Biology subjects

Hsieh, C.-Y.

Publications and source records attributed to Hsieh, C.-Y..

10 recordsLinked to original sources

bioRxiv2026

SubcellPert: Benchmarking Subcellular Perturbation Prediction

Perturbations change both RNA abundance and intracellular organization, but gene-count predictors omit the positions needed to model spatial responses. To address this limitation, we introduce \textbf{\textit{SubcellPert}}, a benchmark for perturbation-response prediction at single-RNA-molecule resolution. It comprises 14 imaging datasets with 13.88 million cells and 3.65 billion RNA molecules. Given source observations and a directed condition change, models generate target RNA populations in normalized two-dimensional coordinates. Two independently trained tasks separate abundance from localization: \textit{Oracle} predicts coordinates given exact target counts; \textit{Joint} predicts both. Eight baselines span source resampling, classical transport, and frozen foundation-model adapters. Fixed perturbation and cellular-context holdouts evaluate abundance error, target-gene coverage, and conditional spatial error. Through within-dataset analyses, we find context-dependent gains from spatial features in condition decoding. Interpreting spatial responses requires distinguishing changes within spatial clusters from population composition, and RNA fractions from enrichment relative to compartment geometry. Through comprehensive experiments, we find that Gaussian transport achieves the lowest Oracle spatial error on eight of eleven cellular-context OOD datasets. Audits show that better centroid accuracy can accompany worse radial distributions, and that 19 of 190 Joint adapter fits retain source counts after validation. SubcellPert opens a spatial frontier for virtual cell research, laying the groundwork for predictive models that capture both the molecular composition and intracellular organization of cellular responses. The code is available at \url{https://github.com/king159/SubcellPert}.

bioinformatics↗

bioRxiv2026

ProtFlow: Flow Matching-based Protein Sequence Design with Comprehensive Protein Semantic Distribution Learning and High-quality Generation

Designing protein sequences with desired properties is a fundamental task in protein engineering. Recent advances in deep generative models have greatly accelerated this design process. However, most existing models face the issue of distribution centralization and focus on local compositional statistics of natural sequences instead of the global semantic organization of protein space, which confines their generation to specific regions of the distribution. These problems are amplified for functional proteins, whose sequence patterns strongly correlate with semantic representations and exhibit a long-tailed functional distribution, causing existing models to miss semantic regions associated with rare but essential functions. Here, we propose ProtFlow, a generative model designed for comprehensive semantic distribution learning of protein sequences, enabling high-quality sequence generation. ProtFlow employs a rectified flow matching algorithm to efficiently capture the underlying semantic distribution of the protein design manifold, and introduces a reflow technique enabling one-step sequence generation. We construct a semantic integration network to reorganize the representation space of large protein language models, facilitating stable and compact incorporation of global protein semantics. We pretrain ProtFlow on 2.6M peptide sequences and fine-tune it on antimicrobial peptides (AMPs), a representative class of therapeutic proteins exhibiting unevenly distributed activities across pathogen targets. Experiments show that ProtFlow outperforms state-of-the-art methods in generating high-quality peptides, and AMPs with desirable activity profiles across a range of pathogens, particularly against underrepresented bacterial species. These results demonstrate ProtFlows effectiveness in capturing the full training distribution and its potential as a general framework for computational protein design.

bioinformatics↗

bioRxiv2026

Ophiuchus-Ab: A Versatile Generative Foundation Model for Advanced Antibody-Based Immunotherapy

Antibodies exhibit extraordinary specificity and diversity in antigen recognition and have become a central class of therapeutics across a wide range of diseases. Despite this clinical success, antibody design remains fundamentally challenging. Antibody function emerges from intricate and highly coupled interactions between heavy and light chains, which complicate sequence-function relationships and limit the rational design of developable antibodies. Here, we reveal that modeling antibody sequence space at the level of paired heavy and light chains is essential to faithfully capture inter-chain dependencies, enabling a deeper understanding of antibody function and facilitating antibody discovery. We present Ophiuchus-Ab, a generative foundation model pre-trained on largescale paired antibody repertoires within a diffusion language modeling framework, unifying antibody generation and representation learning in a single probabilistic formulation. This framework excels diverse antibody design tasks, including CDR infilling, antibody humanization, and light-chain pairing. Beyond generation, diffusion-based pre-training yields transferable representations that enable accurate prediction of antibody properties, including developability, binding affinity, and specificity, even in low-data regimes. Together, these results establish Ophiuchus-Ab as a versatile foundation model for modeling antibodies, providing a foundation for next-generation antibody-based immunotherapy.

bioinformatics↗

bioRxiv2025

SynPROTAC: synthesizable PROTACs design through synthesis constrained generative model and reinforcement learning

Protein hydrolysis targeting chimeric (PROTAC) has emerged as a promising technology in degrading disease-related proteins for drug design. Recent deep generative models can accelerate PROTAC design, but the generated molecules are often difficult to synthesize. Here we develop SynPROTAC model, which employs Graphormer encoded warhead or E3 ligand as input, and autoregressively samples reaction templates and building blocks through transformer based decoder for PROTAC construction. The model is also fine-tuned via reinforcement learning for generating PROTACs with favorable binding properties. The comprehensive evaluations indicated that SynPROTAC is capable of generating novel PROTACs with feasible synthetic routes, reasonable physico-chemical and binding related properties.

bioinformatics↗

bioRxiv2024

Interpretable PROTAC degradation prediction with structure-informed deep ternary attention framework

Proteolysis Targeting Chimeras (PROTACs) are heterobifunctional ligands that form ternary complexes with Protein Of Interests (POIs) and E3 ligases, exploiting the ubiquitin-proteasome system to degrade disease-associated proteins, promising to drug the undruggable. While PROTAC research primarily relies on costly and time-consuming wet experimental approaches, deep learning offers a promising avenue to accelerate development and reduce expenses. However, existing deep learning methods for PROTAC degradation prediction often overlook the significance of hierarchical molecular representation and protein structural information, hindering effective data modeling. Moreover, their black-box nature limits the interpretability of computational outcomes, failing to provide intuitive insights into substructure interactions within the PROTAC system. This study introduces PROTAC-STAN, a structure-informed deep ternary attention network (STAN) framework for interpretable PROTAC degradation prediction. PROTAC-STAN represents PROTAC molecules across atom, molecule, and property hierarchies and incorporates structure information for POIs and E3 ligases using a protein language model infused with structural data. Furthermore, it simulates interactions among three entities at the atom and amino acid levels via a novel ternary attention network tailored for the PROTAC system, providing unprecedented insights into the degradation mechanism. By integrating hierarchical PROTAC molecule representation, structural embedding of POI and E3 ligase, and ternary attention network modeling interactions, our approach substantially improves prediction accuracy by 10.95% while enabling significant model interpretability via atomic and residue level visualization of molecule and complex. Experiments on the refined public PROTAC dataset demonstrate that PROTAC-STAN outperforms state-of-the-art baselines in overall performance. The excellent performance of PROTAC-STAN is anticipated to establish it as a foundational tool in future research on PROTAC-related drugs, thereby accelerating the development of this field.

bioinformatics↗

bioRxiv2024

Multi-Modal CLIP-Informed Protein Editing

AO_SCPLOWBSTRACTC_SCPLOWProteins govern most biological functions essential for life, but achieving controllable protein discovery and optimization remains challenging. Recently, machine learning-assisted protein editing (MLPE) has shown promise in accelerating optimization cycles and reducing experimental workloads. However, current methods struggle with the vast combinatorial space of potential protein edits and cannot explicitly conduct protein editing using biotext instructions, limiting their interactivity with human feedback. To fill these gaps, we propose a novel method called ProtET for efficient CLIP-informed protein editing through multi-modality learning. Our approach comprises two stages: in the pretraining stage, contrastive learning aligns protein-biotext representations encoded by two large language models (LLMs), respectively. Subsequently, during the protein editing stage, the fused features from editing instruction texts and original protein sequences serve as the final editing condition for generating target protein sequences. Comprehensive experiments demonstrated the superiority of ProtET in editing proteins to enhance human-expected functionality across multiple attribute domains, including enzyme catalytic activity, protein stability and antibody specific binding ability. And ProtET improves the state-of-the-art results by a large margin, leading to significant stability improvements of 16.67% and 16.90%. This capability positions ProtET to advance real-world artificial protein editing, potentially addressing unmet academic, industrial, and clinical needs.

bioinformatics↗

bioRxiv2024

ECloudGen: Access to Broader Chemical Space for Structure-based Molecule Generation

Structure-based molecule generation represents a significant advancement in AI-aided drug design (AIDD). However, progress in this domain is constrained by the scarcity of structural data on protein-ligand complexes, a challenge we term the Paradox of Sparse Chemical Space Generation. To address this limitation, we propose a novel latent variable approach that bridges the data gap between ligand-only and protein-ligand complexes, enabling the target-aware generative models to explore a broader chemical space and enhancing the quality of molecular generation. Drawing inspiration from quantum molecular simulations, we introduce ECloudGen, a generative model that leverages electron clouds as meaningful latent variables--an innovative integration of physical principles into deep learning frameworks. ECloudGen incorporates modern techniques, including latent diffusion models, Llama architectures, and a newly proposed contrastive learning task, which organizes the chemical space into a structured and highly interpretable latent representation. Benchmark studies demonstrate that ECloudGen outperforms state-of-the-art methods by generating more potent binders with superior physiochemical properties and by covering a significantly broader chemical space. The incorporation of electron clouds as latent variables not only improves generative performance but also introduces model-level interpretability, as illustrated in a case study designing V2R inhibitors. Furthermore, ECloudGens structurally ordered modeling of chemical space enables the development of a model-agnostic optimizer, extending its utility to molecular optimization tasks. This capability has been validated through a single-objective oracle benchmark and a complex multi-objective optimization scenario involving the redesign of endogenous BRD4 ligands. In conclusion, ECloudGen effectively addresses the Paradox of Sparse Chemical Space Generation through its integration of theoretical insights, advanced generative techniques, and real-world validation. The newly proposed technique of leveraging physical entities (such as electron clouds) as latent variables within a deep learning framework may prove useful for computational biology fields beyond AIDD.

bioinformatics↗

bioRxiv2023

OICNet: A Neural Network for Online EEG Source Separation using Independent Component Analysis

Online source separation of EEG signals plays a crucial role in understanding and interpreting brain dynamics in real-time applications such as brain-computer interfaces (BCIs). In this paper, we propose OICNet, a novel neural network designed specifically for online EEG source separation using independent component analysis, aiming to address the challenges of real-time computational efficiency and reliable extraction of independent components from EEG data streams. The OICNet is trained on a loss function that integrates non-Gaussianity measurement and an orthogonality constraint to achieve effective decomposition of multi-channel EEG signals. We conducted comprehensive evaluation of OICNet on both task-related and task-free EEG datasets with comparison against conventional and network-based ICA counterparts. The results demonstrate that OICNet outperforms existing methods in terms of accuracy and computational efficiency. Overall, OICNet provides high-efficiency real-time EEG source separation capabilities and paves the way for advancements in deep-learning EEG processing in real-world BCI applications.

bioinformatics↗

bioRxiv2021

Fuzzy Integral With Particle Swarm Optimization for CNN-Based Motor-Imagery EEG Classification

Recently, decoding human electroencephalographic (EEG) data using convolutional neural network (CNN) has driven the state-of-the-art recognition of motor-imagery EEG patterns for brain-computer interfacing (BCI). While a variety of CNN models have been used to classify motor-imagery EEG data, it is unclear if aggregating an ensemble of heterogeneous CNN models could further enhance the classification performance. To integrate the outputs of ensemble classifiers, this work utilizes fuzzy integral with particle swarm optimization (PSO) to estimate optimal confidence levels assigned to classifiers. The proposed frame-work aggregates CNN classifiers and fuzzy integral with PSO, achieving robust performance in single-trial classification of motor-imagery EEG data across various CNN model training schemes depending on the scenarios of BCI usage. This proof-of-concept study demonstrates the feasibility of applying fuzzy fusion techniques to enhance CNN-based EEG decoding and benefit practical applications of BCI.

bioengineering↗

bioRxiv2021

Estimating Intraclonal Heterogeneity and Subpopulation Changes from Perturbational Bulk Gene Expression Profiles in LINCS L1000 CMap by Premnas

BackgroundThe connectivity among signatures upon perturbations curated in the CMap library provides a valuable resource for understanding therapeutic pathways and biological processes associated with the drugs and diseases. MotivationHowever, due to the nature of bulk-level expression profiling by the L1000 assay, intraclonal heterogeneity and subpopulation compositional change that could contribute to the responses to perturbations are largely neglected, hampering the interpretability and reproducibility of the connections. ResultsIn this work, we proposed a computational framework, Premnas, to estimate the abundance of undetermined subpopulations from L1000 profiles in CMap directly according to an ad hoc subpopulation representation learned from a well-normalized batch of single-cell RNA-seq datasets by the archetypal analysis. By recovering the information of subpopulation changes upon perturbation, the potentials of searching for drug cocktails and drug-resistant/susceptible subpopulations with CMap L1000 were further explored and examined. ConclusionsThe proposed framework enables a new perspective to understand the connectivity among cellular signatures and expands the scope of the CMAP and other similar perturbation datasets limited by the bulk profiling technology. The executable and source code of Premnas is freely available at https://github.com/jhhung/Premnas. HighlightsO_LIA computational framework (i.e., Premnas) for learning subpopulation characteristics by the archetypal analysis and deconvoluting bulk expression profiles using the digital cytometry into subpopulation composition was proposed and validated. C_LIO_LIIntraclonal heterogeneity and subpopulation changes upon different perturbagen treatments in LINCS CMap L1000 datasets were estimated by Premnas. C_LIO_LIWith Premnas, we introduced a new strategy of finding effective drug cocktails and further linked the drug-resistant subpopulation found in CMap L1000 to a known drug-resistant clone (i.e., the pre-adapted [PA] cell). C_LIO_LITo our best knowledge, this work is the first attempt to provide a new subpopulation perspective to CMap database. C_LIO_LIWe believe Premnas can be applied to all the perturbation datasets, of which intraclonal/intratumoral heterogeneity was concealed by the bulk profiling and hereafter provides a new dimension of interpreting the connectivity. C_LI

bioinformatics↗