Biology subjects

Li, J. D.

Publications and source records attributed to Li, J. D..

4 recordsLinked to original sources

bioRxiv2026

The Role of Training Paradigms in Shaping Connectivity and Dynamics: A Comparison of Gradient-Based and Evolutionary Methods in Recurrent Neural Network Models

Recurrent neural networks (RNNs) trained with backpropagation through time (BPTT) use gradients and global error signals to solve tasks, while evolutionary algorithms (EAs) offer an alternative solution through gradient-free optimization. Both classes of methods can solve the same tasks by modifying network weights, but it remains unclear how the choice of training paradigm biases the final connectivity structure and resulting dynamics. When drawing conclusions from task-trained RNNs, especially as proxies for neurobiological computation, it is important to consider whether the resulting network structure is due to the training method itself. Here, we compare four training paradigms--BPTT, evolution strategies (ES), genetic algorithms (GA), and GA combined with Ojas Hebbian plasticity rule (GA+Oja) -- across RNNs of 32, 64, and 128 neurons. These training paradigms were applied to two contrasting tasks that are exemplary of tasks which animals perform: a discrete working memory task and a continuous sensorimotor integration task. For both tasks, we analyzed how each training algorithm modified the network through four measurements: weight change allocations across input (Win), recurrent (Wrec), and output (Wout) layers, the effective rank of the recurrent weight matrix, the dimensionality of hidden-state dynamics, and task accuracy. BPTT achieved near-perfect accuracy across all task conditions, progressively allocated more weight changes to Wout as the working memory tasks difficulty increased, and confined hidden-state activity to a lower-dimensional subspace than any evolutionary method. Evolutionary methods maintained higher recurrent effective rank, higher activity dimensionality, and no comparable difficulty-dependent reallocation toward the readout, all while maintaining comparable task accuracy as BPTT. These findings show how gradient-based and gradient-free algorithms discover distinct structural and dynamical solutions to the same computational problems, especially in tasks involving working memory, with important implications for analyzing task-trained RNNs as models of biological neural computation. Biological neural circuits, which are shaped by evolution and local plasticity rather than gradient descent, may operate in higher-dimensional regimes than gradient-trained RNN models predict.

neuroscience↗

bioRxiv2026

BiDBiC: A novel ultra-high-throughput pipeline for Bead-in-Droplet Biofilm Cultivation and Characterization

Biofilms are a form of microbial growth consisting of cells, often attached to a surface, embedded in a structured 3D extracellular matrix that confers important emergent properties such as increased resistance to physical removal and antimicrobials. Despite the importance of biofilms to a variety of systems and despite increasing attention from both the public and private sectors, high-throughput approaches to study them are scarce, limiting investigations of complex mechanisms critical for the structure and function of biofilms, such as interactions in multispecies communities. We thus developed a novel workflow to grow and analyze bacterial cells adhered to plastic beads encapsulated within highly parallel nanoliter-scale water-in-oil microfluidic droplets. We term this pipeline for bead-in-droplet biofilm cultivation and characterization BiDBiC. To benchmark BiDBiC, we utilized a well-characterized biofilm former, Stenotrophomonas maltophilia, as well as a poorly studied drinking water biofilm isolate, Sphingopyxis sp. OPL5. Each bacterium exhibited strong adherent growth when co-encapsulated with polystyrene beads in droplets. Furthermore, we retrieved beads from the droplets and removed planktonic cells, enabling focused analysis of adhered cells. From bead-associated biomass, we extracted DNA and RNA for molecular analysis and recovered viable cells for subculturing. We conclude with a discussion of further development of the platform as well as suggestions for microbial biofilm systems that may benefit from ultra-high-throughput droplet-enabled cultivation and analysis. Insight BoxBiofilms are an important yet understudied form of microbial growth. In this study, we developed bead-in-droplet biofilm cultivation (BiDBiC), a novel ultra-high-throughput workflow to culture biofilms. The droplets act as massively parallelized miniature bioreactors, with co-encapsulated plastic beads providing a surface for cell attachment and growth. Using a well-characterized biofilm former, Stenotrophomonas maltophilia, as well as a drinking water biofilm isolate, Sphingopyxis sp. OPL5, we demonstrated robust adherent cell growth in droplets. We additionally efficiently separated beads from planktonic cells, enabling targeted molecular analysis and outgrowth of adherent cells. Adapting and extending BiDBiC could facilitate the study of numerous complex biofilm systems, such as diverse isolates or combinatorial subcommunities of microbiomes, to observe their phenotypes and probe underlying mechanisms.

microbiology↗

bioRxiv2025

Massively parallel metabarcoding of droplet co-cultures

A systems approach to microbial community ecology requires high-throughput tools for identifying key microbial interactions. While microfluidic droplets enable the high throughput generation, co-cultivation, and sorting of miniaturized co-cultures in sub-nanoliter, uniform water-in-oil emulsions, analyzing composition of individual droplet co-cultures at comparable throughputs has been challenging, particularly with environmentally isolated or less genetically tractable strains. To address this bottleneck, we present Cocoa-seq (combinatorial co-cultivation and amplicon sequencing), a droplet-based microfluidic workflow that enables 16S rRNA gene amplicon sequencing for thousands of droplet co-cultures in parallel. In summary, after co-cultivation of microbial co-cultures in agarose droplets, which are then set and recovered as discrete gel beads, Cocoa-seq pairs individual gel beads with barcoded primer beads in new droplets to produce multiplexed amplicon libraries via droplet PCR for sequencing. To benchmark Cocoa-seq, we used a model two-species co-culture and four-member mock communities with three different compositions. The community profiles derived from Cocoa-seq were qualitatively consistent with fluorescence microscopy-based estimates for the two-species co-culture and correlated well with mock community expectations, even for low-abundance representatives. Attempts to incorporate spike-in 16S standards for quantification of absolute abundance were hindered by PCR stochasticity at the single-molecule level, and we provide a simulation-based explanation of this bias and recommend against relying on spike-ins for quantitative inference. ImportanceResearchers in microbial ecology are increasingly utilizing sub-nanoliter microfluidic droplets to construct, grow, sort, and analyze miniaturized microbial co-cultures to identify and characterize critical microbial interactions. However, while the generation, incubation, and sorting of droplet co-cultures is scalable, analyzing the composition of these droplet co-cultures in large number is much more difficult. We demonstrate and benchmark a workflow for multiplexing 16S rRNA amplicon libraries derived from thousands of individual droplet co-cultures in a single sequencing run. This addition in the arsenal of growing microfluidic capabilities will support future study of critical interactions within complex microbial systems. Study fundingNational Science Foundation (Awards 2120909, 2426415) (awarded to XNL)

microbiology↗

bioRxiv2024

Within-host competition sparks pathogen molecular evolution and perpetual microbiota dysbiosis.

Pathogens newly invading a host must compete with resident microbiota. This within-host microbial warfare could lead to more severe disease outcomes or constrain the evolution of virulence. Using experimental evolution of a widespread pathogen (Staphylococcus aureus) and a native microbiota community in C. elegans nematode hosts, we show that a competitively superior pathogen displaced microbiota and reduced species richness, whilst maintaining virulence across generations. Conversely, pathogen populations and microbiota passaged separately caused more host harm relative to their respective ancestral controls. We find the evolved increase in virulence exhibited by pathogen populations passaged independently (compared to ancestral controls) was partly mediated by enhanced expression of the global virulence regulator agr and increased biofilm formation. Whole genome sequencing revealed shifts in the mode of selection from directional (on pathogens evolving alone) to fluctuating (on pathogens evolving with a host microbiota), with competitive interactions driving early diversification among pathogen populations. Metagenome sequencing of the evolved microbiota shows that evolution in infected hosts caused a significant reduction in community stability, along with restrictions on the co- existence of some species based on nutrient competition. Our study reveals how microbial competition during emerging infection determines the patterns and processes of evolution with major consequences for host health.

evolutionary biology↗