Biology subjects

Yoon, B.-J.

Publications and source records attributed to Yoon, B.-J..

12 recordsLinked to original sources

bioRxiv2026

UPLiFT: Uncertainty Parameterized Activation for Lightweight Fine-Tuning

Learned protein representations contribute to the broad utility of pretrained protein language models (PLMs), often enhanced by task-specific fine-tuning where parameter-efficient techniques are imperative due to large network size. Our work takes an orthogonal approach to such fine-tuning: rather than optimizing the network weights, we formulate adaptation as an optimization task over the parameterized activation space. Motivated by the prior work on curvature-tuning, we propose UPLiFT, Uncertainty Parameterized activation for Lightweight Fine-Tuning, with enhanced adaptation capacity when the underlying model--whether pretrained or randomly initialized--remains fixed, for significantly improved parameter efficiency. Furthermore, its learnable non-linearity naturally allows to model uncertainty over the activation space through fine-tuning in a variational inference framework. Evaluation across multiple protein datasets shows that UPLiFT performs competitively while requiring 3-11x fewer trainable parameters than weight-only adaptations. Bayesian UPLiFT further improves calibration while largely preserving the predictive performance of deterministically fine-tuned models. The narrow predictive performance margin with the low-rank adaptation baseline in residue-level tasks highlights the potential of our proposed approach as a parameter-efficient fine-tuning technique for PLM-based predictors.

bioinformatics↗

bioRxiv2026

An Integrated Computational Antigen Discovery Pipeline with Hierarchical Filtering for Emerging Viral Variants

Emerging and evolving viral diseases, such as SARS-CoV-2, continue to pose significant global health challenges, underscoring the urgent need for rapid and scalable antigen discovery pipelines. This work presents a computational pipeline that integrates diverse computational tools and machine learning models to accelerate the identification and optimization of antigen candidates. The pipeline employs efficient filtering and consensus-based strategies to highlight epitopes with high therapeutic potential. We demonstrate its utility by significantly narrowing the antigen search space for Rift Valley fever virus (RVFV) and Mayaro virus (MAYV), and by effectively identifying conserved neutralizing epitopes in SARS-CoV-2. Our proposed computational antigen pipeline offers a powerful framework for expediting the development of future vaccines and therapeutics in response to emerging pathogens.

bioinformatics↗

bioRxiv2026

Rapid sequence-based screening of structure-disrupting protein mutations

Recent advances in AI-based protein structure prediction have dramatically reduced the cost of obtaining three-dimensional protein models and have become integral to modern protein engineering workflows. However, full structure prediction remains computationally prohibitive in high-throughput settings for mutation-based protein engineering, where thousands of candidate variants may need to be evaluated. In many such scenarios, the primary objective is not to resolve the complete structure of a candidate mutant, but rather to identify whether the introduced mutations are likely to induce substantial structural changes for rapid down-selection of candidates that conserve the wildtype structure. Protein language models (PLMs), trained solely on unlabeled natural protein sequence data, are known to encode rich structural information within their hidden representations. Motivated by this observation, we investigate a range of sequence-based ranking metrics derived from PLMs as efficient surrogates for structural deformation prediction. Through systematic evaluation across multiple proteins, mutation regimes, and structure-prediction backbones, we show that embedding distance--particularly the L1 distance between representations--provides a robust and computationally efficient signal for identifying structure-disrupting mutations. Our results demonstrate that sequence-level screening can substantially reduce the need for expensive structure prediction while preserving sensitivity to large structural perturbations, thereby providing the means to significantly speed up mutation-based protein design.

bioinformatics↗

bioRxiv2026

A Spatio-Temporal Analysis Framework for Characterizing Radiation-Induced Genomic Instability

Chronic low-dose ionizing radiation induces complex genomic instability encompassing both structural variants and point mutations, yet these alterations are typically analyzed as independent events, limiting detection of mechanistic coupling between rearrangement formation and localized mutagenesis at breakpoint junctions. This gap is particularly consequential given the widespread occupational and environmental exposure contexts such as nuclear energy, medical imaging, and environmental contamination, where coupled genomic alterations may contribute to cancer risk through mechanisms invisible to type-agnostic analyses. We developed an integrated analytical framework combining temporal pattern tracking, breakpoint-proximal mutation enrichment analysis, and systematic testing across all structural variant types to resolve these coupled dynamics across dose and time. Applying this framework to whole-genome sequencing data from primary human endothelial cells (HUVEC) exposed to chronic low-dose gamma radiation (0.20-2.62 mGy/hr) over three weeks, we identified inversion-specific mutagenic coupling; doublet base substitutions (DBS) were 7.13-fold enriched within 10bp of inversion breakpoints, a signal absent from other structural variant types, with sharp distance-dependent decay indicating localized mutagenesis at these junctions. Temporal analysis further revealed divergent fates of co-occurring alterations: inversions appeared transiently while DBS mutations showed greater persistence. These results illustrate how systematic integration of dose, time, and variant-type dimensions can uncover coupled mutagenic mechanisms that remain invisible in static or type-agnostic analyses. The framework is broadly applicable to longitudinal sequencing studies of genotoxic exposures, with applications to cancer genomics, radiation risk assessment, and mechanistic studies of DNA repair fidelity.

bioinformatics↗

bioRxiv2026

Structure-Based TCR-pMHC Binding Prediction and Generalization to Unseen Peptides

The interaction between T-cell receptors (TCRs) with the peptide-bound major histocompatibility complex (MHC) intricately impacts the functional specificity of T-cell-mediated adaptive immune response. Consequently, implication in immunotherapy has contributed to the ever-growing computational methods for TCR recognition, which have recently attracted structure-based approaches due to advancements in protein structure modeling. Despite access to structural information of the predicted binding interface, graph neural network (GNN)-based TCR-pMHC binding specificity classifiers tend to show poor accuracy for samples with unseen peptides. In this work, we comprehensively assess the potential factors that critically impact the generalization performance of classifiers trained with computationally predicted structures. Specifically, our experiments focus on analyzing the sensitivity of such predictors to the interaction features in the TCR-pMHC interface and the structural uncertainty. Building on the analysis, we demonstrate how the design of classifier architecture with auxiliary training objectives can improve the generalization performance to novel peptides not yet seen during model training. Overall, our work highlights the challenges of unseen peptide generalization from different perspectives of the GNN-based classifier paradigm, showcasing the strengths and weaknesses of the current state-of-the-art approaches in the generalization landscape.

bioinformatics↗

bioRxiv2026

Interpretable transcriptome-to-phenotype modeling of cell-painting nuclear morphology features from RNA-seq under low-dose radiation exposure

With rapid advancements in high-throughput multi-modal profiling techniques across molecular, cellular to tissue scales, translating such multi-modal data into knowledge discovery for foundational understanding of cellular mechanisms is central in modern biomedical sciences. In this study, we focus on understanding how low-dose radiation exposure perturbs cellular morphology by linking high-dimensional transcriptomic responses to quantitative cell-phenotype readouts over time. We present a time-resolved inverse modeling framework that associates gene expression changes via RNA-sequencing with nuclear morphology features obtained from cell-painting imaging. Morphology responses were defined as treated-control differences for multiple nuclear features including size, shape, intensity, and textures, indexed by radiation dose and week. To capture time-dependent associations while maintaining interpretability, both RNA-sequencing and cell-painting data were stratified into four temporal phases (weeks 1-2, 3-4, 5-6, 7-9) and phase-dependent effects are encoded via gene-phase interaction predictors. To reduce confounding by dose trends and to evaluate generalization across time, we used a two-stage leave-one-week-out procedure: (i) a dose-only baseline model produced out-of-week residuals for each morphology feature, and (ii) elastic-net regression on phase-aware predictors modeled residual variation not explained by dose. Hyperparameters were selected via an exhaustive grid search scored by the correlation between observed residuals and out-of-week residual predictions, with additional sparsity diagnostics based on nonzero coefficient counts per fold. Stable predictors were identified by selection frequency and sign consistency across folds, then pruned further for multicollinearity and parsimony. Final reduced models were fit using ordinary least squares with heteroskedasticity-consistent standard errors to report effect estimates robust to non-constant variance. This workflow yields a transparent, time-stratified set of transcriptomic predictors associated with longitudinal nuclear morphology changes and provides a reproducible foundation for downstream biological interpretation and validation.

bioinformatics↗

bioRxiv2024

Biologically Interpretable VAE with Supervision for Transcriptomics Data Under Ordinal Perturbations

AO_SCPLOWBSTRACTC_SCPLOWLatent variable models such as the Variational Auto-Encoders (VAEs) have shown impressive performance for inferring expression patterns for cell subtyping and biomarker identification from transcriptomics data. However, the limited interpretability of their latent variables obscures deriving meaningful biological understanding of cellular responses to different external and internal perturbations. We here propose a novel deep learning framework, EXPORT (EXPlainable VAE for ORdinally perturbed Transcriptomics data), for analyzing ordinally perturbed transcriptomics data that can incorporate any biological pathway knowledge in the VAE latent space. With the corresponding pathway-informed decoder, the learned latent expression patterns can be explained as pathway-level responses to perturbations, offering direct interpretability with biological understanding. More importantly, we explicitly model the ordinal nature of many real-world perturbations into the EXPORT framework by training an auxiliary ordinal regressor neural network to capture corresponding expression changes in the VAE latent representations, for example under different dosage levels of radiation exposure. By incorporating ordinal constraints during the training of our proposed framework, we further enhance the model interpretability by guiding the VAE latent space to organize perturbation responses in a hierarchical manner. We demonstrate the utility of the inferred guided latent space for downstream tasks, such as identifying key regulatory pathways associated with specific perturbation changes by analyzing transcriptomics datasets on both bulk and single-cell data. Overall, we envision that our proposed approach can unravel unprecedented biological intricacies in cellular responses to various perturbations while bringing an additional layer of interpretability to biology-inspired deep learning models.

bioinformatics↗

bioRxiv2023

Pharmacodynamic model of PARP1 inhibition and global sensitivity analyses can lead to cancer biomarker discovery

Pharmacodynamic models provide inroads to understanding key mechanisms of action and may significantly improve patient outcomes in cancer with improved ability to determine therapeutic benefit. Additionally, these models may also lead to insights into potential biomarkers that can be utilized for prediction in prognosis and therapeutic decisions. As an example of this potential, here we present an advanced computational Ordinary Differential Equation (ODE) model of PARP1 signalling and downstream effects due to its inhibition. The model has been validated experimentally and further evaluated through a global sensitivity analysis. The sensitivity analysis uncovered two model parameters related to protein synthesis and degradation rates that were also found to contribute the most variability to the therapeutic prediction. Because this variability may define cancer patient subpopulations, we interrogated genomic, transcriptomic, and clinical databases, to uncover a biomarker that may correspond to patient outcomes in the model. In particular, GSPT2, a GTPase with translation function, was discovered and if mutations serve to alter catalytic activity, its presence may explain the variability in the models parameters. This work offers an analysis of ODE models, inclusive of model development, sensitivity analysis, and ensuing experimental data analysis, and demonstrates the utility of this methodology in uncovering biomarkers in cancer. Author summaryBecause biochemical reaction networks are complex, dynamic, and typically provide output that results from non-linear interactions, mathematical models of such offer insight into cell function. In the clinic, models including drug action further their usefulness in that they may predict therapeutic outcome and other useful markers such as those for prognosis. In this study, we report a model of drug action that targets a critical protein, that when inhibited, promotes tumor cell death and documented remissions. Because all patients do not respond to the described treatment, a means to find cancer patient subpopulations that might benefit continues to be a challenge. Therefore, we analyzed the pharmacodynamic model by defining the parameters of the greatest variability and interrogated genomic, transcriptomic, and clinical cohort databases with this information and discovered a novel biomarker associated with prognosis in some ovarian and uterine cancer patients and separately, associated with the potential to respond to treatment.

bioinformatics↗

bioRxiv2022

Sensitivity analysis of genome-scale metabolic flux prediction

TRIMER, Transcription Regulation Integrated with MEtabolic Regulation, is a genome-scale modeling pipeline targeting at metabolic engineering applications. Using TRIMER, regulated metabolic reactions can be effectively predicted by integrative modeling of metabolic reactions with Transcription Factor (TF)-gene regulatory network (TRN), where the TRN is modeled via Bayesian network (BN). In this paper, we focus on sensitivity analysis of metabolic flux prediction considering potential model uncertainty in TRIMER. We propose a computational strategy to construct the uncertainty class of TRN models based on the inferred regulatory order uncertainty when learning from given transcriptomic expression data and analyze the prediction sensitivity of the TRIMER pipeline for the metabolite yield of interest. The obtained sensitivity analyses can provide a useful guidance for Optimal Experimental Design (OED) to help acquire new data that can enhance TRN modeling and effectively achieve specific metabolic engineering objectives, including metabolite yield alterations. We have performed simulation experiments to demonstrate the effectiveness of our developed sensitivity analysis strategy and its potential to effectively guide OED. ACM Reference FormatPuhua Niu, Maria J. Soto, Shuai Huang, Byung-Jun Yoon, Edward R. Dougherty,, Francis J. Alexander, Ian Blaby, Xiaoning Qian. 2018. Sensitivity analysis of genome-scale metabolic flux prediction. In Proceedings of Make sure to enter the correct conference title from your rights confirmation email (CNB-MAC 2022). ACM, New York, NY, USA, 9 pages. https://doi.org/XXXXXXX.XXXXXXX

bioinformatics↗

bioRxiv2021

AI-driven high-throughput automation of behavioral analysis and dual-channel wireless optogenetics for multiplexing brain dynamics

Advances in behavioral optogenetics are limited by the absence of high-throughput pipelines for the automated analysis of behavior in freely behaving animals. Although a variety of AI algorithms has been proposed that enable automation of behavioral analysis, existing methods are generally low-throughput. In addition, optogenetic manipulation of neural circuits typically requires physical tethers to light sources, which limits the number of brain areas that can be targeted and thus constrains behavioral testing. Here, we develop a new wireless platform that combines a battery-free dual-channel optogenetic implant with an AI algorithm for high-throughput behavioral analysis. In our platform, a customized AI algorithm detected and quantified freezing behavior of rats that had undergone fear conditioning. Notably, our platform allows up to four enclosures to be monitored simultaneously. Wireless dual-channel optogenetic devices were implanted in the basolateral amygdala (BLA) to permit independent modulation of BLA principal neurons (red light, AAV-CaMKII-JAWS) or BLA interneurons (blue light, AAV-mDlx-ChR2) in freely behaving animals. In vivo validation with behaving rats demonstrates the utility of the telemetry system for large-scale optogenetic studies. SignificanceAI algorithms can enable automation of behavioral analysis and thereby facilitate the progress on behavioral optogenetics. Successful integration of advanced wireless dual-channel optoelectronic devices with biological systems can also yield new tools and techniques for neuroscience research, particularly in the context of techniques for optogenetics. Here, we propose a new approach that combines an advanced AI algorithm with a low power wireless telemetry system, yielding powerful capabilities in the understanding of brain functions and the evaluation of the behavioral consequences of neural circuit manipulations. In vivo studies using optimized systems demonstrate high-throughput automation of behavioral manipulation and analysis via AI-powered wireless telemetry.

neuroscience↗

bioRxiv2021

TRIMER: Transcription Regulation Integrated with MEtabolic Regulation

Advances in bioengineering have enabled numerous bio-based commodities. Yet most traditional approaches do not extend beyond a single metabolic pathway and do not attempt to modify gene regulatory networks in order to buffer metabolic perturbations. This is despite access to near universal technologies allowing genome-scale engineering. To help overcome this limitation, we have developed a pipeline enabling analysis of Transcription Regulation Integrated with MEtabolic Regulation (TRIMER). TRIMER utilizes a Bayesian network (BN) inferred from transcriptomic data to model the transcription factor regulatory network. TRIMER then infers the probabilities of gene states that are of relevance to the metabolism of interest, and predicts metabolic fluxes resulting from deletion of transcription factors at the genome scale. Additionally, we have developed a simulation framework to mimic the TF-regulated metabolic network, capable of generating both gene expression states and metabolic fluxes, thereby providing a fair evaluation platform for benchmarking models and predictions. Here, we present this computational pipeline. We demonstrate TRIMERs applicability to both simulated and experimental data and show that it outperforms current approaches on both data types.

bioinformatics↗

bioRxiv2017

Characterizing co-expression networks underpinning maize stalk rot virulence in Fusarium verticillioides through computational subnetwork module analyses

Fusarium verticillioides is recognized as an important stalk rot pathogen of maize worldwide, but our knowledge of genetic mechanisms underpinning this pathosystem is limited. Previously, we identified a striatin-like protein Fsr1 that plays an important role in stalk rot. To further characterize transcriptome networks downstream of Fsr1, we performed next-generation sequencing (NGS) to investigate relative read abundance and also to infer co-expression networks utilizing the preprocessed expression data through partial correlation. We used a probabilistic pathway activity inference strategy to identify functional subnetwork modules likely involved in virulence. Each subnetwork modules consisted of multiple correlated genes with coordinated expression patterns, but the collective activation levels were significantly different in F. verticillioides wild type versus the mutant. We also identified putative hub genes from predicted subnetworks for functional validation and network robustness studies through mutagenesis, virulence and qPCR studies. Our results suggest that these genes are important virulence genes that regulate the expression of closely correlated genes, demonstrating that these are important hubs of their respective subnetworks. Lastly, we used key F. verticillioides virulence genes to computationally predict a subnetwork of maize genes that potentially respond to fungal genes by applying cointegration-correlation-expression strategy.

microbiology↗