for taxa with estimated pH preferences below pH 4 and above pH 9, the model can only predict accurately within that range. Applicants also note that Applicants model predicted bacterial pH preferences across all five datasets, not necessarily pH optima for growth. Observed pH preferences correspond to the pH at which taxa achieve maximal abundance in nature, reflecting both the pH optimum for growth and the biotic and abiotic factors that may constrain bacterial abundances — the realized niche. Applicants have made a detailed description of this model and how it can be applied to any genome of interest in doi.org / 10.6084 / m9.fi share.22588963 (such data being specifically incorporated herein in its entirety).

[0071] Applicants further validated the model using information on bacterial pH preferences estimated in a completely independent study of bacterial distributions in soils across the United Kingdom. The linear regression between the statistically estimated pH preference of taxa in the dataset and Applicants model predictions had R2 = 0.21 and MAE = 0.93 (Fig. 12). Note that the estimated pH preferences in that study were obtained using a different approach from Applicants study and that the pH preferences for many taxa were inferred using models with relatively weak fits, a point that the authors were careful to acknowledge. Despite the limitations of this independent dataset and the limitations associated with Applicants model, the correspondence between predicted and observed pH preferences from this independent study (and an independent set of genomes from this study; Fig. 5) supports the value of the approach and the model Applicants developed. Applicants emphasize the importance of including data from future studies that could measure actual pH preferences of bacterial isolates studied in vitro. While curated bacterial phenotypic information from culture collections can provide information on bacterial pH preferences, preexisting data on bacterial pH preferences of cultured isolates are now limited to the very narrow distribution of putative pH preferences [85.4% of pH preferences falling between 6 and 8], a pattern that most likely reflects the limited breadth of culturing conditions most commonly used in isolation efforts. Further quantification of microbial growth responses across large pH gradients under laboratory conditions coupled with whole genome sequencing can improve Applicants knowledge of the genetic underpinnings of microbial adaptations to pH.

[0072] Applicants show that biogeographical information can be combined with genomic information to infer and predict the pH preferences of bacterial taxa. This is important given the considerable effort often required to directly measure pH preferences of cultivated taxa in vitro and given that such in vitro assays are impossible for the maj ority of bacterial taxa that are resistant to cultivation. Applicants show that pH preferences can be inferred from genomic information alone, making it feasible to leverage the ever expanding genomic databases, including those which include MAGs and SAGs, to determine a key ecological attribute that cannot be readily determined from taxonomic or phylogenetic information alone. Applicants not only identified genes that had been previously linked to pH tolerance via detailed studies of select bacteria but also identified genes that warrant further investigation as they have not previously been associated with pH adaptations.

[0073] Applicants approach demonstrates the feasibility of using genomic information to make predictions for other important traits that can be difficult to infer directly. In this sense, this work is similar to previous studies that have used genomic data to predict maximum potential growth rates, oxygen tolerances, and temperature optima, among other traits. However, in those cases, data from cultivated isolates were used to develop the models linking genomic attributes to the trait values of interest. This represents an important bottleneck given that well-characterized cultivated isolates represent only a fraction of the phylogenetic and ecological diversity found in many environments. Instead, Applicants demonstrate how biogeographical information, specifically distributions of taxa across environmental gradients of interest, combined with data from representative genomes, can be used to predict environmental preferences and identify the specific adaptations that may be associated with the ecological attributes. Applicants expect that a similar approach could be used to quantify other relevant traits that have traditionally been difficult to infer directly for uncultivated taxa, including tolerances to changes in moisture, salinity, or heavy metals and other potentially toxic compounds. The machine learning model presented here has the potential to aid the rational design of microbiomes where information on the pH preferences of bacterial taxa is needed. For example, several studies with N2-fixing rhizobia have successfully improved the symbiotic benefits to legume crops via the isolation and inoculation of acid resistant Rhizobium and Bradyrhizobium strains. Forecasting invasive species spread can also benefit from genome-based predictions of pH, given the likely importance of pH as a factor limiting bacterial colonization of habitats. In addition, by predicting bacterial pH preferences, Applicants model can aid the optimization of culturing conditions for any bacterial taxon with available genomic information. The coupling of biogeographical and genomic information can be successfully used to predict the environmental preferences of bacterial taxa, presenting opportunities for improving Applicants trait-based understanding of microbial life.

[0074] Example 5: Materials and Methods.

[0075] Datasets and sequence processing

[0076] Applicants compiled five 16S rRNA gene sequencing datasets that spanned broad gradients in pH and represented distinct ecosystem types. These datasets included two previously published soil datasets [soils collected from across Panama and across Australia; (18)], two previously published freshwater datasets [samples collected from streams, rivers, and lakes in Canada; (21, 22)], and one dataset of soils collected from the Savannah River Site (SRS) in South Carolina, United States.

[0077] For the SRS dataset, Applicants collected soil samples in May 2021 from patches of savanna and surrounding forests of longleaf and loblolly pine that have been maintained since 2000 as part of the Corridor Project experiment [e.g., (57)]. The SRS is a National Environmental Research Park. It is a U.S. Department of Energy site that is managed by the U.S. Department of Agriculture Forest Service. Each soil sample consisted of a homogenate of eight 5-cm-deep soil core subsamples. Applicants extracted DNA using the DNeasy PowerSoil HTP 96 Kit (Qiagen) from well-mixed soil slurries consisting of 1 g of soil and 2 ml of deionized and autoclaved water. Applicants amplified the V4 region of the 16S rRNA gene using universal primers 515 forward and 806 reverse in duplicate polymerase chain reactions (PCRs). Applicants cleaned and normalized amplicon samples using the SequelPrep Normalization Plate Kit (Applied Biosystems, Waltham, MA, USA) and sequenced a total of 240 soil samples, 22 extraction blanks (400 pl of deionized and autoclaved water), and three PCR negative controls using paired-end Illumina MiSeq sequencing (300-cycle flow cell). For pH measurements, Applicants created soil slurries consisting of 1 g of soil and 10 ml of deionized water, vortexed the slurries at maximum speed for 20 s, and then let them rest for about 1 hour.

[0078] The five datasets were selected as each one encompassed a broad range in measured sample pH values, each included a sufficiently large number of samples across the five pH gradients (>200 samples per dataset), and pH was strongly correlated with differences in overall bacterial community composition across each dataset (Fig. 6). From all but the SRS dataset, Applicants compiled the metadata from open-source databases and from the authors upon request and downloaded the DNA sequences from the Sequence Read Archive (SRA) of the National Center for Biotechnology Information. The raw sequences of the SRS project were deposited in the SRA under bioproject ID PRJNA898410. Additional details on these datasets are provided in Fig. 6.

[0079] All datasets were analyzed using the same bioinformatic pipeline. Briefly, primers, adapters, and ambiguous bases were initially removed from the 16S rRNA gene reads using cutadapt [vl.18; (58)]. Sequences were then quality-filtered, and ASVs were inferred using the DADA2 pipeline [vl.14.1; (59)]. Chimeric sequences were also removed, and taxonomic affiliations were determined against the SILVA SSU database [release 138; (60)]. The outputs were loaded into R (61) using the phyloseq package [vl.38.0; (62)] for downstream analyses.

[0080] Statistical inference of pH preferences

[0081] Singletons (ASVs represented by only a single read within a given dataset) were removed, and samples were rarefied to the minimum read number that ensured sufficient sequencing depth based on rarefaction curves (see Fig. 14). As archaeal reads were relatively rare in most of these samples, Applicants focused Applicants analyses only on bacteria. Those bacterial ASVs that occurred in fewer than 20 samples per dataset were excluded from the analysis as ASVs needed to occur in a sufficiently large number of samples to effectively infer pH preferences (2 to 10% of ASVs passed this minimum threshold of occurrence per dataset; Fig. 14). To infer pH preferences for individual ASVs in each dataset, Applicants first generated 1000 randomized distributions of the relative abundance of each ASV across samples with replacement (i.e., each relative abundance value could be sampled more than once) and calculated the maximum value for each of these distributions. Applicants then calculated 95% confidence intervals of these relative abundance maxima using the boot package (vl.3.28) in R. The extremes of these intervals in relative abundance maxima were then matched to the pH of the samples where these ASVs achieved these relative abundance values, thus obtaining an estimated range of preferential pH for each ASV (i.e., the range of pH in which a given ASV consistently achieved maximal abundance across randomizations). All ASVs with inferred pH preferences that had ranges greater than 0.5 pH units were removed from downstream analyses as Applicants were not confident that Applicants could accurately infer a specific pH preference for those taxa. This stems from the assumption that Applicants could only obtain confident statistical inferences for taxa with narrower pH preferences, excluding taxa with broader ranges in pH preferences to yield more accurate inferences. Applicants visually inspected whether the abundance of each ASV in the nonrandomized dataset was within the identified preferred pH range to verify that Applicants pH range filter was sufficiently stringent to exclude unreliable associations between ASVs and pH preferences. The midpoint of the pH range was taken as the estimated pH preference for each ASV. Preliminary tests showed this bootstrapping approach to be more robust to zero inflation compared to alternative randomization approaches, generalized additive model, or logistic regression model fits that have previously been used for estimating environmental optima (34). Applicants note that many factors, besides just pH, can influence the distributions of bacteria, and Applicants restricted Applicants analyses just to those taxa (ASVs) for which pH preferences could effectively be determined (4568 ASVs in total, 468 to 1614 ASVs per dataset; Fig. 14).

[0082] Genome search and annotation

[0083] The ASVs with estimated pH preferences were matched to the annotated GTDB (release 207; (40)], using vsearch [v2.21.1; (63)] to identify representative genomes. Applicants acknowledge that the representative genome identified for any given ASV will not necessarily be an identical match to the genome of that taxon in situ as even bacterial taxa with identical 16S rRNA genes can have distinct genomes (64). However, this limitation poses a challenge to nearly all genomic analyses of bacteria from environmental samples, as even collections of MAGs would not necessarily capture the genomic heterogeneity that could exist at finer levels of taxonomic resolution. Applicants also note that, when selecting representative genomes, Applicants only allowed a single-base pair mismatch between the genomes and the 16S rRNA gene amplicons, corresponding to a conservative 99.6% sequence similarity for the 250-base pair amplicons. In situations where a single ASV matched multiple genomes with the same similarity, Applicants selected the most complete genome. Predicted coding sequences for the -62,000 unique bacterial genomes (“species clusters”) available in the GTDB representative set were identified using Prodigal [v2.6.3; (65)]. The predicted coding sequences for each genome were aligned to the Pfam database [v35.0; (66)] using HMMER [v3; (67)] to obtain annotation of potential domains, genes, and gene families in each coding sequence. Applicants discarded matches with a bit score lower than 10. This pipeline yielded a list of putative genes and domains found in each of the GTDB genomes that matched the ASVs identified from the samples included in this study with estimated pH preferences (580 genomes in total, 57 to 293 genomes per dataset; Fig. 14). The copy numbers of genes and domains were binarized to presence / absence for further analyses.

[0084] Identification of genes associated with pH preference

[0085] Applicants next determined associations between the estimated pH preferences of ASVs and the annotated genes in the corresponding representative genomes. Applicants identified genes associated with pH preference by fitting generalized linear models with a binomial distribution and using the logit as a link function with core R functions. For each binarized gene function, Applicants fitted a single model with the presence / absence of that gene type as the response variable and the estimated pH preference of each genome as the independent variable. Applicants evaluated the statistical significance of the model coefficients using the Wald test as implemented with R core functions. Applicants obtained the slope (positive or negative) of the relationship between pH preference and the presence / absence of each gene from the model estimates. Applicants verified that the model estimates reliably reflected whether the relationship was positive or negative by plotting the proportion of genomes containing a particular gene type against the estimated pH preferences of those genomes. Last, Applicants filtered out genes that had nonsignificant associations to pH preference in all datasets as well as those gene types that had significant associations in only one of the five datasets (as Applicants goal was to identify consistent associations between genes and pH preferences across multiple datasets). Applicants thus considered those gene types with the same significant association (P < 0.05 and same positive or negative direction of the model estimate) that occurred in two or more datasets to be associated with bacterial pH preferences. Applicants verified that phylogenetic signal of pH preference among the representative genomes was very weak (in two datasets, Pearson’s r - 0.06 to 0.07) or nonexistent (in three datasets) to justify not penalizing the generalized linear model estimates with phylogenetic information (Fig. 13). Phylogenetic visualization and analysis

[0086] Applicants generated phylogenetic trees for each dataset that included only those ASVs for which pH preferences could be inferred. For each of those ASVs, Applicants selected the corresponding highest quality full-length 16S rRNA gene sequence from the SILVA SSU database [release 138; (60)], aligned those sequences with MUSCLE [v5; (68)], and created maximum likelihood (ML) trees using the RAxML approach with standard parameters [v8; (69)]. Applicants visualized and edited the trees using iTOL [v6; (70)]. Applicants additionally tested whether pH preference had a phylogenetic signal by calculating Blomberg’s K (71) and Pagel’s X (72) with the depth at which pH preference had a phylogenetic signal estimated using phylogenetic correlograms (73). These three types of phylogenetic analyses were conducted using the R package phylosignal [vl .3; (74)] using ML trees constructed as described above with all ASVs from a given dataset for which pH preference could be inferred (Fig. 14).

[0087] Prediction and independent validation of pH preferences from genomic information

[0088] Applicants used the full set of genes associated with pH preference across habitats to train a gradient boosted decision tree model to predict bacterial pH preference from genomic information. The gene type presence / absence table was imported into Python, and pH preferences were predicted for all ASVs from a one-hot encoded gene matrix using gradient boosted decision trees created with the XGB Regressor function from Python’s xgboost package [vl.6.2; (75)]. Hyperparameter optimization was implemented with the hyperopt package [vO.2.7; (76)] to select the best hyperparameters for the XGBRegressor model. Applicants validated the accuracy of the model by measuring its MAE on an independent 10% of the genomes in each data set (i.e., the test set). Applicants estimated the impact of each gene type on model predictions using Shapley additive explanations as integrated in Python’s xgboost package. Applicants also tested the model using a dataset of bacterial taxa with estimated pH preferences from soils across the United Kingdom (38), from which Applicants selected taxa with unimodal relationships of abundance with pH and therefore more confident estimation of pH preference [models IV and V in (38)]. For each of these taxa, Applicants obtained reference genomes using the same approach described for the ASVs in the other datasets and recorded the presence / absence of the predictive genes in those genomes to run the model. REFERENCES

[0089] 1. M. Delgado-Baquerizo, A. M. Oliverio, T. E. Brewer, A. Benavent-Gonzalez, D. J. Eldridge, R. D. Bardgett, F. T. Maestre, B. K. Singh, N. Fierer, A global atlas of the dominant bacteria found in soil. Science 359, 320-325 (2018).

[0090] 2. D. C. Laughlin, Applying trait-based models to achieve functional targets for theory- driven ecological restoration. Ecol. Lett. 17, 771-784 (2014).

[0091] 3. M. G. Mathipa, M. S. Thantsha, Probiotic engineering: Towards development of robust probiotic strains with enhanced functional properties and for targeted control of enteric pathogens. Gut Pathog. 9, 1-17 (2017).

[0092] 4. S. J. Green, C. B. Brookson, N. A. Hardy, L. B. Crowder, Trait-based approaches to global change ecology: Moving from description to prediction. Proc. Biol. Sci. 289, 20220071 (2022).

[0093] 5. J. S. Madin, D. A. Nielsen, M. Brbic, R. Corkrey, D. Danko, K. Edwards, M. K. M. Engqvist, N. Fierer, J. L. Geoghegan, M. Gillings, N. C. Kyrpides, E. Litchman, C. E. Mason, L. Moore, S. L. Nielsen, I. T. Paulsen, N. D. Price, T. B. K. Reddy, M. A. Richards, E. P. C. Rocha, T. M. Schmidt, H. Shaaban, M. Shukla, F. Supek, S. G. Tetu, S. Vieira-Silva, A. R. Wattam, D. A. Westfall, M. Westoby, A synthesis of bacterial and archaeal phenotypic trait data. Sci. Data 7, 170 (2020).

[0094] 6. A. Cebron, E. Zeghal, P. Usseglio-Polatera, A. Meyer, P. Bauda, F. Lemmel, C. Leyval, F. Maunoury -Danger, BactoTraits - A functional trait database to evaluate how natural and man- induced changes influence the assembly of bacterial communities. Ecol. Indic. 130, 108047 (2021).

[0095] 7. A. D. Steen, A. Crits-Christoph, P. Carini, K. M. DeAngelis, N. Fierer, K. G. Lloyd, J. Cameron Thrash, High proportions of bacteria and archaea across most biomes remain uncultured. ISME J. 13, 3126-3130 (2019).

[0096] 8. E. Baath, E. Kritzberg, pH tolerance in freshwater bacterioplankton: Trait variation of the community as measured by leucine incorporation. Appl. Environ. Microbiol. 81, 7411-7419 (2015).

[0097] 9. K. K. Treseder, S. N. Kivlin, C. V. Hawkes, Evolutionary trade-offs among decomposers determine responses to nitrogen enrichment. Ecol. Lett. 14, 933-938 (2011). 10. N. Lustenhouwer, D. S. Maynard, M. A. Bradford, D. L. Lindner, B. Oberle, A. E. Zanne, T. W. Crowther, A trait-based understanding of wood decomposition by fungi. Proc. Natl. Acad. Sci. U.S.A. 117, 11551-11558 (2020).

[0098] 11. J. T. Lennon, Z. T. Aanderud, B. K. Lehmkuhl, D. R. Schoolmaster Jr., Mapping the niche space of soil microorganisms using taxonomy and traits. Ecology 93, 1867-1879 (2012).

[0099] 12. D. B. Sauer, D.-N. Wang, Predicting the optimal growth temperatures of prokaryotes using only genome derived features. Bioinformatics 35, 3224-3231 (2019).

[0100] 13. P.-A. Chaumeil, A. J. Mussig, P. Hugenholtz, D. H. Parks, GTDB-Tk: A toolkit to classify genomes with the Genome Taxonomy Database. Bioinformatics 36, 1925-1927 (2020).

[0101] 14. A. Barberan, H. Caceres Velazquez, S. Jones, N. Fierer, Hiding in plain sight: Mining bacterial species records for phenotypic trait information. mSphere 2, e00237-17 (2017).

[0102] 15. J. Jablohska, D. S. Tawfik, The number and type of oxygen-utilizing enzymes indicates aerobic vs. anaerobic phenotype. Free Radic. Biol. Med. 140, 84-92 (2019).

[0103] 16. A. O. C. Dupont, R. I. Griffiths, T. Bell, D. Bass, Differences in soil micro-eukaryotic communities over soil pH gradients are strongly driven by parasites and saprotrophs. Environ. Microbiol. 18, 2010-2024 (2016).

[0104] 17. J. Davison, M. Moora, M. Semchenko, S. B. Adenan, T. Ahmed, A. A. Akhmetzhanova, J. M. Alatalo, S. Al-Quraishy, E. Andriyanova, S. Anslan, M. Bahram, A. Batbaatar, C. Brown, C. G. Bueno, J. Cahill, J. J. Cantero, B. B. Casper, M. Cherosov, S. Chideh, A. P. Coelho, M. Coghill, G. Decocq, S. Dudov, E. C. Fabiano, V. E. Fedosov, L. Fraser, S. I. Glassman, A. Helm, H. A. L. Henry, B. Herault, I. Hiiesalu, I. Hiiesalu, W. N. Hozzein, P. Kohout, U. Koljalg, K. Koorem, L. Laanisto, U. Mander, L. Mucina, J.-P. Munyampundu, L. Neuenkamp, U. Niinemets, C. Nyamukondiwa, J. Oja, V. Onipchenko, M. Partel, C. Phosri, S. Polme, K. Piissa, A. Ronk, A. Saitta, O. Semboli, S.-K. Sepp, A. Seregin, S. Sudheer, C. P. Pena- Venegas, C. Paz, T. Vahter, M. Vasar, A. J. Veraart, L. Tedersoo, M. Zobel, M. Opik, Temperature and pH define the realised niche space of arbuscular mycorrhizal fungi. New Phytol. 231, 763-776 (2021).

[0105] 18. A. M. Oliverio, A. Bissett, K. McGuire, K. Saltonstall, B. L. Turner, N. Fierer, The role of phosphorus limitation in shaping soil bacterial communities and their metabolic capabilities. mBio 11, e01718-e01720 (2020).

[0106] 19. N. Fierer, R. B. Jackson, The diversity and biogeography of soil bacterial communities. Proc. Natl. Acad. Sci. U.S.A. 103, 626-631 (2006). 20. R. Ortiz-Alvarez, J. Caliz, L. Camarero, E. O. Casamayor, Regional community assembly drivers and microbial environmental sources shaping bacterioplankton in an alpine lacustrine district (Pyrenees, Spain). Environ. Microbiol. 22, 297-309 (2020).

[0107] 21. J. P. Nino-Garcia, C. Ruiz-Gonzalez, P. A. Del Giorgio, Interactions between hydrology and water chemistry shape bacterioplankton biogeography across boreal freshwater networks. ISME J. 10, 1755-1766 (2016).

[0108] 22. M. Stadler, P. A. del Giorgio, Terrestrial connectivity, upstream aquatic history and seasonality shape bacterial community assembly within a large boreal aquatic network. ISME J. 16, 937-947 (2022).

[0109] 23. J. F. Power, C. R. Carere, C. K. Lee, G. L. J. Wakerley, D. W. Evans, M. Button, D. White, M. D. Climo, A. M. Hinze, X. C. Morgan, I. R. McDonald, S. C. Cary, M. B. Stott, Microbial biogeography of 925 geothermal springs in New Zealand. Nat. Commun. 9, 2876 (2018).

[0110] 24. Q. Jin, M. F. Kirk, pH as a primary control in environmental microbiology: 1. Thermodynamic perspective. Front. Environ. Sci. 6, 21 (2018).

[0111] 25. M. A. Oberhardt, R. Zarecki, S. Gronow, E. Lang, H.-P. Klenk, U. Gophna, E. Ruppin, Harnessing the landscape of microbial culture media to predict new organism-media pairings. Nat. Commun. 6, 8493 (2015).

[0112] 26. W. H. Hartman, C. J. Richardson, R. Vilgalys, G. L. Bruland, Environmental and anthropogenic controls over bacterial communities in wetland soils. Proc. Natl. Acad. Sci. U.S.A. 105, 17842-17847 (2008).

[0113] 27. C. L. Lauber, M. Hamady, R. Knight, N. Fierer, Pyrosequencing-based assessment of soil pH as a predictor of soil bacterial community structure at the continental scale. Appl. Environ. Microbiol. 75, 5111-5120 (2009).

[0114] 28. A. A. Malik, B. C. Thomson, A. S. Whiteley, M. Bailey, R. I. Griffiths, Bacterial physiological adaptations to contrasting edaphic conditions identified using landscape scale metagenomics. mBio 8, 799-816 (2017).

[0115] 29. R. T. Jones, M. S. Robeson, C. L. Lauber, M. Hamady, R. Knight, N. Fierer, A comprehensive survey of soil acidobacterial diversity using pyrosequencing and clone library analyses. ISME J. 3, 442-453 (2009). 30. J. Rousk, E. Baath, P. C. Brookes, C. L. Lauber, C. Lozupone, J. G. Caporaso, R. Knight, N. Fierer, Soil bacterial and fungal communities across a pH gradient in an arable soil. ISME J. 4, 1340-1351 (2010).

[0116] 31. N. Guan, L. Liu, Microbial response to acid stress: Mechanisms and applications. Appl. Microbiol. Biotechnol. 104, 51-65 (2020).

[0117] 32. Y. Derman, H. Soderholm, M. Lindstrom, H. Korkeala, Role of csp genes in NaCl, pH, and ethanol stress response and motility in Clostridium botulinum ATCC 3502. Food Microbiol. 46, 463-470 (2015).

[0118] 33. M. Ito, M. Morino, T. A. Krulwich, Mrp antiporters have important roles in diverse bacteria and archaea. Front. Microbiol. 8, 2325 (2017).

[0119] 34. J. C. Benavides, D. H. Vitt, Response curves and the environmental limits for peatforming species in the northern Andes. Plant Ecol. 215, 937-952 (2014).

[0120] 35. N. Fierer, J. L. Morse, S. T. Berthrong, E. S. Bernhardt, R. B. Jackson, Environmental controls on the landscape-scale biogeography of stream bacterial communities. Ecology 88, 2162- 2173 (2007).

[0121] 36. A. M. Kielak, C. C. Barreto, G. A. Kowalchuk, J. A. van Veen, E. E. Kuramae, The ecology of Acidobacteria: Moving beyond genes and genomes. Front. Microbiol. 7, 744 (2016).

[0122] 37. J. A. F. Diniz-Filho, T. Santos, T. F. Rangel, L. M. Bini, A comparison of metrics for estimating phylogenetic signal under alternative evolutionary models. Genet. Mol. Biol. 35, 673- 679 (2012).

[0123] 38. B. Jones, T. Goodall, P. B. L. George, H. S. Gweon, J. Puissant, D. S. Read, B. A. Emmett, D. A. Robinson, D. L. Jones, R. I. Griffiths, Beyond taxonomic identification: Integration of ecological responses to a soil bacterial 16S rRNA gene database. Front. Microbiol. 12, 682886 (2021).

[0124] 39. S. Mukheijee, R. Seshadri, N. J. Varghese, E. A. Eloe-Fadrosh, J. P. Meier-Kolthoff, M. Gbker, R. C. Coates, M. Hadjithomas, G. A. Pavlopoulos, D. Paez-Espino, Y. Yoshikuni, A. Visel, W. B. Whitman, G. M. Garrity, J. A. Eisen, P. Hugenholtz, A. Pati, N. N. Ivanova, T. Woyke, H.- P. Klenk, N. C. Kyrpides, 1,003 reference genomes of bacterial and archaeal isolates expand coverage of the tree of life. Nat. Biotechnol. 35, 676-683 (2017).

[0125] 40. D. H. Parks, M. Chuvochina, C. Rinke, A. J. Mussig, P.-A. Chaumeil, P. Hugenholtz, GTDB: An ongoing census of bacterial and archaeal diversity through a phylogenetically consistent, rank normalized and complete genome-based taxonomy. Nucleic Acids Res. 50, D785- D794 (2022).

[0126] 41. P. A. Lund, D. De Biase, O. Liran, O. Scheier, N. P. Mira, Z. Cetecioglu, E. N. Fernandez, S. Bover-Cid, R. Hall, M. Sauer, C. O’Byrne, Understanding how microorganisms respond to acid pH is central to their control and successful exploitation. Front. Microbiol. 11, 556140 (2020).

[0127] 42. P. D. Cotter, C. Hill, Surviving the acid test: Responses of gram-positive bacteria to low pH. Microbiol. Mol. Biol. Rev. 67, 429-453 (2003).

[0128] 43. T. A. Krulwich, G. Sachs, E. Padan, Molecular aspects of bacterial pH sensing and homeostasis. Nat. Rev. Microbiol. 9, 330-343 (2011).

[0129] 44. H. A. Danhof, S. Vylkova, E. M. Vesely, A. E. Ford, M. Gonzalez-Garay, M. C. Lorenz, Robust extracellular pH modulation by Candida albicans during growth in carboxylic acids. mBio 7, e01646-16 (2016).

[0130] 45. Y. Wen, E. A. Marcus, U. Matrubutham, M. A. Gleeson, D. R. Scott, G. Sachs, Acid- adaptive genes of Helicobacter pylori. Infect. Immun. 71, 5921-5939 (2003).

[0131] 46. E. T. Hayes, J. C. Wilks, P. Sanfilippo, E. Yohannes, D. P. Tate, B. D. Jones, M. D. Radmacher, S. S. BonDurant, J. L. Slonczewski, Oxygen limitation modulates pH regulation of catabolism and hydrogenases, multidrug transporters, and envelope composition in Escherichia coli K-12. BMC Microbiol. 6, 89 (2006).

[0132] 47. P. Lund, A. Tramonti, D. De Biase, Coping with low pH: Molecular strategies in neutralophilic bacteria. FEMS Microbiol. Rev. 38, 1091-1125 (2014).

[0133] 48. J. B. Russell, D. B. Dombrowski, Effect of pH on the efficiency of growth by pure cultures of rumen bacteria in continuous culture. Appl. Environ. Microbiol. 39, 604-610 (1980).

[0134] 49. J. J. Therion, A. Kistner, J. H. Kornelius, Effect of pH on growth rates of rumen amylolytic and lactilytic bacteria. Appl. Environ. Microbiol. 44, 428-434 (1982).

[0135] 50. S. Vieira-Silva, E. P. C. Rocha, The systemic imprint of growth and its uses in ecological (meta)genomics. PLOS Genet. 6, el000808 (2010).

[0136] 51. J. L. Weissman, S. Hou, J. A. Fuhrman, Estimating maximal microbial growth rates from cultures, metagenomes, and single cells via codon usage patterns. Proc. Natl. Acad. Sci. U.S.A. 118, e2016810118 (2021). 52. A. Chinnaswamy, B. Dhar, Symbiotic effectiveness of acid tolerant Bradyrhizobium strains with soybean in low pH soil. African J. Biotechnol. 5, 842-845 (2006).

[0137] 53. D. P. Oliveira, M. A. de Figueiredo, B. L. Soares, O. H. Stivanin Teixeira, F. A. Dias Martins, M. Rufini, C. Peixoto Chain, R. Pereira Reis, A. Ramalho de Morais, F. M. de Souza Moreira, M. J. B. de Andrade, Acid tolerant Rhizobium strains contribute to increasing the yield and profitability of common bean in tropical soils. J. Soil Sci. Plant Nutr. 17, 922-933 (2017).

[0138] 54. J. F. Briand, C. Leboulanger, J. F. Humbert, C. Bernard, P. Dufour, Cylindrospermopsis raciborskii (cyanobacteria) invasion at mid-latitudes: Selection, wide physiological tolerance, orglob al warming? J. Phycol. 40, 231-238 (2004).

[0139] 55. S. Pertola, H. Kuosa, R. Olsonen, Is the invasion of Prorocentrum minimum (Dinophyceae) related to the nitrogen enrichment of the Baltic Sea? Harmful Algae 4, 481-492 (2005).

[0140] 56. P. C. Reid, D. G. Johns, M. Edwards, M. I. Starr, M. Poulin, P. Snoeijs, A biological consequence of reducing Arctic ice cover: Arrival of the Pacific diatom Neodenticula seminae in the North Atlantic for the first time in 800000 years. Glob. Chang. Biol. 13, 1910-1921 (2007).

[0141] 57. J. J. Tewksbury, D. J. Levey, N. M. Haddad, S. Sargent, J. L. Orrock, A.Weldon, B. J. Danielson, J. Brinkerhoff, E. I. Damschen, P. Townsend, Corridors affect plants, animals, and their interactions in fragmented landscapes. Proc. Natl. Acad. Sci. U.S.A. 99, 12923-12926 (2002).

[0142] 58. M. Martin, Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet.journal 17, 10-12 (2011).

[0143] 59. B. J. Callahan, P. J. McMurdie, S. P. Holmes, Exact sequence variants should replace operational taxonomic units in marker-gene data analysis. ISME J. 11, 2639-2643 (2017).

[0144] 60. C. Quast, E. Pruesse, P. Yilmaz, J. Gerken, T. Schweer, P. Yarza, J. Peplies, F. O. Glbckner, The SILVA ribosomal RNA gene database project: Improved data processing and webbased tools. Nucleic Acids Res. 41, D590-D596 (2013).

[0145] 61. R: The R Project for Statistical Computing; r-project.org / .

[0146] 62. P. J. McMurdie, S. Holmes, phyloseq: An R package for reproducible interactive analysis and graphics of microbiome census data. PLOS ONE 8, e61217 (2013).

[0147] 63. T. Rognes, T. Flouri, B. Nichols, C. Quince, F. Mahe, VSEARCH: Aversatile open source tool for metagenomics. PeerJ. 4, e2584 (2016). 64. H. B. Hassler, B. Probert, C. Moore, E. Lawson, R. W. Jackson, B. T. Russell, V. P. Richards, Phylogenies of the 16S rRNA gene and its hypervariable regions lack concordance with core genome phylogenies. Microbiome 10, 104 (2022).

[0148] 65. D. Hyatt, G. L. Chen, P. F. LoCascio, M. L. Land, F. W. Larimer, L. J. Hauser, Prodigal: Prokaryotic gene recognition and translation initiation site identification. BMC Bioinformatics 11, 119 (2010).

[0149] 66. J. Mistry, S. Chuguransky, L. Williams, M. Qureshi, G. A. Salazar, E. L. L. Sonnhammer, S. C. E. Tosatto, L. Paladin, S. Raj, L. J. Richardson, R. D. Finn, A. Bateman, Pfam: The protein families database in 2021. Nucleic Acids Res. 49, D412-D419 (2021).

[0150] 67. R. D. Finn, J. Clements, S. R. Eddy, HMMER web server: Interactive sequence similarity searching. Nucleic Acids Res. 39, W29-W37 (2011).

[0151] 68. R. C. Edgar, MUSCLE: A multiple sequence alignment method with reduced time and space complexity. BMC Bioinformatics 5, 113 (2004).

[0152] 69. A. Stamatakis, RAxML version 8: A tool for phylogenetic analysis and post-analysis of large phylogenies. Bioinformatics 30, 1312-1313 (2014).

[0153] 70. 1. Letunic, P. Bork, Interactive Tree Of Life (iTOL) v5: An online tool for phylogenetic tree display and annotation. Nucleic Acids Res. 49, W293-W296 (2021).

[0154] 71. S. P. Blomberg, T. Garland Jr., A. R. Ives, Testing for phylogenetic signal in comparative data: Behavioral traits are more labile. Evolution 57, 717-745 (2003).

[0155] 72. M. Pagel, Inferring the historical patterns of biological evolution. Nature 401, 877-884 (1999).

[0156] 73. J. L. Gittleman, M. Kot, Adaptation: Statistics and a null model for estimating phylogenetic effects. Syst. Biol. 39, 227-241 (1990).

[0157] 74. F. Keck, F. Rimet, A. Bouchez, A. Franc, phylosignal: An R package to measure, test, and explore the phylogenetic signal. Ecol. Evol. 6, 2774-2780 (2016).

[0158] 75. T. Chen, C. Guestrin, XGBoost: A scalable tree boosting system, in Proceedings of the 22nd

[0159] ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (Association for Computing Machinery, 2016), 785-794.

[0160] 76. J. Bergstra, D. Yamins, D. Cox, Making a science of model search: Hyperparameter optimization in hundreds of dimensions for vision architectures. 28, 115-123 (2013). 77. J. Snider, T. Gutsche, M. Lin, S. Baby, B. Cox, G. Butland, J. Greenblatt, A. Emili, W. A. Houry, Formation of a distinctive complex between the inducible bacterial lysine decarboxylase and a novel AAA+ ATPase. J. Biol. Chem. 281, 1532-1546 (2006).

[0161] 78. A. Alam, J. E. Brbms, R. Kumar, A. Sjostedt, The role of ClpB in bacterial stress responses and virulence. Front. Mol. Biosci. 8, 668910 (2021).

[0162] 79. J. J. Lensbouer, R. P. Doyle, Secondary transport of metal-citrate complexes: The CitMHS family. Crit. Rev. Biochem. Mol. Biol. 45, 453-462 (2010).

[0163] 80. Z.-C. Yuan, P. Liu, P. Saenkham, K. Kerr, E.W. Nester, Transcriptome profding and functional analysis of Agrobacterium tumefaciens reveals a general conserved response to acidic conditions (pH 5.5) and a complex acid-mediated signaling involved in Agrobacteriumplant interactions. J. Bacteriol. 190, 494-507 (2008).

[0164] 81. E. Khalikova, P. Susi, T. Korpela, Microbial dextran-hydrolyzing enzymes: Fundamentals and applications. Microbiol. Mol. Biol. Rev. 69, 306-325 (2005).

[0165] 82. W.-J. Jung, J.-H. Kuk, K.-Y. Kim, K.-C. Jung, R.-D. Park, Purification and characterization of exo-P-d-glucosaminidase from Aspergillus fumigatus S-26. Protein Expr. Purif. 45, 125-131 (2006).

[0166] 83. M. C. Cholo, M. T. Matjokotja, A. G. Osman, R. Anderson, Role of the kdpDE regulatory operon of Mycobacterium tuberculosis in modulating bacterial growth in vitro . Front. Genet. 12, 698875 (2021).

[0167] 84. A. Ballal, B. Basu, S. K. Apte, The Kdp-ATPase system and its regulation. J. Biosci. 32, 559-568 (2007).

[0168] 85. H. Asha, J. Gowrishankar, Regulation of kdp operon expression in Escherichia coli: Evidence against turgor as signal for transcriptional control. J. Bacteriol. 175, 4528-4537 (1993).

[0169] 86. T. Hiramatsu, K. Kodama, T. Kuroda, T. Mizushima, T. Tsuchiya, A putative multisubunit Na+ / H+ antiporter from Staphylococcus aureus. J. Bacteriol. 180, 6642-6648 (1998).

[0170] 87. P. Putnoky, A. Kereszt, T. Nakamura, G. Endre, E. Grosskopf, P. Kiss, A. Kondorosi, The pha gene cluster of Rhizobium meliloti involved in pH adaptation and symbiosis encodes a novel type of K+ efflux system. Mol. Microbiol. 28, 1091-1101 (1998).

[0171] 88. M. Ito, A. A. Guffanti, B. Oudega, T. A. Krulwich, Mrp, a multigene, multifunctional locus in Bacillus subtilis with roles in resistance to cholate and to Na+ and in pH homeostasis. J. Bacteriol. 181, 2394-2402 (1999). 89. J. M. Chirgwin, T. F. Parsons, E. A. NoItmann, Mechanistic implications of the pH independence of inhibition of phosphoglucose isomerase by neutral sugar phosphates. J. Biol. Chem. 250, 7277-7279 (1975).

[0172] 90. N. S. DeSilva, P. A. Quinn, Characterization of phospholipase Al, A2, C activity in Ureaplasma urealyticum membranes. Mol. Cell. Biochem. 201, 159-167 (1999).

[0173] 91. M. B. Marques, P. F. Weller, A. Nicholson-Weller, Growth in acidic media increases production of phosphatidylinositol-specific phospholipase C by Staphylococcus aureus. Curr. Microbiol. 25, 125-128 (1992).

[0174] 92. M. Comellas-Bigler, K. Maskos, R. Huber, H. Oyama, K. Oda, W. Bode, 1.2 A crystal structure of the serine carboxyl proteinase pro-kumamolisin; structure of an intact pro-subtilase.

[0175] Structure 12, 1313-1323 (2004).

[0176] 93. M. Caza, J. W. Kronstad, Shared and distinct mechanisms of iron acquisition by bacterial and fungal pathogens of humans. Front. Cell. Infect. Microbiol. 3, 80 (2013).

Claims

CLAIMSWhat is claimed is1. A method of identifying a growth condition of a microorganism, the method comprising:- annotating the genomes of a first training sample of microorganisms to identify genes and / or proteins that are associated with a known target growth condition;- training a neural network using a training dataset comprising the annotated genomes of the first sample to infer a target growth condition;- establishing a reference sample of microorganisms wherein the target growth condition is unknown;- annotating the genomes of the reference sample of microorganisms to identify genes and / or proteins that are associated with a target growth condition;- predicting a growth condition of one or more microorganisms based on the presence of absence of the one or more genes and / or proteins associated with a target growth condition; and- optionally validating the predicted growth condition on one or more of the microorganisms from the reference sample.

2. The method of claim 1, wherein said plurality of microorganisms comprises a plurality of bacteria.

3. The method of claim 1, wherein said plurality of bacteria comprise a plurality of bacteria having different target growth conditions.

4. The method of claim 1, wherein said plurality of bacteria are identified via the sequence of their 16S ribosomal RNA (rRNA).

5. The method of claim 1, wherein said target growth condition is selected from pH, culture media, nutrients, temperature, osmotic potential, water, oxygen, or a combination of the same.

6. The method of claim 1, the target growth condition is pH.

7. The method of claim 6, wherein the annotated genes and / or proteins include one or more genes and / or proteins are selected from Figure 15.

8. The method of claim 1, wherein the reference sample of microorganisms comprises one or more previously uncultured microorganisms.

9. The method of claim 1, wherein said step for validating comprises the step of culturing one or more or the microorganisms under the predicted growth condition.

10. The method of claim 1, wherein said predicted growth condition comprises an optimal growth condition.

11. The system identifying a growth condition of a microorganism, the method comprising- a first training sample of microorganisms, wherein the genomes of the microorganisms of the training sample are annotates to identify genes and / or proteins that are associated with a known target growth condition;- a neural network configured to generate a training dataset comprising the annotated genomes of the first training sample to infer a target growth condition;- a reference sample of microorganisms having an unknown target growth condition, wherein the genomes of the reference sample are annotated to identify the presence of absence of the genes and / or proteins that are associated with a target growth condition;- one or more hardware processors configured by machine-readable instructions to generate a predicted growth condition of the one or more microorganisms of the reference sample based on the presence or absence of the one or more genes and / or proteins associated with a target growth condition.

12. The system of claim 11, wherein said plurality of microorganisms comprises a plurality of bacteria.

13. The system of claim 11, wherein said plurality of bacteria comprise a plurality of bacteria having different target growth conditions.

14. The system of claim 13, wherein said plurality of bacteria are identified via the sequence of their 16S ribosomal RNA (rRNA).

15. The system of claim 11, wherein said target growth condition is selected from pH, culture media, nutrients, temperature, osmotic potential, water, oxygen, or a combination of the same.

16. The system of claim 11, the target growth condition is pH.

17. The system of claim 16, wherein the annotated genes and / or proteins include one or more genes and / or proteins are selected from Figure 15.

18. The system of claim 11, wherein the reference sample of microorganisms comprises one or more previously uncultured microorganisms.

19. The system of claim 11, further comprising one or more or the microorganisms from the reference sample cultured under the predicted growth condition.

20. The system of claim 1, wherein said predicted growth condition comprises an optimal growth condition.

Images