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.

Distributed information encoding and decoding using self-organized spatial patterns.

Dynamical systems often generate distinct outputs according to different initial conditions, and one can infer the corresponding input configuration given an output. This property captures the essence of information encoding and decoding. Here, we demonstrate the use of self-organized patterns that generate high-dimensional outputs, combined with machine learning, to achieve distributed information encoding and decoding. Our approach exploits a critical property of many natural pattern-formation systems: in repeated realizations, each initial configuration generates similar but not identical output patterns due to randomness in the patterning process. However, for sufficiently small randomness, different groups of patterns that arise from different initial configurations can be distinguished from one another. Modulating the pattern-generation and machine learning model training can tune the tradeoff between encoding capacity and security. We further show that this strategy is scalable by implementing the encoding and decoding of all characters of the standard English keyboard.

Intra- and interpopulation transposition of mobile genetic elements driven by antibiotic selection.

The spread of genes encoding antibiotic resistance is often mediated by horizontal gene transfer (HGT). Many of these genes are associated with transposons, a type of mobile genetic element that can translocate between the chromosome and plasmids. It is widely accepted that the translocation of antibiotic resistance genes onto plasmids potentiates their spread by HGT. However, it is unclear how this process is modulated by environmental factors, especially antibiotic treatment. To address this issue, we asked whether antibiotic exposure would select for the transposition of resistance genes from chromosomes onto plasmids and, if so, whether antibiotic concentration could tune the distribution of resistance genes between chromosomes and plasmids. We addressed these questions by analysing the transposition dynamics of synthetic and natural transposons that encode resistance to different antibiotics. We found that stronger antibiotic selection leads to a higher fraction of cells carrying the resistance on plasmids because the increased copy number of resistance genes on multicopy plasmids leads to higher expression of those genes and thus higher cell survival when facing antibiotic selection. Once they have transposed to plasmids, antibiotic resistance genes are primed for rapid spread by HGT. Our results provide quantitative evidence for a mechanism by which antibiotic selection accelerates the spread of antibiotic resistance in microbial communities.

Modulation of microbial community dynamics by spatial partitioning.

Microbial communities inhabit spatial architectures that divide a global environment into isolated or semi-isolated local environments, which leads to the partitioning of a microbial community into a collection of local communities. Despite its ubiquity and great interest in related processes, how and to what extent spatial partitioning affects the structures and dynamics of microbial communities are poorly understood. Using modeling and quantitative experiments with simple and complex microbial communities, we demonstrate that spatial partitioning modulates the community dynamics by altering the local interaction types and global interaction strength. Partitioning promotes the persistence of populations with negative interactions but suppresses those with positive interactions. For a community consisting of populations with both positive and negative interactions, an intermediate level of partitioning maximizes the overall diversity of the community. Our results reveal a general mechanism underlying the maintenance of microbial diversity and have implications for natural and engineered communities.

Predicting plasmid persistence in microbial communities by coarse-grained modeling.

Plasmids are a major type of mobile genetic elements (MGEs) that mediate horizontal gene transfer. The stable maintenance of plasmids plays a critical role in the functions and survival for microbial populations. However, predicting and controlling plasmid persistence and abundance in complex microbial communities remain challenging. Computationally, this challenge arises from the combinatorial explosion associated with the conventional modeling framework. Recently, a plasmid-centric framework (PCF) has been developed to overcome this computational bottleneck. This framework enables the derivation of a simple metric, the persistence potential, to predict plasmid persistence and abundance. Here, we discuss how PCF can be extended to account for plasmid interactions. We also discuss how such model-guided predictions of plasmid fates can benefit from the development of new experimental tools and data-driven computational methods.

Large-scale tethered function assays identify factors that regulate mRNA stability and translation.

The molecular functions of the majority of RNA-binding proteins (RBPs) remain unclear, highlighting a major bottleneck to a full understanding of gene expression regulation. Here, we develop a plasmid resource of 690 human RBPs that we subject to luciferase-based 3'-untranslated-region tethered function assays to pinpoint RBPs that regulate RNA stability or translation. Enhanced UV-cross-linking and immunoprecipitation of these RBPs identifies thousands of endogenous mRNA targets that respond to changes in RBP level, recapitulating effects observed in tethered function assays. Among these RBPs, the ubiquitin-associated protein 2-like (UBAP2L) protein interacts with RNA via its RGG domain and cross-links to mRNA and rRNA. Fusion of UBAP2L to RNA-targeting CRISPR-Cas9 demonstrates programmable translational enhancement. Polysome profiling indicates that UBAP2L promotes translation of target mRNAs, particularly global regulators of translation. Our tethering survey allows rapid assignment of the molecular activity of proteins, such as UBAP2L, to specific steps of mRNA metabolism.

Machine learning applied to enzyme turnover numbers reveals protein structural correlates and improves metabolic models.

Knowing the catalytic turnover numbers of enzymes is essential for understanding the growth rate, proteome composition, and physiology of organisms, but experimental data on enzyme turnover numbers is sparse and noisy. Here, we demonstrate that machine learning can successfully predict catalytic turnover numbers in Escherichia coli based on integrated data on enzyme biochemistry, protein structure, and network context. We identify a diverse set of features that are consistently predictive for both in vivo and in vitro enzyme turnover rates, revealing novel protein structural correlates of catalytic turnover. We use our predictions to parameterize two mechanistic genome-scale modelling frameworks for proteome-limited metabolism, leading to significantly higher accuracy in the prediction of quantitative proteome data than previous approaches. The presented machine learning models thus provide a valuable tool for understanding metabolism and the proteome at the genome scale, and elucidate structural, biochemical, and network properties that underlie enzyme kinetics.

Finding relevant biomedical datasets: the UC San Diego solution for the bioCADDIE Retrieval Challenge.

Database URL: https://github.com/w2wei/dataset_retrieval_pipeline.