Biology subjects

Kim, S.

Publications and source records attributed to Kim, S..

At least 235 records · Page 13Linked to original sources

bioRxiv2024

Skeletal Metastasis of Prostate Cancer Is Augmented by Activation of EphA2 Noncanonical Signaling and Ligand-Deficient Bone Microenvironment

EphA2, a member of Eph family receptor tyrosine kinases (RTKs), is overexpressed in multiple types of solid human tumors, particularly at the late stages. However, whether and how it drives specific malignant processes remain elusive. We report that EphA2 is elevated during prostate cancer (PCa) progression in multiple syngeneic murine models. Interestingly, human metastatic PCa specimens from two Rapid Autopsy Programs showed selective overexpression of EphA2 in metastasis to the bone, but not to the lymph nodes or viscera. Serine 897 phosphorylation that mediates the pro-oncogenic, noncanonical signaling by EphA2 was also upregulated in bone metastasis. Analysis of human datasets shows EphA2 overexpression is associated with skeletal but not visceral metastases. Ephrin-A1, a major cognate ligand for EphA2, is lost in PCa bone metastasis, which is correlated with poor prognosis. Further, the bone microenvironment is unique in expressing little of the five EFNA genes, providing permissive microenvironment for bone colonization. S897A mutation that ablates EphA2 noncanonical signaling, suppressed PCa development. Restoration of ephrin-A1 expression in PC-3, a model cell line for double negative prostate cancer derived from bone metastasis and devoid of ephrin-As, profoundly changed global tyrosine phosphorylation profiles, inhibited basal ERK and Src activities in vitro, and suppressed tumor development in the bone. Together these results demonstrate EphA2 overexpression and concomitant loss of ligands in PCa lead to activation of noncanonical signaling that is sustained in ephrin-A1-deficient skeletal milieu to promote bone metastasis.

cancer biology↗

bioRxiv2024

Tailoring a CRISPR/Cas-based Epigenome Editor for Programmable Chromatin Acylation and Decreased Cytotoxicity

CRISPR-based epi-editors can robustly modulate cellular transcription and chromatin structure, but off-target activity and cytotoxicity limit their utility. Here, we engineer the acyl-CoA binding pocket of the human p300 histone acyltransferase to reduce its cytotoxicity and tune its acylation profile when fused to dCas9. We discover a single amino acid substitution (I1417N) that decreases cytotoxicity related to exogenous p300 overexpression yet preserves dCas9-p300 mediated histone acetylation and gene activation. We find that dCas9-p300 I1417N is less perturbative to the transcriptome and proteome of human cells, and that this behavior is driven by favorable stability kinetics. We also develop a crotonylation-biased dCas-p300 variant (I1395G) that selectively deposits histone crotonylation and activates transcription from endogenous promoters at levels comparable to wild-type p300. The p300 variants generated here enhance epi-editing capabilities and demonstrate that engineering of catalytic domains can be a powerful strategy for tailoring enzymatic activities and mitigating effector-driven toxicity in epi-editing.

synthetic biology↗

bioRxiv2024

Cholesterol binding to VCAM-1 promotes vascular inflammation

Hypercholesterolemia has long been implicated in endothelial cell (EC) dysfunction, but the mechanisms by which excess cholesterol causes vascular pathology are incompletely understood. Here we used a cholesterol-mimetic probe to map cholesterol-protein interactions in primary human ECs and discovered that cholesterol binds to and stabilizes the adhesion molecule VCAM-1. We show that accessible plasma membrane (PM) cholesterol in ECs is acutely responsive to inflammatory stimuli and that the nonvesicular cholesterol transporter Aster-A regulates VCAM-1 stability in activated ECs by controlling the size of this pool. Deletion of Aster-A in ECs increases VCAM-1 protein, promotes immune cell recruitment to vessels, and impairs pulmonary immune homeostasis. Conversely, depleting cholesterol from the endothelium in vivo dampens VCAM-1 induction in response to inflammatory stimuli. These findings identify cholesterol binding to VCAM-1 as a key step during EC activation and provide a biochemical explanation for the ability of excess membrane cholesterol to promote immune cell recruitment to the endothelium.

physiology↗

bioRxiv2024

Cytoplasmic ribosomes hitchhike on mitochondria to dendrites

Neurons rely on local protein synthesis to rapidly modify the proteome of neurites distant from the cell body. A prerequisite for local protein synthesis is the presence of ribosomes in the neurite, but the mechanisms of ribosome transport in neurons remain poorly defined. Here, we find that ribosomes hitchhike on mitochondria for their delivery to the dendrite of a sensory neuron in C. elegans. Ribosomes co-transport with dendritic mitochondria, and their association requires the atypical Rho GTPase MIRO-1. Disrupting mitochondrial transport prevents ribosomes from reaching the dendrite, whereas ectopic re-localization of mitochondria results in a concomitant re-localization of ribosomes, demonstrating that mitochondria are required and sufficient for instructing ribosome distribution in dendrites. Endolysosomal organelles that are involved in mRNA transport and translation can associate with mitochondria and ribosomes but do not play a significant role in ribosome transport. These results reveal a mechanism for dendritic ribosome delivery, which is a critical upstream requirement for local protein synthesis.

cell biology↗

bioRxiv2024

Context-Seq: CRISPR-Cas9 Targeted Nanopore Sequencing for Transmission Dynamics of Antimicrobial Resistance

Precisely understanding how and to what extent antimicrobial resistance (AMR) is exchanged between animals and humans is needed to inform control strategies. Metagenomic sequencing has low detection for rare targets such as antibiotic resistance genes, while whole genome sequencing of isolates misses exchange between uncultured bacterial species. We introduce Context-Seq, CRISPR-Cas9 targeted sequencing of ARGs and their genomic context with long-reads. Using Context-Seq, we investigate genetically similar AMR elements containing the ARGs blaCTX-M and blaTEM between adults, children, poultry, and dogs in Nairobi, Kenya. We identify genetically distinct clusters containing blaTEM and blaCTX-M that are shared between animals and humans within and between households. We also uncover potentially pathogenic hosts of ARGs including Escherichia coli, Klebsiella pneumonia, and Haemophilus influenzae in this study context. Context-Seq complements conventional methods to obtain an additional view of bacterial and mammalian hosts in the proliferation of AMR.

microbiology↗

bioRxiv2024

A blended genome and exome sequencing method captures genetic variation in an unbiased, high-quality, and cost-effective manner

We deployed the Blended Genome Exome (BGE), a DNA library blending approach that generates low pass whole genome (1-4x mean depth) and deep whole exome (30-40x mean depth) data in a single sequencing run. This technology is cost-effective, empowers most genomic discoveries possible with deep whole genome sequencing, and provides an unbiased method to capture the diversity of common SNP variation across the globe. To evaluate this new technology at scale, we applied BGE to sequence >53,000 samples from the Populations Underrepresented in Mental Illness Associations Studies (PUMAS) Project, which included participants across African, African American, and Latin American populations. We evaluated the accuracy of BGE imputed genotypes against raw genotype calls from the Illumina Global Screening Array. All PUMAS cohorts had R2 concordance [≥]95% among SNPs with MAF[≥]1%, and never fell below [≥]90% R2 for SNPs with MAF<1%. Furthermore, concordance rates among local ancestries within two recently admixed cohorts were consistent among SNPs with MAF[≥]1%, with only minor deviations in SNPs with MAF<1%. We also benchmarked the discovery capacity of BGE to access protein-coding copy number variants (CNVs) against deep whole genome data, finding that deletions and duplications spanning at least 3 exons had a positive predicted value of [~]90%. Our results demonstrate BGE scalability and efficacy in capturing SNPs, indels, and CNVs in the human genome at 28% of the cost of deep whole-genome sequencing. BGE is poised to enhance access to genomic testing and empower genomic discoveries, particularly in underrepresented populations.

genomics↗

bioRxiv2024

YOLOv8 Enables Automated Dragonfly Species Classification Using Wing Images

This study investigates the digitization of insect collections and the application of deep learning models to improve this process. During a one-hour filming session, 141 images of dragonfly specimens from the Cornell University Insect Collection were captured and preprocessed using five distinct methods: (1) adding box annotations around the wings, (2) adding polygon annotations to outline the forewings and hindwings, (3) removing vein system images, (4) retaining only the wing outline images, and (5) grouping by automatically measured wing size and classifying species within those groups. By comparing YOLOv8 models trained on datasets with these different preprocessing methods, the study revealed three key findings: (1) datasets with bounding box annotations result in shorter preprocessing times and superior model performance compared to polygon annotations; (2) although models trained with polygon annotations may have lower accuracy, they provide more detailed information on wing length and phenotypic traits; and (3) the wing vein system, rather than the wing outline, is the critical factor in classification accuracy.

ecology↗

bioRxiv2024

Lower-Limb Muscle Synergies in Musician's Dystonia: A Case Study of a Drummer

Musicians dystonia (MD) is a movement disorder characterized by involuntary muscle contractions specifically triggered by playing an instrument. This condition often leads to a loss of fine motor control, threatening the careers of affected musicians. While MD is commonly associated with the hands, it can also affect the lower limbs, particularly in drummers. Understanding the muscle coordination involved in MD is crucial for comprehending its neurological mechanisms, yet the muscle coordination of lower-limb dystonia has not been thoroughly explored. This study aimed to investigate the differences in lower-limb muscle synergies in a drummer with MD, utilizing Non-negative Matrix Factorization (NMF) to analyze coordinated muscle activity patterns during drumming tasks. A 36-year-old male professional drummer with lower-limb MD was instructed to play a drum set in time with a metronome set at 80 beats per minute. The task involved striking the bass drum pedal in time with the downbeat. Electromyographic (EMG) data were collected from ten muscles in the right lower limb. The data were analyzed using NMF to extract muscle synergies and compare the number of synergies, spatial modules, and temporal modules between the data with and without dystonia symptoms. The number of muscle synergies did not differ significantly between the data with and without symptoms. Notably, changes were observed in both the spatial and temporal modules of muscle synergies. Spatial modules revealed the appearance of dystonia-specific muscle synergy, which is considered related to compensatory movement. Temporal modules showed significant earlier overactivation in timing, which is considered the direct manifestation of dystonia symptoms. These findings indicate that lower-limb dystonia in drummers affects the spatial and temporal profiles of muscle synergies. This study underscores the importance of considering both spatial and temporal modules of muscle synergy in understanding and treating lower-limb dystonia in drummers. Further research is needed to validate these findings and apply muscle synergy analysis for the clinical assessment of lower-limb dystonia in drummers.

neuroscience↗

bioRxiv2024

Heterologous Production of Cyprosin B in Nicotiana benthamiana: Unveiling the Role of the Plant-Specific Insert Domain in Protein Function and Subcellular Localization

The aqueous extract of Cynara cardunculus flowers is traditionally used in cheese production across Mediterranean countries. To meet the growing industrial demand for plant-based milk-clotting enzymes and to explore potential biotechnological applications, we initiated a study to heterologously produce cyprosin B (CYPB), a key milk-clotting enzyme from C. cardunculus, in Nicotiana benthamiana. We also investigated the role of its plant-specific insert (PSI) domain in the CYPBs activity and its localization. In this study, full-length CYPB and a PSI domain deleted CYPB (CYPB{Delta}PSI) were transiently expressed in N. benthamiana leaves using Agrobacterium-mediated infiltration. The leaves were harvested nine days post-infiltration, and proteins were purified, yielding approximately 81 mg/kg (CYPB) and 60 mg/kg (CYPB{Delta}PSI) fresh weight. CYPB{Delta}PSI showed significantly higher proteolytic activity (156.72 IU/mg) than CYPB (57.2 IU/mg), indicating that the PSI domain is not essential for enzymatic activity and that its removal results in enhanced enzymatic efficiency. In the milk-clotting activity assay, CYPB{Delta}PSI demonstrated a significantly faster clotting time than full-length CYPB, indicating enhanced milk-clotting efficiency for CYPB{Delta}PSI. Subcellular localization studies revealed that CYPB and PSI were localized in the vacuole and endocytic vesicles. In contrast, CYPB{Delta}PSI was primarily localized in the endoplasmic reticulum (ER) and the tonoplast, suggesting that the PSI domain is critical for vacuolar targeting and membrane permeabilization that affects overall protein yield. This study demonstrates the feasibility of using N. benthamiana as a platform for the scalable production of more efficient recombinant CYPB. It highlights the multifunctional role of the PSI domain in vacuolar sorting without impairing its functionality. These results underscore the potential of plant-based expression systems as a viable alternative for the industrial production of plant milk-clotting enzymes, with significant implications for sustainable cheese production.

molecular biology↗

bioRxiv2024

Amyloid-β deposition in basal frontotemporal cortex is associated with selective disruption of temporal mnemonic discrimination

Cerebral amyloid-beta (A{beta}) accumulation, a hallmark pathology of Alzheimers disease (AD), precedes clinical impairment by two to three decades. However, it is unclear whether A{beta} contributes to subtle memory deficits observed during the preclinical stage. The heterogenous emergence of A{beta} deposition may selectively impact certain memory domains, which rely on distinct underlying neural circuits. In this context, we tested whether specific domains of mnemonic discrimination, a neural computation essential for episodic memory, exhibit specific deficits related to early A{beta} deposition. We tested 108 cognitively unimpaired human older adults (66% female) who underwent 18F-florbetapir positron emission tomography (A{beta}-PET), and a control group of 35 young adults, on a suite of mnemonic discrimination tasks taxing object, spatial, and temporal domains. We hypothesized that A{beta} pathology would be selectively associated with temporal discrimination performance due to A{beta}s propensity to accumulate in the basal frontotemporal cortex, which supports temporal processing. Consistent with this hypothesis, we found a dissociation in which generalized age-related deficits were found for object and spatial mnemonic discrimination, while A{beta}-PET levels were selectively associated with deficits in temporal mnemonic discrimination. Further, we found that higher A{beta}-PET levels in medial orbitofrontal and inferior temporal cortex, regions supporting temporal processing, were associated with greater temporal mnemonic discrimination deficits, pointing to the selective vulnerability of circuits related to temporal processing early in AD progression. These results suggest that A{beta} accumulation within basal frontotemporal regions may disrupt temporal mnemonic discrimination in preclinical AD, and may serve as a sensitive behavioral biomarker of emerging AD progression.

neuroscience↗

bioRxiv2024

Rough Grouping Enhances YOLO's Pollinator Classification and Detection from Small Datasets

This study addresses the challenges in pollinator monitoring by proposing an effective data structure for automated systems, with a focus on the use of machine learning to handle underrepresented groups in small datasets. By experimenting with grouping the top three pollinators (bee, butterfly, hoverfly) and non-pollinators in datasets of fewer than 300 samples, the research aims to enhance classification and detection accuracy. During 4-hour filming sessions, 181 images of insects larger than 1 cm were captured and classified into three grouping methods: "Pollinator/ Non-pollinator", "Bee/Butterfly/Hoverfly/Ant", and "Bumblebee/Honeybee/Butterfly/Hoverfly/ Ant". YOLO V8 models were trained, validated, and tested with these datasets based on different class grouping methods. The study found that the "Pollinator/Non-pollinator" YOLOv8 model performed best across all metrics, suggesting it is more reliable for categorizing groups and detecting target objects, especially with smaller, imbalanced datasets. This finding aligns with the trend in machine learning that providing more training opportunities for individual classes improves accuracy. Therefore, using broader categorization methods can enhance the reliability and accuracy of automated monitoring systems when training data is insufficient.

ecology↗

bioRxiv2024

The MRE11-RAD50-NBS1 complex both starts and extends DNA end resection in mouse meiosis

Nucleolytic resection of DNA ends is critical for homologous recombination, but its mechanism is not fully understood, particularly in mammalian meiosis. Here we examine roles of the conserved MRN complex (MRE11, RAD50, and NBS1) through genome-wide analysis of meiotic resection in mice with various MRN mutations, including several that cause chromosomal instability in humans. Meiotic DSBs form at elevated levels but remain unresected if Mre11 is conditionally deleted, thus MRN is required for both resection initiation and regulation of DSB numbers. Resection lengths are reduced to varying degrees in MRN hypomorphs or if MRE11 nuclease activity is attenuated in a conditional nuclease-dead Mre11 model. These findings unexpectedly establish that MRN is needed for longer-range extension of resection, not just resection initiation. Finally, resection defects are additively worsened by combining MRN and Exo1 mutations, and mice that are unable to initiate resection or have greatly curtailed resection lengths experience catastrophic spermatogenic failure. Our results elucidate multiple functions of MRN in meiotic recombination, uncover unanticipated relationships between short- and long-range resection, and establish the importance of resection for mammalian meiosis.

genetics↗

bioRxiv2024

Phase Separation Clustering of Poly Ubiquitin Cargos on the Ternary Mixture Lipid Membranes by Synthetically Cross-Linked Ubiquitin Binder Peptides.

Ubiquitylation is involved in various physiological processes in our bodies such as signaling and vesicle trafficking, thus Ubiquitin (UB) is an important medical target system of interest. Polymeric addition of UB allows cargo molecules to be recognized specifically by multivalent binding interaction with UB binding proteins leading to various downstream processes. Recently, implication of protein condensate formation by Ubiquitylated proteins have been reported in many independent UB processes suggesting its potential role in governing the spatial organization of Ubiquitylated cargo proteins. We created modular polymeric UB binder motifs and polymeric UB cargos by synthetic bioconjugation and protein purification. Giant unilamellar vesicles with lipid raft composition were created to reconstitute polymeric UB cargo organization on the membranes. Fluorescence imaging was used to observe the outcome. We found that polymeric UB cargos clustered on the membranes by forming phase separation codomains in interaction with multivalent UB binder conjugate. This phase separation was valency dependent and was strongly correlated to their potency to form protein condensate droplets in solution. Multivalent UB binding interactions showed a general trend toward the formation of phase separated condensates, and the resulting condensates were either in liquid-like or solid-like state depending on the conditions and interactions used. It implies that polymeric UB cargos on the plasma and endosomal membranes may use the codomain phase separation to assist clustering of UB cargos on the membranes for cargo sorting. Our work also shows that model systems of such phase behavior can be created by modular synthetic approach that can potentially be used to further engineer biomimetic interactions in vitro. Graphical abstracts (TOC) O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=102 SRC="FIGDIR/small/608403v1_ufig1.gif" ALT="Figure 1"> View larger version (30K): org.highwire.dtl.DTLVardef@eca49dorg.highwire.dtl.DTLVardef@2b64d2org.highwire.dtl.DTLVardef@e82b98org.highwire.dtl.DTLVardef@862558_HPS_FORMAT_FIGEXP M_FIG C_FIG

biochemistry↗

bioRxiv2024

HSP90 interacts with VP37 to facilitate the cell-to-cell movement of broad bean wilt virus 2

The systemic spread of viruses in plants requires successful viral cell-to-cell movement through plasmodesmata (PD). Viral movement proteins (MPs) interact with cellular proteins to modify and utilize host transport routes. Broad bean wilt virus 2 (BBWV2), a widespread plant RNA virus, moves from cell to cell as a virion through the PD gated by VP37, the MP of BBWV2. However, the host proteins that function in the cell-to-cell movement of BBWV2 remain unclear. In this study, we identified cellular heat shock protein 90 (HSP90) as an interacting partner of VP37. The interaction between HSP90 and VP37 was assessed using the yeast two-hybrid assay, co-immunoprecipitation, and bimolecular fluorescence complementation. Tobacco rattle virus-based virus-induced gene silencing analysis revealed that HSP90 silencing significantly inhibited the systemic spread of BBWV2 in N. benthamiana plants. Furthermore, in planta treatment with geldanamycin (GDA), an inhibitor of the chaperone function of HSP90, demonstrated the necessity of HSP90 in successful cell-to-cell movement and systemic infection of BBWV2. Interestingly, GDA treatment inhibited the HSP90-VP37 interaction at the PD, resulting in the inhibition of VP37-derived tubule formation through the PD. Our results suggest that the HSP90-VP37 interaction regulates VP37-derived tubule formation through the PD, thereby facilitating the cell-to-cell movement of BBWV2. IMPORTANCEThis study highlights the regulatory role of heat shock protein 90 (HSP90) in facilitating the cell-to-cell movement of broad bean wilt virus 2 (BBWV2). HSP90 interacted with VP37, the movement protein of BBWV2, specifically at plasmodesmata (PD). This study demonstrated that the HSP90-VP37 interaction is crucial for viral cell-to-cell movement and the formation of VP37-derived tubules, which are essential structures for virus transport through the PD. The ATP-dependent chaperone activity of HSP90 is integral to this interaction, as demonstrated by the inhibition of virus movement upon treatment with geldanamycin (GDA), which disrupts the function of HSP90. These findings elucidate the molecular mechanisms underlying the cell-to-cell movement of plant viruses, and highlight the role of HSP90 in viral infection. This study suggests that the chaperone activity of HSP90 may function in changing the conformational structure of VP37, thereby facilitating the assembly and function of virus-induced structures required for viral cell-to-cell movement.

microbiology↗

bioRxiv2024

Optimized methods for mapping DNA double-strand-break ends and resection tracts and application to meiotic recombination in mouse spermatocytes

DNA double-strand breaks (DSBs) made by SPO11 protein initiate homologous recombination during meiosis. Subsequent to DNA strand breakage, endo- and exo-nucleases process the DNA ends to resect the strands whose 5 termini are at the DSB, generating long 3-terminal single-stranded tails that serve as substrates for strand exchange proteins. DSB resection is essential for meiotic recombination, but a detailed understanding of its molecular mechanism is currently lacking. Genomic approaches to mapping DSBs and resection endpoints, e.g., S1-sequencing (S1-seq) and similar methods, play a critical role in studies of meiotic DSB processing. In these methods, nuclease S1 or other enzymes that specifically degrade ssDNA are used to trim resected DSBs, allowing capture and sequencing of the ends of resection tracts. Here, we present optimization of S1-seq that improves its signal:noise ratio and allows its application to analysis of spermatocyte meiosis in adult mice. Furthermore, quantitative features of meiotic resection are evaluated for reproducibility, and we suggest approaches for analysis and interpretation of S1-seq data. We also compare S1-seq to variants that use exonuclease T and/ or exonuclease VII from Escherichia coli instead of nuclease S1. Detailed step-by-step protocols and suggestions for troubleshooting are provided.

genomics↗

bioRxiv2024

GPR43 in Eosinophils Prevents the Emergence of Pathogenic Siglec-Fhi Neutrophils in Allergic Airway Inflammation

Eosinophils are major effector cells in type 2 immune responses, contributing to host defense and allergic diseases. They also play critical roles in maintaining tissue homeostasis by regulating various immune cell types. However, evidence of the crosstalk between eosinophils and neutrophils is limited. Here, we show that eosinophils directly associate with neutrophils in the lungs of asthma-induced mice. Eosinophil-specific deficiency of the short-chain fatty acid receptor GPR43 results in hyperactivation of eosinophils and increases the expression of neutrophil chemoattractants and PECAM-1, thus enhancing the interaction between eosinophils and neutrophils. This binding event exposes the neutrophils to eosinophil-derived IL-4 and GM-CSF, which induces the conversion of conventional neutrophils into more pathogenic Siglec-Fhi neutrophils that strongly enhance Th17 cell differentiation and aggravate asthma symptoms. These results reveal GPR43 as a critical regulator of eosinophils and highlight that eosinophils have a hitherto little-known ability to directly modulate neutrophil differentiation and function. One Sentence SummaryEosinophils directly recruit neutrophils and induce their differentiation into a pathogenic Siglec-Fhi subset in allergic airway inflammation.

immunology↗

bioRxiv2024

Context-Aware Hierarchical Fusion for Drug Relational Learning

Drug relational learning, focused on understanding drug-pair relationships within specific contexts of interest, has emerged as a critical area of investigation for its pivotal role in enhancing the efficacy of disease treatment. The nature of drug relationships exhibits significant variations across diverse contexts, such as different types of cancer cell lines. Existing methods often encounter limitations by either neglecting the incorporation of context information or lacking explicit modeling of the intricate connections within drug-drug-context triplets, due to the difficulty in handling heterogeneous relationships between drugs and context. In this study, we present a novel context-aware hierarchical cross-fusion architecture tailored for drug relational learning. By formulating the problem as the label prediction of drug-drug-context triplets, we explicitly calculate all the relations among the triplets. Considering drugs often function as causes and contexts serve as results, our model enhances the learning of intricate drug pair relations hierarchically fusing the information from drug to context through the learned relations. Empirical results across multiple prediction tasks, including synergy, polypharmacy side effects, and drug-drug interactions, highlight the models capability to capture essential information relevant to drug relational learning. Notably, our model demonstrates robust predictive performance even in scenarios of heightened contextual complexity, demonstrating its adaptability in learning context-aware drug relations. CCS CONCEPTS* Applied computing [->] Bioinformatics; Molecular structural biology; * Computing methodologies [->] Artificial intelligence. ACM Reference FormatYijingxiu Lu, Yinhua Piao, Sangseon Lee, and Sun Kim. 2024. Context-Aware Hierarchical Fusion for Drug Relational Learning. In Proceedings of Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (BIOKDD 24). ACM, New York, NY, USA, 11 pages. https://doi.org/10.1145/nnnnnnn.nnnnnnn

bioinformatics↗

bioRxiv2024

Mixture-of-Experts Approach for Enhanced Drug-Target Interaction Prediction and Confidence Assessment

In recent years, numerous deep learning models have been developed for drug-target interaction (DTI) prediction. These DTI models specialize in handling data with distinct distributions and features, often yielding inconsistent predictions when applied to unseen data points. This inconsistency poses a challenge for researchers aiming to utilize these models in downstream drug development tasks. Particularly in screening potential active compounds, providing a ranked list of candidates that likely interact with the target protein can guide scientists in prioritizing their experimental efforts. However, achieving this is difficult as each current DTI model can provide a different list based on its learned feature space. To address these issues, we propose EnsDTI, a Mixture-of-Experts architecture designed to enhance the performance of existing DTI models for more reliable drug-target interaction predictions. We integrate an inductive conformal predictor to provide confidence scores for each prediction, enabling EnsDTI to offer a reliable list of candidates for a specific target. Empirical evaluations on four benchmark datasets demonstrate that EnsDTI not only improves DTI prediction performance with an average accuracy improvement of 2.7% compared to the best performing baseline, but also offers a reliable ranked list of candidate drugs with the highest confidence, showcasing its potential for ranking potential active compounds in future applications. CCS CONCEPTS* Applied computing [->] Bioinformatics; * Computing methodologies [->] Artificial intelligence.

bioinformatics↗