Biology subjects

Baek, M.

Publications and source records attributed to Baek, M..

At least 19 recordsLinked to original sources

bioRxiv2026

DAG-HEART: Directed Acyclic Graph-Guided Health Equity-Aware Representation Transfer Learning Framework for Breast Cancer

Breast cancer outcome prediction remains challenging for underrepresented populations because genomic datasets are demographically imbalanced and conventional multi-omics integration largely relies on undirected molecular similarity. We developed DAG-HEART, a directed acyclic graph-guided multi-omics transfer-learning framework that extends our previous transfer learning strategy with data augmentation. Using TCGA-BRCA mRNA, miRNA, and DNA-methylation data, DAG-HEART was evaluated for progression-free interval prediction in a data-minority group. DAG-guided nonlinear integration consistently improved predictive performance relative to direction-agnostic and correlation-based representations, while biologically motivated directional constraints generally outperformed reversed or unconstrained structures. Recurrently selected features converged on extracellular-matrix and regulatory pathways and supported clinically meaningful risk stratification. DAG-HEART provides an interpretable strategy for combining directed multi-omics structure with transfer learning under data imbalance across racial groups.

bioinformatics↗

bioRxiv2026

Ancient Somatosensory Circuit Architectures Employ Flexible Molecular Strategies

The extent to which conserved neural circuit architectures depend on shared molecular specification programs remains unclear. Here, we address this question by examining the somatosensory system of the little skate, Leucoraja erinacea, an early-diverging vertebrate that retains ancestral features of both finned and limb-based body plans. We show that core features of somatosensory circuit organization, including laminar organization of the spinal cord and dorsally restricted targeting of sensory afferents, are deeply conserved. Unexpectedly, the molecular programs specifying dorsal root ganglion (DRG) sensory subtypes diverge extensively from those of mammals. Although DRG neuron subtype specification and spinal connectivity rely on target-derived cues, skates employ distinct neurotrophin receptor and transcription factor identity codes. These findings support a model in which conserved spinal circuit architectures provide a stable scaffold that leverages flexible sensory neuron specification programs, enabling the evolutionary diversification of vertebrate somatosensory systems.

neuroscience↗

bioRxiv2026

Thal-Kak: unifying biomolecular structure predictors reveals a sampling-selection gap

Complementary all-atom structure predictors sample different solutions, but how to allocate a fixed sampling budget across them and select the best output remains unclear. Thal-Kak unifies five released predictors under shared upstream inputs and a common schema. Across FoldBench and CASP16, model mixing improves oracle sampling over single-model runs, but selection remains a bottleneck because confidence scores do not transfer across models and existing quality-assessment methods cannot resolve this gap.

bioinformatics↗

bioRxiv2026

Landscape-scale navigation unlocks antibody CDR structural logic for AI-guided rescue and therapeutic optimization

While AI offers transformative potential for therapeutic antibody design, the lack of ground-truth data fundamentally constrains our ability to model the epistatic topology of fitness landscapes. Here, we establish a high-throughput workflow to characterize tens of thousands of antibody variants per week with gold-standard biophysical precision. By combinatorially assembling functional variants from deep mutational scanning, we charted antibody fitness landscapes comprising over 17,000 data points, which revealed an extremely rugged, non-navigable epistatic topology. Yet, navigating at this unprecedented scale enabled the discovery of rare peak clusters exhibiting simultaneous enhancements in affinity and productivity. Strikingly, ProteinMPNN predicted the CDR-dependent productivity landscape with remarkable accuracy, suggesting that sequence-structure compatibility within CDRs gates cellular productivity. This insight enabled a structure-guided rescue strategy combining AlphaFold3 and ProteinMPNN, which successfully restored the cellular productivity of high-affinity, low-productivity clones via single amino acid substitutions. Two elite variants drawn directly from peak clusters further demonstrated 20- to 100-fold in vivo efficacy gains in a murine psoriasis model. Our findings establish CDR structural fitness as a fundamental determinant of antibody cellular productivity and validate landscape-scale navigation as a powerful framework for therapeutic antibody optimization.

bioengineering↗

bioRxiv2025

Prediction of zoonotic virus-host transmissibility using comparative airway organoids

AbstractThe emergence of new pandemic zoonotic diseases is accelerating. However, current approaches to control after outbreaks often allow the epidemic or pandemic spread of new zoonotic diseases, exemplified by the case of COVID-19. Prediction of zoonotic transmissibility will help minimize the outbreaks and socioeconomic damage. Here, we established highly efficient comparative airway organoids with minimal components from a panel of mammalian species for long-term culture and passaging and validated their utility to predict the transmissibility of zoonotic pathogens. Airway organoids from each species recapitulated the airway histology and contained the major cell types in the airway. They recapitulated the known susceptibility of their host species and tissues to diverse viruses. Transcriptomic analyses revealed that organoids from susceptible species induced genes of anti-viral immune responses only in common upon exposure to the same virus, although each species organoids displayed differential gene expression profiles. Also, organoids of the same species displayed differential transcriptomic responses against different viruses but induced anti-viral immune genes only in common. Our work provides a roadmap for comprehensive species- and tissue-level organoid panels for quick and reproducible uncharacterized pathogen surveillance to predict and prevent the next pandemic.

developmental biology↗

bioRxiv2025

LetA defines a structurally distinct transporter family involved in lipid trafficking

Membrane transport proteins translocate diverse cargos, ranging from small sugars to entire proteins, across cellular membranes. A few structurally distinct protein families have been described that account for most of the known membrane transport processes. However, many membrane proteins with predicted transporter functions remain uncharacterized. We determined the structure of E. coli LetAB, a phospholipid transporter involved in outer membrane integrity, and found that LetA adopts a distinct architecture that is structurally and evolutionarily unrelated to known transporter families. LetA functions as a pump at one end of a ~225 [A] long tunnel formed by its binding partner, MCE protein LetB, creating a pathway for lipid transport between the inner and outer membranes. Unexpectedly, the LetA transmembrane domains adopt a fold that is evolutionarily related to the eukaryotic tetraspanin family of membrane proteins, including TARPs and claudins. LetA has no detectable homology to known transport proteins, and defines a new class of membrane transporters. Through a combination of deep mutational scanning, molecular dynamics simulations, AlphaFold-predicted alternative states, and functional studies, we present a model for how the LetA-like family of membrane transporters may use energy from the proton-motive force to drive the transport of lipids across the bacterial cell envelope.

biophysics↗

bioRxiv2024

Computing the Human Interactome

Protein-protein interactions (PPI) are essential for biological function. Recent advances in coevolutionary analysis and Deep Learning (DL) based protein structure prediction have enabled comprehensive PPI identification in bacterial and yeast proteomes, but these approaches have limited success to date for the more complex human proteome. Here, we overcome this challenge by 1) enhancing the coevolutionary signals with 7-fold deeper multiple sequence alignments harvested from 30 petabytes of unassembled genomic data, and 2) developing a new DL network trained on augmented datasets of domain-domain interactions from 200 million predicted protein structures. These advancements allow us to systematically screen through 200 million human protein pairs and predict 18,316 PPIs with an expected precision of 90%, among which 5,578 are novel predictions. 3D models of these predicted PPIs nearly triple the number of human PPIs with accurate structural information, providing numerous insights into protein function and mechanisms of human diseases.

biochemistry↗

bioRxiv2024

FDA approved PDE4 inhibitors reduce the dominant toxicity of ALS/FTD associated CHCHD10S59L

Mutations in coiled-coil-helix-coiled-coil-helix domain containing 10(CHCHD10) have been identified as a genetic cause of amyotrophic lateral sclerosis and/or frontotemporal dementia(ALS-FTD). In our previous studies using in vivo Drosophila model expressing C2C10HS81L, and human cell models expressing CHCHD10S59L, we have identified that the PINK1/Parkin pathway is activated and causes cellular toxicity. Furthermore, we demonstrated that pseudo-substrate inhibitors for PINK1 and mitofusin2 agonists mitigated the cellular toxicity of CHCHD10S59L. Evidences using in vitro/ in vivo genetic and chemical tools indicate that inhibiting PINK1 would be the most promising treatment for CHCHD10S59L-induced diseases. Therefore, we have investigated cellular pathways that can modulate the PINK1/Parkin pathway and reduce CHCHD10S59L-induced cytotoxicity. Here, we report that FDA-approved PDE4 inhibitors reduced CHCHD10S59L-induced morphological and functional mitochondrial defects in human cells and an in vivo Drosophila model expressing C2C10HS81L. Multiple PDE4 inhibitors decreased PINK1 accumulation and downstream mitophagy induced by CHCHD10S59L. These findings suggest that PDE4 inhibitors currently available in the market may be repositioned to treat CHCHD10S59L-induced ALS-FTD and possibly other related diseases.

neuroscience↗

bioRxiv2024

Mapping and Engineering Antibody-Antigen Interaction Landscapes for Systematic Affinity Enhancement

Antibodies, crucial in adaptive immunity, recognize antigens through specific interactions facilitated by Complementarity Determining Regions (CDRs), diversified via Variable-Diversity-Joining (VDJ) recombination. Traditional antibody development, limited by the scope of animal models and phage display libraries, captures a fraction of the potential antibody-antigen interactions. This underscores a gap in understanding antibody specificity and the relationship between antibody sequence and binding affinity. Here we introduce an approach using the Single-Protein Interaction Detection (SPID) platform, repurposed to systematically map local landscapes of antibody-antigen interactions with unprecedented depth and speed, aiming to rival the precision of methods like Surface Plasmon Resonance (SPR) and Bio-Layer Interferometry (BLI) while significantly boosting throughput. By editing CDR sequences and measuring effects on dissociation constants, we elucidated pathways for optimizing antibody affinity, enhancing predictive models for interactions. Our findings demonstrate the capability of the SPID platform to characterize thousands of variants weekly, offering a deeper insight into antibody-antigen interactions and advancing antibody development with finely-tuned affinities.

bioengineering↗

bioRxiv2024

Floral Cues and Flower Handling Tactics Affect Switching Decisions by Nectar-Foraging Bumble Bees

Nectar foraging bees change their use of floral resources as plant species appear in the environment and disappear over their lifetimes. The new flowers used may involve different cues and different nectar extraction tactics. Although bumble bees can adapt to changes in floral cues and required tactics, little is known about whether bees prioritize switching tactics or floral cues when deciding which plant species to switch to. In a laboratory assay, we forced Bombus impatiens (common eastern bumble bee) workers either to switch the handling tactic they were using or to continue using the tactic but to switch the colour of artificial flowers foraged on. We examined whether bees tendency to change their tactics was influenced by how similar in colour novel flowers were to familiar ones. We conducted a 2 x 2 factorial experiment using artificial flowers, manipulating the handling tactic that bees were initially trained (legitimate visitation or nectar robbing) and the similarity between novel and trained colours (similar or distinct). We found that under most conditions bees preferred to switch flower colours and retain handling tactics. However, when given experience with legitimate visitation and when novel flowers were markedly different in colour from those they had experienced previously, bees tended to switch tactic while continuing to forage on flowers of the same colour. These findings suggest that the similarity in colour of a new floral resource to the currently exploited resource, along with the flower handling tactic employed by bees, jointly plays an important role in decision-making by foraging bumble bees.

animal behavior and cognition↗

bioRxiv2024

The Role of Similarity of Stimuli and Responses in Learning by Nectar-Foraging Bumble Bees: A Test of Osgood's Model

Learning stimulus - response associations helps animals to adjust to changing environments. Sequentially learned associations may interact with each other, either reinforcing memory, a process referred to as transfer, or hindering memory, a process referred to as interference. According to Osgoods (1949) model, close similarity between new and previously learned stimuli can enhance the transfer of memory through a process of stimulus generalization. In contrast, the model proposes that if responses are different from those previously learned, generalizing stimuli may lead to confusion, resulting in the interference of memory. Except for some work in humans, the interaction between stimulus similarity and response similarity is poorly documented. Here, we tested Osgoods model using bumble bees (Bombus impatiens) foraging for sucrose on artificial flowers with varied colours (= stimuli) that required either legitimate visitation or nectar robbing (= responses). Bees were first allowed to forage on one type of flower, were then switched to another, and finally were returned to the initial flower type. We measured learning performance via flower handling time and the number of failed visits. Consistent with Osgoods model, bees made more failed visits when they switched between similarly coloured flowers requiring different foraging techniques but made fewer failed visits when switching between similarly coloured flowers with the same technique. Regardless of similarities in stimuli or responses, however, experienced bees were faster in handling flowers than were naive bees. Results taken together thus provided mixed support for Osgoods model. Possible explanations for the mixed results are discussed.

animal behavior and cognition↗

bioRxiv2024

Essential and virulence-related protein interactions of pathogens revealed through deep learning

Identification of bacterial protein-protein interactions and predicting the structures of the complexes could aid in the understanding of pathogenicity mechanisms and developing treatments for infectious diseases. Here, we developed a deep learning-based pipeline that leverages residue-residue coevolution and protein structure prediction to systematically identify and structurally characterize protein-protein interactions at the proteome-wide scale. Using this pipeline, we searched through 78 million pairs of proteins across 19 human bacterial pathogens and identified 1923 confidently predicted complexes involving essential genes and 256 involving virulence factors. Many of these complexes were not previously known; we experimentally tested 12 such predictions, and half of them were validated. The predicted interactions span core metabolic and virulence pathways ranging from post-transcriptional modification to acid neutralization to outer membrane machinery and should contribute to our understanding of the biology of these important pathogens and the design of drugs to combat them.

biochemistry↗

bioRxiv2023

Antibody-based immunotherapy targeting FAM19A5 reverses synaptic loss and improves cognitive function in Alzheimer's disease.

IntroductionAlzheimers disease (AD) is characterized by the dysregulation of synaptic balance, with progressive loss of synapses outpacing formation, ultimately leading to cognitive decline. However, the lack of effective strategies for restoring lost synapses poses a major barrier to improving clinical outcomes. MethodsWe developed NS101, a monoclonal antibody targeting FAM19A5, a brain-secreted protein. Its preclinical efficacy in restoring synapses and cognition was evaluated using APP/PS1 and P301S mice. The clinical safety and target engagement of NS101 were examined in human participants. ResultsFAM19A5 binds to LRRC4B, a postsynaptic adhesion molecule, leading to synapse reduction. Blocking this interaction with NS101 normalized the rate of synapse elimination in AD mice. This synaptic rebalancing restored the number and function of synapses, resulting in improved cognition. Systemically administered NS101 facilitated the transport of brain FAM19A5 into the bloodstream. DiscussionTargeting FAM19A5 may hold clinical promise for treating AD by restoring synaptic balance.

neuroscience↗

bioRxiv2023

Democratizing Protein Language Models with Parameter-Efficient Fine-Tuning

Proteomics has been revolutionized by large pre-trained protein language models, which learn unsupervised representations from large corpora of sequences. The parameters of these models are then fine-tuned in a supervised setting to tailor the model to a specific downstream task. However, as model size increases, the computational and memory footprint of fine-tuning becomes a barrier for many research groups. In the field of natural language processing, which has seen a similar explosion in the size of models, these challenges have been addressed by methods for parameter-efficient fine-tuning (PEFT). In this work, we newly bring parameter-efficient fine-tuning methods to proteomics. Using the parameter-efficient method LoRA, we train new models for two important proteomic tasks: predicting protein-protein interactions (PPI) and predicting the symmetry of homooligomers. We show that for homooligomer symmetry prediction, these approaches achieve performance competitive with traditional fine-tuning while requiring reduced memory and using three orders of magnitude fewer parameters. On the PPI prediction task, we surprisingly find that PEFT models actually outperform traditional fine-tuning while using two orders of magnitude fewer parameters. Here, we go even further to show that freezing the parameters of the language model and training only a classification head also outperforms fine-tuning, using five orders of magnitude fewer parameters, and that both of these models outperform state-of-the-art PPI prediction methods with substantially reduced compute. We also demonstrate that PEFT is robust to variations in training hyper-parameters, and elucidate where best practices for PEFT in proteomics differ from in natural language processing. Thus, we provide a blueprint to democratize the power of protein language model tuning to groups which have limited computational resources.

bioinformatics↗

bioRxiv2023

Generalized Biomolecular Modeling and Design with RoseTTAFold All-Atom

Although AlphaFold2 (AF2) and RoseTTAFold (RF) have transformed structural biology by enabling high-accuracy protein structure modeling, they are unable to model covalent modifications or interactions with small molecules and other non-protein molecules that can play key roles in biological function. Here, we describe RoseTTAFold All-Atom (RFAA), a deep network capable of modeling full biological assemblies containing proteins, nucleic acids, small molecules, metals, and covalent modifications given the sequences of the polymers and the atomic bonded geometry of the small molecules and covalent modifications. Following training on structures of full biological assemblies in the Protein Data Bank (PDB), RFAA has comparable protein structure prediction accuracy to AF2, excellent performance in CAMEO for flexible backbone small molecule docking, and reasonable prediction accuracy for protein covalent modifications and assemblies of proteins with multiple nucleic acid chains and small molecules which, to our knowledge, no existing method can model simultaneously. By fine-tuning on diffusive denoising tasks, we develop RFdiffusion All-Atom (RFdiffusionAA), which generates binding pockets by directly building protein structures around small molecules and other non-protein molecules. Starting from random distributions of amino acid residues surrounding target small molecules, we design and experimentally validate proteins that bind the cardiac disease therapeutic digoxigenin, the enzymatic cofactor heme, and optically active bilin molecules with potential for expanding the range of wavelengths captured by photosynthesis. We anticipate that RFAA and RFdiffusionAA will be widely useful for modeling and designing complex biomolecular systems.

biochemistry↗

bioRxiv2023

Protein Ensemble Generation through VariationalAutoencoder Latent Space Sampling

Mapping the ensemble of protein conformations that contribute to function and can be targeted by small molecule drugs remains an outstanding challenge. Here we explore the use of soft-introspective variational autoencoders for reducing the challenge of dimensionality in the protein structure ensemble generation problem. We convert high-dimensional protein structural data into a continuous, low-dimensional representation, carry out search in this space guided by a structure quality metric, then use RoseTTAFold to generate 3D structures. We use this approach to generate ensembles for the cancer relevant protein K-Ras, training the VAE on a subset of the available K-Ras crystal structures and MD simulation snapshots, and assessing the extent of sampling close to crystal structures withheld from training. We find that our latent space sampling procedure rapidly generates ensembles with high structural quality and is able to sample within 1 angstrom of held out crystal structures, with a consistency higher than MD simulation or AlphaFold2 prediction. The sampled structures sufficiently recapitulate the cryptic pockets in the held-out K-Ras structures to allow for small molecule docking.

biochemistry↗

bioRxiv2023

Efficient and accurate prediction of protein structure using RoseTTAFold2

AlphaFold2 and RoseTTAFold predict protein structures with very high accuracy despite substantial architecture differences. We sought to develop an improved method combining features of both. The resulting method, RoseTTAFold2, extends the original three-track architecture of RoseTTAFold over the full network, incorporating the concepts of Frame-aligned point error, recycling during training, and the use of a distillation set from AlphaFold2. We also took from AlphaFold2 the idea of structurally coherent attention in updating pair features, but using a more computationally efficient structure-biased attention as opposed to triangle attention. The resulting model has the accuracy of AlphaFold2 on monomers, and AlphaFold2-multimer on complexes, with better computational scaling for large proteins and complexes. This excellent performance is achieved without hallmark features of AlphaFold2, invariant point attention and triangle attention, indicating that these are not essential for high accuracy prediction. Almost all recent work on protein structure prediction has re-used the basic AlphaFold2 architecture; our results show that excellent performance can be achieved with a broader class of models, opening the door for further exploration.

biochemistry↗

bioRxiv2022

Identification of hyperosmotic stress-responsive genes in Chinese hamster ovary cells via genome-wide virus-free CRISPR/Cas9 screening

Chinese hamster ovary (CHO) cells are the preferred mammalian host cells for therapeutic protein production that have been extensively engineered to possess the desired attributes for high-yield protein production. However, empirical approaches for identifying novel engineering targets are laborious and time-consuming. Here, we established a genome-wide CRISPR/Cas9 screening platform for CHO-K1 cells with 111,651 guide RNAs (gRNAs) targeting 21,585 genes using a virus-free recombinase-mediated cassette exchange-based gRNA integration method. Using this platform, we performed a positive selection screening under hyperosmotic stress conditions and identified 180 genes whose perturbations conferred resistance to hyperosmotic stress in CHO cells. Functional enrichment analysis identified hyperosmotic stress responsive gene clusters, such as tRNA wobble uridine modification and signaling pathways associated with cell cycle arrest. Furthermore, we validated 32 top-scoring candidates and observed a high rate of hit confirmation, demonstrating the potential of the screening platform. Knockout of the novel target genes, Zfr and Pnp, in monoclonal antibody (mAb)-producing recombinant CHO (rCHO) cells and bispecific antibody (bsAb)-producing rCHO cells enhanced their resistance to hyperosmotic stress, thereby improving mAb and bsAb production. Overall, the collective findings demonstrate the value of the screening platform as a powerful tool to investigate the functions of genes associated with hyperosmotic stress and to discover novel targets for rational cell engineering on a genome-wide scale in CHO cells.

bioengineering↗