Getting back to the nature of the microbial world: from the description and inductive reasoning to deductive study after 'meta-omics'.

'Omics' studies have by now deposited massive amounts of data into the databases, and it is now time to return to the question as to what can we actually learn from them. Increased application of the deductive approach in synthetic microbial ecology and synthetic microbiome research will undoubtedly provide exciting new opportunities for advancing our understanding of microbial ecology.

On 25th September 2015, after near 1‐year revision–rejection–revision cycles, my first Molecular Microbiology paper was finally accepted for publication. I was so proud of this paper, as I thought it was the best paper of my research career. However, if the impact factor (IF) is used to estimate the quality of my research, I wondered why other omics articles written by my group could be published in journals of much higher IF, even though they were not the same calibre as my last MM paper. Worse even, the IF of Molecular Microbiology keeps declining, for reasons only known to those who invented this neat and devilish ‘performance indicator’.

Impact factors may reflect the quality of the journal, but only within a particular professional field. For comparison in a broader field, it reflects the hotspot of the academic activities, which may not only be related to the science itself, but also to how researchers behave. The fact that a lot of researchers are following the research hotspots without real scientific questions is exactly the same as my own ill‐gotten behaviour when perpetually chasing hot stocks, only to lose my money again and again.

‘Meta‐omics’ undoubtedly represent a major hotspot in the field of microbial ecology. Since the development of next‐generation sequencing (NGS) at the beginning of this century (Metzker 2010), the IFs of journals publishing ‘meta‐omics’ works have been increasing. More and more microbiologists and microbial ecologist changed their focus to ‘meta‐omics’‐related fields based on NGS technologies. One important reason is the declining cost of sequencing and the mass of data generated/obtained. Sequencing a metagenome is now cheaper (about $7/Gb data) than the expression and purification of a protein, due to its higher labour cost (Goodwin et al., 2016).

Based on metagenomic technologies, we can describe in great detail the microbial compositions and their functional genes in any type of environment we are interested in. The scientists of microbiology and microbial ecology have investigated the microbiomes in the ocean (Sunagawa et al., 2015), in the soil (Delgado‐Baquerizo et al., 2018), in lakes (Zorzet al., 2019), in the air (Adams et al., 2013), inside buildings (Lax et al., 2017) and we even know about the microbial communities living in space stations (Sielaff et al., 2019). Besides, using sequencing technologies such as DNA/RNA stable‐isotope probing and sequencing (Chen and Murrell, 2010), and meta‐transcriptome sequencing (Urich et al., 2008), we can now bridge the gap between function and taxonomy information. The emergent bioinformatic and mathematic tools also promote the study of ‘meta‐omics’. The binning tools enable us to construct the genomes of single strains from a stack of chaotic sequence data (Sangwan et al., 2016). We can predict the microbial interactions using network analysis (Layeghifard et al., 2017) and speculate on the assembly and succession process of complex microbial communities using inductive reasoning (Zhou and Ning, 2017). Although most of the studies investigating microbial communities are descriptive, we can still learn much, for example what the biogeographic distribution of microorganisms is on earth (Thompson et al., 2017), or how microorganisms are related to soil productivity (Fierer 2017), how gut microbiota affect human health (Shreiner et al., 2015) and what roles microorganisms in the biogeochemical cycles of earth play (Falkowski et al., 2008), apart from discovering more and more unculturable microorganisms hitherto unknown to humanity (Parks et al., 2017).

‘Meta‐omics’‐related research undoubtedly improves our understanding of the microbial world. However, because most of the ‘meta‐omics’ studies use descriptive methods and inductive reasoning, what exactly do these studies achieve? Whether these hypotheses or so‐called theories can be tested and applied in other environments remains unknown. Also, whether the scientific questions in the present ‘meta‐omics’ studies are real scientific questions or merely technical problems, and whether they are meaningful, remains in doubt and thus needs to be discussed. Generally, the answer to a real scientific question can explain a phenomenon, and the scientific hypothesis is a proposed explanation for the phenomenon, an explanation that is testable using scientific methods.

What else can we learn from ‘meta‐omics’ and NGS? Imagine a time when the cost of sequencing becomes negligibly small, when automatic tools for bioinformatic and mathematical analysis are developed (online automatic platforms are now emerging, allowing users without prior experience to upload their sequencing raw data and perform popular bioinformatic and embedded mathematical analysis), ‘meta‐omics’ analysis based on NGS would become common technologies like PCR used in every biological laboratory today. Critically, once artificial intelligence is mature enough, will it be able to assist, or even replace us in the analysis of patterns emerging from ‘meta‐omics’ data? Or even develop hypotheses that explain those patterns?

To find the answer to these questions, we need to return to the scientific questions and hypotheses themselves. In my opinion, future scientists should focus less on using new technologies to simply generate vast amounts of data and instead return to first developing scientific hypotheses aimed at real scientific questions based on these data and pay more attention to test the hypotheses in a deductive approach. Although the inductive and deductive approaches are different, they are also complementary. For a complete cycle of scientific research, investigators/researchers begin with a set of phenomena, then look for patterns, before exploring different theories to explain these patterns and developing a working hypothesis. Next, they design experiments that allow the testing of their hypothesis, before either verifying or amending their theories. Recently, using a similar approach, several studies on ‘synthetic microbial community’ or ‘synthetic microbiome’ have emerged (see Lawson et al., 2019 for references?). These studies are characterized by a ‘design–build–test–learn’ (DBTL) cycle. The cycle begins with a hypothesis from a microbial ecological theory, before a model is designed that embodies the hypothesis. Subsequently, microbiomes are built on the basis of the design and then tested for their performances, including how function changes over time, to determine whether the results are in line with the hypothesis, and to ultimately obtain a deeper understanding from the results. This approach has also been used for industrial applications, such as biosynthesis of valuable products or degradation of refractory pollutants, based on rational design and construction of microbiomes (Gilmore et al., 2019; Sgobba and Wendisch, 2020).

The basis of design is the hypothesis and assumptions supporting the hypothesis. Hypotheses are constructed not to simply fit the data, but to explain phenomena observed. For example, the statement ‘the community changed with temperature’ is not really a hypothesis, as it only describes a phenomenon of the community. It fails to regard the mechanisms and cannot explain how the community forms in a given environment. By contrast, ‘higher temperatures should favour the slower‐growing species in a bacterial community with microbial interactions’ is a much better hypothesis (Lax et al., 2020), which considers the mechanistic assumption (i.e. the difference in growth rates between the two species decreases when the temperature is not optimal for both species), and can be tested by experiments.

On the basis of such a hypothesis, a quantitative model can actually be proposed, and experiments can be designed. For synthetic microbial ecology research, a quantitative mathematical model is needed to simulate the dynamics of microbiomes (Zomorrodi and Segre, 2016). The Lotka–Volterra model is one of the most commonly used population dynamic models that describes pairwise microbial interactions. Another useful model is the individual‐based model (IBM), which treats each individual cell as a discrete independent entity for exploring microbial interactions in space. In addition, the genome‐scale stoichiometric model based on flux balance analysis (FBA) quantitatively predicts the cellular metabolic fluxes to build a bridge between systems biology and microbial ecology. These models greatly simplify the complexity of microbiomes and identify the critical parameters suitable for experimental interrogation. Experiments can now be designed to test the prediction of the hypothesis and the quantitative models.

Synthetic microbiomes can be constructed using either a top‐down or bottom‐up approach to meet the assumptions of a hypothesis of synthetic microbiomes.

In contrast to the top‐down approach, the bottom‐up assembly begins with defined axenic strains or engineered strains. To test the fitness, the axenic strains are usually co‐cultured to test the fitness to each other and used to explore the effects of microbial interactions and random processes on the stability of microbial communities (Friedman et al., 2017; Gokhale et al., 2018; Abreu et al., 2019). They are also used to study the trade‐offs of different parameters such as temperature and growth rate (Lax et al., 2020). Compared with the axenic strains, the engineered strains have a clearer genetic background, and the interactions between X and Y are under better control. Using genetic circuits, more and more types of microbial interactions have recently been constructed artificially by engineering strains. Using quorum‐sensing regulatory modules, investigators/scientists/researchers can now construct consortia: is mutualism a consortium? how can ‘competition’ be ‘constructed?’ including mutualism, competition and predation interaction modes, and explore the critical factors on these interactions (Song et al., 2014). Moreover, interactions based on metabolite exchange such as cross‐feeding can also be constructed between inter‐ or intraspecies: inter already means ‘between/across’, intra already means ‘within’ ‐> for both within and across species (Shou et al., 2007). However, the consortia constructed with engineered strains usually contain fewer members than those from axenic strains. They are mainly used to explore microbial interactions. Ultimately, no matter what approach is being used, each is driven by a scientific question or hypothesis.

One of the bottlenecks for efficiently studying synthetic microbiomes is the lack of high‐throughput tests of the co‐cultures. Although mathematical models point out the critical parameters for the construction of microbiomes and simplify the design of experiments, the number of possible co‐cultures is extremely high, especially for synthetic microbiomes designed to contain multiple species. Moreover, for those from axenic strains, mass replicates are needed to ensure the stability of the results. A small number of high‐throughput methods have recently been developed for constructing and testing the synthetic microbiomes, containing up to several thousands of co‐cultures per day (Kehe et al., 2019). Another challenge to X is presented by Y the high‐throughput investigation of microbial compositions of mass co‐cultures. For co‐cultures from a few species, the morphology of colonies can be used to discriminate different species (Celiker and Gore, 2014). Therefore, to match the need for synthetic microbiome research, it is critical to develop low‐cost methods based on NGS and techniques of intelligent image recognition. The inductive methods are also important for the iterative process between model prediction and the testing of the hypothesis.

In summary, the development of low‐cost/high‐yield technologies greatly affects the way we do science. If publishing work solely based on new technologies, without actually asking a good scientific question becomes the norm, then our progress in understanding how nature works will slow considerably. New technologies might be the focus for tech‐focused researchers, but it should never be the focus of an entire research discipline. ‘Omics’ studies have by now deposited massive amounts of data into the databases, and it is now time to return to the question as to what can we actually learn from them. Without testing, the hypotheses and theories from ‘omics’ data based on inductive reasoning are only in the air. It is now time to return to ask hypothesis‐driven scientific questions in the field of microbial ecology and find approaches suitable to finding answers. Increased application of the deductive approach in synthetic microbial ecology and synthetic microbiome research will undoubtedly provide exciting new opportunities for advancing our understanding of microbial ecology.