Data-driven learning of structure augments quantitative prediction of biological responses.

Multi-factor screenings are commonly used in diverse applications in medicine and bioengineering, including optimizing combination drug treatments and microbiome engineering. Despite the advances in high-throughput technologies, large-scale experiments typically remain prohibitively expensive. Here we introduce a machine learning platform, structure-augmented regression (SAR), that exploits the intrinsic structure of each biological system to learn a high-accuracy model with minimal data requirement. Under different environmental perturbations, each biological system exhibits a unique, structured phenotypic response. This structure can be learned based on limited data and once learned, can constrain subsequent quantitative predictions. We demonstrate that SAR requires significantly fewer data comparing to other existing machine-learning methods to achieve a high prediction accuracy, first on simulated data, then on experimental data of various systems and input dimensions. We then show how a learned structure can guide effective design of new experiments. Our approach has implications for predictive control of biological systems and an integration of machine learning prediction and experimental design.

Vertical and horizontal gene transfer tradeoffs direct plasmid fitness.

Plasmid fitness is directed by two orthogonal processes-vertical transfer through cell division and horizontal transfer through conjugation. When considered individually, improvements in either mode of transfer can promote how well a plasmid spreads and persists. Together, however, the metabolic cost of conjugation could create a tradeoff that constrains plasmid evolution. Here, we present evidence for the presence, consequences, and molecular basis of a conjugation-growth tradeoff across 40 plasmids derived from clinical Escherichia coli pathogens. We discover that most plasmids operate below a conjugation efficiency threshold for major growth effects, indicating strong natural selection for vertical transfer. Below this threshold, E. coli demonstrates a remarkable growth tolerance to over four orders of magnitude change in conjugation efficiency. This tolerance fades as nutrients become scarce and horizontal transfer attracts a greater share of host resources. Our results provide insight into evolutionary constraints directing plasmid fitness and strategies to combat the spread of antibiotic resistance.

Temporal encoding of bacterial identity and traits in growth dynamics.

In biology, it is often critical to determine the identity of an organism and phenotypic traits of interest. Whole-genome sequencing can be useful for this but has limited power for trait prediction. However, we can take advantage of the inherent information content of phenotypes to bypass these limitations. We demonstrate, in clinical and environmental bacterial isolates, that growth dynamics in standardized conditions can differentiate between genotypes, even among strains from the same species. We find that for pairs of isolates, there is little correlation between genetic distance, according to phylogenetic analysis, and phenotypic distance, as determined by growth dynamics. This absence of correlation underscores the challenge in using genomics to infer phenotypes and vice versa. Bypassing this complexity, we show that growth dynamics alone can robustly predict antibiotic responses. These findings are a foundation for a method to identify traits not easily traced to a genetic mechanism.

Autoencoder neural networks enable low dimensional structure analyses of microbial growth dynamics.

The ability to effectively represent microbiome dynamics is a crucial challenge in their quantitative analysis and engineering. By using autoencoder neural networks, we show that microbial growth dynamics can be compressed into low-dimensional representations and reconstructed with high fidelity. These low-dimensional embeddings are just as effective, if not better, than raw data for tasks such as identifying bacterial strains, predicting traits like antibiotic resistance, and predicting community dynamics. Additionally, we demonstrate that essential dynamical information of these systems can be captured using far fewer variables than traditional mechanistic models. Our work suggests that machine learning can enable the creation of concise representations of high-dimensional microbiome dynamics to facilitate data analysis and gain new biological insights.

Checks and Balances with Use of the Keio Collection for Phenotype Testing.

The Keio single gene knockout collection comprises approximately 4000 mutants of E. coli K12 strain BW25113, where each mutant contains a kanamycin resistance cassette in place of a single nonessential gene. This mutant library has proven to be incredibly useful in the fields of bacteriology, chemical genomics, biotechnology, and systems biology, which is evidenced by the greater than 3800 citations that the article describing its construction has garnered in the approximate first 11 years since its publication. Among the various applications of the collection, the most extensive use has been in the assessment of how loss of specific gene function influences phenotypes. In this chapter, we describe pitfalls with use of the collection and procedures that can be employed to ensure robust phenotype assessment of mutations in the library. These include procedures for thorough confirmation of gene deletions by PCR, phage transduction of mutated loci to new host strains, and strategies for genetic complementation.

Applying ecological resistance and resilience to dissect bacterial antibiotic responses.

An essential property of microbial communities is the ability to survive a disturbance. Survival can be achieved through resistance, the ability to absorb effects of a disturbance without a notable change, or resilience, the ability to recover after being perturbed by a disturbance. These concepts have long been applied to the analysis of ecological systems, although their interpretations are often subject to debate. Here, we show that this framework readily lends itself to the dissection of the bacterial response to antibiotic treatment, where both terms can be unambiguously defined. The ability to tolerate the antibiotic treatment in the short term corresponds to resistance, which primarily depends on traits associated with individual cells. In contrast, the ability to recover after being perturbed by an antibiotic corresponds to resilience, which primarily depends on traits associated with the population. This framework effectively reveals the phenotypic signatures of bacterial pathogens expressing extended-spectrum ß-lactamases (ESBLs) when treated by a ß-lactam antibiotic. Our analysis has implications for optimizing treatment of these pathogens using a combination of a ß-lactam and a ß-lactamase (Bla) inhibitor. In particular, our results underscore the need to dynamically optimize combination treatments based on the quantitative features of the bacterial response to the antibiotic or the Bla inhibitor.