Biology subjects

Weng, G.

Publications and source records attributed to Weng, G..

9 recordsLinked to original sources

bioRxiv2026

Cross-domain confidence reliability and remappability of frozen single-cell representations

Single-cell foundation models are increasingly adopted for downstream applications such as cell-type prediction. However, these predictions are often utilized without assessing their reliability, or by relying on a simple cutoff applied to the maximum softmax probability (MSP) derived from the classifier. This raises a critical question: can raw MSP be trusted as a reliable measure of confidence on unseen data (target) exhibiting diverse technical and biological variations? To address this, we introduce an audit framework to systematically evaluate the confidence estimates derived from the training set (source) against the realities of the target test set. We demonstrate that raw MSP fundamentally fails to represent true confidence when applied to target datasets. By decomposing the confidence gap between source and target data, we reveal that these discrepancies stem from a combination of rank-ordering errors and systemic probability drift. We further show that the drift can be successfully recalibrated by leveraging a small subset of labeled target data. While this recalibration improves the reliability of automated acceptance, it inherently introduces a trade-off by increasing the volume of cells requiring manual review. Ultimately, our framework establishes that the safe deployment of these models necessitates target-specific confidence adjustment using representative local labels.

bioinformatics↗

bioRxiv2026

A population readout of extrastriate activity reveals biased and smoothed temporal representations across saccades

Perception of visual time is transiently distorted around saccadic eye movements, yet the neuronal mechanisms constructing perisaccadic representations of time remain unclear. Here, we investigate how perisaccadic temporal information is encoded and read out in extrastriate cortex, by combining electrophysiological recordings in V4 and MT neuronal population in macaque monkeys under high spatiotemporal resolution visual stimulation, and a statistical modeling framework capturing perisaccadic response modulations at single-trial precision. Our analyses show that perisaccadic neuronal responses systematically shift the temporal representation of presaccadic stimuli within receptive fields-biasing them toward earlier times, and also reduce temporal sensitivity for presaccadic stimuli-impairing discrimination of stimulus onset times. Model-based readout using time-varying spatiotemporal sensitivity maps in neuronal ensembles enables quantitative characterization of these effects and identifies their specific neuronal response components at millisecond resolution. In silico manipulations further demonstrate a causal role of representational bias in reducing temporal sensitivity. These findings suggest that extrastriate cortex implements an active encoding strategy to stabilize presaccadic temporal information by favoring the most recent reliable input, revealing a fundamental tradeoff between temporal precision and robustness that supports a continuous visual percept across saccades. This finding also establishes a general role for extrastriate populations in constructing the perception of visual time.

neuroscience↗

bioRxiv2026

Limits of V4 perisaccadic firing rate modulations in explaining perceptual mislocalization

Whether and how various visual sensory areas contribute to the perceived location of visual stimuli remains unknown. To test the role of neurons in extrastriate area V4 in generating alterations in spatial perception during saccadic eye movements (saccades), we examined perisaccadic mislocalization--the perceptual phenomenon in which visual stimuli appearing around the time of a saccade are perceived at a different position than their actual location. We designed and implemented a combined behavioral and electrophysiological framework in non-human primates to directly relate trial-by-trial spatial perception reports during saccades to neuronal firing rates in V4 populations. We measured monkeys perception of stimulus location behaviorally and found perisaccadic mislocalization opposite to the saccade direction. We also quantified population responses by computing the center of mass of firing rate activity across probe locations for V4 neurons with receptive fields close to the saccade target. While perisaccadic neuronal responses showed shifts toward the saccade target, these shifts did not systematically vary with the magnitude of perceptual mislocalization across trials. In conclusion, receptive field shifts based on the perisaccadic firing rate of V4 neurons are not sufficient to account for the magnitude of perceptual mislocalization in each trial, suggesting that more complex neural representation of perisaccadic visual information may be critical for linking extrastriate neural activity to saccade-induced perception.

neuroscience↗

bioRxiv2026

FALCON: Closed-Loop Multi-Objective Optimization of Lipid Nanoparticles for Cell-Selective mRNA Delivery

Efficient, cell type-selective delivery of genetic payloads remains a central challenge in the development of gene and cell therapies. Lipid nanoparticles (LNPs) offer a versatile delivery platform, but their optimization is hindered by reliance on brute-force screening methods that are laborious, resource-intensive, and focus on single targets. Here, we present FALCON (Framework for Active Learning-driven Compositional Optimization of Nanoparticles), a closed-loop pipeline that leverages iterative screening, surrogate modeling, and multi-objective optimization to accelerate LNP compositional design. In B cell-targeted validation experiments, FALCON-optimized LNPs achieved a 1.8-fold increase in splenic B cell transfection in vivo compared with reference compositions. When optimized for selectivity, FALCON LNPs displayed an 84-fold improvement in selective transfection of splenic B cells over off-target liver populations and enabled spleen-tropic behavior across factorial panels of varying ionizable and helper lipid chemistries. In vaccine studies, these LNPs induced higher IgG2c antibody titers and a more Th1-biased immune profile. FALCON was also deployed to optimize LNPs for myeloid cell-selective delivery, achieving enhanced in vivo selectivity following systemic administration both across and within spleen and liver compartments. Our results establish FALCON as a useful tool for data-driven design of LNP compositions for precision gene delivery.

bioengineering↗

bioRxiv2023

Machine Learning Elucidates Design Features of Plasmid DNA Lipid Nanoparticles for Cell Type-Preferential Transfection

For cell and gene therapies to become more broadly accessible, it is critical to develop and optimize non-viral cell type-preferential gene carriers such as lipid nanoparticles (LNPs). Despite the effectiveness of high throughput screening (HTS) approaches in expediting LNP discovery, they are often costly, labor-intensive, and often do not provide actionable LNP design rules that focus screening efforts on the most relevant chemical and formulation parameters. Here we employed a machine learning (ML) workflow using well-curated plasmid DNA LNP transfection datasets across six cell types to maximize chemical insights from HTS studies and has achieved predictions with 5-9% error on average depending on cell type. By applying Shapley additive explanations to our ML models, we unveiled composition-function relationships dictating cell type-preferential LNP transfection efficiency. Notably, we identified consistent LNP composition parameters that enhance in vitro transfection efficiency across diverse cell types, such as ionizable to helper lipid ratios near 1:1 or 10:1 and the incorporation of cationic/zwitterionic helper lipids. In addition, several parameters were found to modulate cell type-preferentiality, including the ionizable and helper lipid total molar percentage, N/P ratio, cholesterol to PEGylated lipid ratio, and the chemical identity of the helper lipid. This study leverages HTS of compositionally diverse LNP libraries and ML analysis to understand the interactions between lipid components in LNP formulations; and offers fundamental insights that contribute to the establishment of unique sets of LNP compositions tailored for cell type-preferential transfection.

bioengineering↗

bioRxiv2023

Neural correlates of perisaccadic visual mislocalization in extrastriate cortex

When interacting with the visual world using saccadic eye movements (saccades), the perceived location of visual stimuli becomes biased, a phenomenon called perisaccadic mislocalization, which is indeed an exemplar of the brains dynamic representation of the visual world. However, the neural mechanism underlying this altered visuospatial perception and its potential link to other perisaccadic perceptual phenomena have not been established. Using a combined experimental and computational approach, we were able to quantify spatial bias around the saccade target (ST) based on the perisaccadic dynamics of extrastriate spatiotemporal sensitivity captured by statistical models. This approach could predict the perisaccadic spatial bias around the ST, consistent with the psychophysical studies, and revealed the precise neuronal response components underlying representational bias. These findings also established the crucial role of response remapping toward ST representation for neurons with receptive fields far from the ST in driving the ST spatial bias. Moreover, we showed that, by allocating more resources for visual target representation, visual areas enhance their representation of the ST location, even at the expense of transient distortions in spatial representation. This potential neural basis for perisaccadic ST representation, also supports a general role for extrastriate neurons in creating the perception of stimulus location.

neuroscience↗

bioRxiv2022

scmTE: multivariate transfer entropy builds interpretable compact gene regulatory networks by reducing false predictions

Gene regulatory network inference from single-cell RNA sequencing (scRNAseq) datasets has an incredible potential to discover new regulatory rules. However, current computational inference methods often suffer from excessive predictions as existing strategies fail to remove indirect or false predictions. Here, we report a new algorithm single-cell multivariate Transfer Entropy, scmTE, that generates interpretable regulatory networks with reduced indirect and false predictions. By utilizing multivariate transfer entropy, scmTE accounts for gene-to-gene interdependence when quantifying regulatory relationships. Benchmarking against other methods using synthetic data manifested that scmTE is the unique algorithm that did not produce a hair-ball structure (due to too many predictions) and recapitulated known ground-truth relationships with high accuracy. In silico knockdown experiments shows that scmTE assigns higher scores for specific interactions important for differentiation processes. We apply scmTE to T-cell differentiation, myelopoiesis and pancreatic development and identified known and novel regulatory interactions. scmTE provides a robust approach to infer interpretable networks by effectively removing unwanted indirect relationships.

bioinformatics↗

bioRxiv2021

Neighbor-specific gene expression revealed from physically interacting cells during mouse embryonic development

Development of multicellular organisms is orchestrated by persistent cell-cell communication between neighboring partners. Direct interaction between different cell types can induce molecular signals that dictate lineage specification and cell fate decisions. Current single cell RNAseq (scRNAseq) technology cannot adequately analyze cell-cell contact-dependent gene expression, mainly due to the loss of spatial information. To overcome this obstacle and resolve cell-cell contact-specific gene expression during embryogenesis, we performed RNA sequencing of physically interacting cells (PIC-seq) and assessed them alongside similar single cell transcriptomes derived from developing mouse embryos between embryonic day (E) 7.5 and E9.5. Analysis of the PIC-seq data identified novel gene expression signatures that were dependent on the presence of specific neighboring cell types. Our computational predictions, validated experimentally, demonstrated that neural progenitor (NP) cells overexpress Lhx5 and Nkx2-1 genes, when exclusively interacting with the definitive endoderm (DE) cell. Moreover, there was a reciprocal impact on the transcriptome of the DE cells, as they tend to overexpress Rax and Gsc genes when in contact with the NP cells. Using individual cell transcriptome data, we formulated a means of computationally predicting the impact of one cell type on the transcriptome of its neighboring cell types. We have further developed a distinctive spatial-tSNE to display the pseudo-spatial distribution of cells in a 2-dimensional space. In summary, we describe an innovative approach to study contact-specific gene regulation during embryogenesis with potential broader implication in other physiologically relevant processes. SignificancePhysical contact between neighboring cells is known to induce transcriptional changes in the interacting partners. Accurate measurement of these cell-cell contact based influences on the transcriptome is a very difficult experimental task. However, determining such transcriptional changes will highly enhance our understanding for the developmental processes. Current scRNAseq technology isolates the tissue into individual cells, making it hard to determine the potential transcriptomic changes due to its interacting partners. Here, we combined PIC-seq and computational algorithms to identify cell-type contact dependent transcriptional profiles focusing on endoderm development. We have computationally identified and experimentally validated specific gene expression patterns depending upon the presence of specific neighboring cell types. Our study suggests a new way to study cell-cell interactions for embryogenesis.

developmental biology↗

bioRxiv2020

VeTra: a new trajectory inference tool based on RNA velocity

Trajectory inference (TI) for single cell RNA sequencing (scRNAseq) data is a powerful approach to interpret dynamic cellular processes such as cell cycle and development. Still, however, accurate inference of trajectory is challenging. Recent development of RNA velocity provides an approach to visualize cell state transition without relying on prior knowledge. To perform TI and group cells based on RNA velocity we developed VeTra. By applying cosine similarity and merging weakly connected components, VeTra identifies cell groups from the direction of cell transition. Besides, VeTra suggests key regulators from the inferred trajectory. VeTra is a useful tool for TI and subsequent analysis.

bioinformatics↗