Biology subjects

Wan, F.

Publications and source records attributed to Wan, F..

At least 19 recordsLinked to original sources

bioRxiv2026

Systematic mapping of the prokaryotic noncoding sequence space with a genomic language model

Prokaryotic genomes encode a vast repertoire of noncoding elements such as ribozymes, riboswitches, and CRISPR RNAs. The sequence-level diversity of these intergenic elements precludes alignment-based clustering that would enable annotation and discovery of novel noncoding RNA families. Here, we show that a genomic language model fine-tuned on prokaryotic intergenic sequences learns representations of noncoding elements from DNA sequence. We used these embeddings to organize 376 million intergenic regions from 1.3 terabase pairs of prokaryotic genomes and metagenomes into 5.14 million clusters. This approach nominated 20,149 clusters of putative structured RNAs with a candidate protein partner. Periodicity in token entropies revealed internal architecture, recovering CRISPR, TIGR and VIPR arrays together with their cognate effector proteins from sequence alone. Targeted clustering of bridge recombinase-adjacent sequences identified bridge RNA variants with longer guide recognition sequences than any characterized ortholog. Finally, we directed teams of AI agents to mine the database under researcher-defined criteria, uncovering divergent riboswitches, ribozymes, and other structured RNAs with biological hypotheses for their function. We release a publicly available database that transforms the dark matter of prokaryotic genomes into an explorable resource for human-agent molecular discovery.

genomics↗

bioRxiv2026

Ancient human mitochondrial genomes encode antimicrobial peptides

Mitochondria are bacteria-derived organelles that coordinate major innate immune pathways, but whether mitochondrial genomes themselves encode direct antimicrobial functions is underexplored. Here, we test the hypothesis that ancient human mitochondrial DNA is not only an evolutionary record, but also contains encrypted peptide sequences with the capacity to contribute to host defense. Because mitochondria descend from a bacterial endosymbiont and retain bacterial-like molecular features, we reasoned that their compact genomes might preserve sequence fragments capable of engaging bacterial-like membranes or bacterial physiology. We mined 2,025 ancient human mitochondrial genomes using a computational pipeline that couples ORF extraction and deep learning, identifying 65 candidate peptides that we term mitochondrins. We synthesized 38 candidates and experimentally validated 14 as antimicrobials against clinically relevant Gram-negative and Gram-positive bacteria. Active mitochondrins were not defined by length, charge, or helicity alone; instead, potency depended on precise hydrophobic-cationic patterning, with nested peptide families revealing how single motif-level changes can switch activity on or off. Mechanistic assays showed that mitochondrins span multiple antibacterial modalities, from strong membrane disruption to potent activity with limited membrane perturbation, suggesting noncanonical or multi-step killing mechanisms. Several peptides displayed low cytotoxicity toward human cells, and one representative mitochondrin reduced bacterial burden in a murine skin abscess model. These findings provide a biochemical basis for the hypothesis that mitochondrial genomes may contribute to innate immunity.

bioengineering↗

bioRxiv2026

Structural basis of substrate recognition and allosteric inhibition in human B0AT2

The SLC6 family is a major target for neuropsychiatric therapeutics. Human B0AT2 (SLC6A15) regulates amino acid homeostasis and glutamatergic transmission and is linked to major depressive disorder, yet its transport and inhibition mechanisms remain unclear. Here we report cryo-EM structures of B0AT2 in apo state and in complex with substrates (proline, leucine, methionine) and inhibitors (loratadine, tiagabine), capturing outward-open, early substrate-bound intermediate, outward-occluded, and inward-open conformations along the transport cycle. These structures reveal an early substrate-sensing mechanism at the substrate-binding pocket (S1), where a substrate-dependent rotameric switch of Phe308 remodels pocket geometry to tune substrate accommodation and selectivity. Loratadine stabilizes an outward-occluded state via allosteric inhibition at the extracellular S2 pocket, whereas tiagabine stabilizes the inward-open state through cooperative multi-site inhibition involving S1 and two previously unrecognized intracellular cavities (S3 and S4). Together with functional assays, these data define the molecular basis of B0AT2 substrate selectivity and state-dependent inhibition. Notably, the two intracellular cavities are conserved across SLC6 transporters, reflecting a shared intracellular vestibular architecture and enabling rational design of conformation-selective modulators for neuropsychiatric disorders.

biophysics↗

bioRxiv2026

A Conditional Variational Autoencoder with QSAR-Guided Surrogate-Weighted Fine-Tuning and Cross-Entropy Optimization for Targeted Antimicrobial Peptide Generation

Machine Learning frameworks have emerged as a promising tool for antimicrobial peptide design; however, generative models remain limited by two persistent problems: the limited availability of experimentally validated peptides and the circular dependency of the models. In this work we present a conditional variational autoencoder pipeline that addresses both limitations through a modular architecture that combines both binary and quantitative experimental data and implements a multimodal approach to externally guide the generation. A transformer-based encoder successfully generated a discriminative 64-dimensional latent space (test AUROC 0.968, F1 0.919) separating antimicrobial from non-antimicrobial sequences. This latent representation conditions a species-specific LoRA fine-tuned ProtGPT2 decoder through a scalar gating function, which generates balanced antimicrobial peptides through two different modes; prior and perturb, depending on their generation starting points. We introduced a Surrogate Weighted Fine-Tuning (SWF) ensemble to eliminate the circular dependency and a Cross-Entropy Method to explore and exploit the latent space, leading to successful antimicrobial peptide generation. The best candidates exhibited competitive physicochemical characteristics, a mean helical fraction of 0.874 (mean pLDDT 83.7), and externally predicted efficacy evaluated by APEX.

bioinformatics↗

bioRxiv2025

Integrative machine learning predicts activating kinase mutations for precision oncology

Kinases are enzymes that catalyze phosphorylation and play crucial roles in a myriad of cellular regulatory processes and hemostasis. Patient-specific genetic mutations that aberrantly activate kinases can profoundly influence cancer progression and alter drug efficacy. Predicting the impact of such missense mutations across the human kinome on protein function and cellular signaling is therefore a critical step toward personalized targeted therapy. Here, we present Kinome-AI, an integrative machine learning framework that classifies kinase missense mutations as activating or non-activating. Kinome-AI is trained on a rich multi-modal feature set, including residue-level biochemical changes, sequence embeddings from a protein language model, and structural descriptors of kinase-ATP-substrate complexes derived from molecular modeling. Notably, detailed structural features were available for only 21% of mutants; we leverage these as privileged information during training to impute missing structural data for the remaining [~]79. This strategy boosts performance without requiring structural inputs for new (unseen) mutations. The resulting classifier achieves an area under the receiver operating characteristic curve (AUROC) of 0.85 and a balanced accuracy (BACC) of 0.76 across 1,003 mutations spanning 110 different kinases --substantially outperforming existing bioinformatics and general-purpose variant effect predictors. This work provides a robust approach to quantify sequence-structure- function relationships of cancer-driving kinase mutations, paving the way for improved personalized cancer treatment. Significance StatementIn cancer patients, numerous mutations in diverse protein kinases lead to marked differences in disease progression and drug response. Identifying which kinase mutations are activating in individual patients is therefore critical for precision oncology. Drawing inspiration from teacher- student (privileged information) learning, we developed a deep learning framework that integrates structural features from molecular simulations with sequence embeddings from protein language models. This approach enables accurate binary classification of the activation status of kinase mutations. Our study demonstrates how data-driven algorithms can leverage accumulated sequence and structural knowledge of known mutations to predict the effects of novel variants a priori. The model, termed Kinome-AI, shows significant promise for incorporation into personalized cancer therapy decision pipelines.

cancer biology↗

bioRxiv2025

Aphid-derived cross-kingdom RNA dynamics underpin maize resistance

Aphids are major pests of maize (Zea mays), yet the molecular mechanisms underlying their interactions with host plants remain poorly understood. Here, we identified and functionally characterized long non-coding RNAs (lncRNAs) from the aphid-specific Ya gene family in the cereal-specialist aphids Rhopalosiphum maidis and R. padi. We showed that Ya genes formed lineage-specific clusters and were transcriptionally active across both species. Multiple Ya transcripts secreted into maize tissues exhibited remarkable stability compared with rapidly degraded aphid mRNAs and migrated systemically, as visualized in planta using an RNA switch-controlled RNA-triggered fluorescence system. RNA interference of Ya genes significantly reduced aphid fecundity, while ectopic expression of Ya lncRNAs in maize enhanced aphid colonization. Importantly, maize genotypes differing in aphid resistance selectively influenced the persistence of aphid mRNAs but not Ya lncRNAs, indicating a decoupled fate of distinct aphid RNA classes in planta. These findings establish Ya lncRNAs as cross-kingdom effectors critical for aphid virulence, and suggest new molecular markers for breeding aphid-resistant maize.

plant biology↗

bioRxiv2025

Multi-omics analysis reveals genetic regulatory mechanismsof muscle traits in wild and domesticated Bactrian camels

Bactrian camels, large domesticated animals adapted to extreme environments, have evolved under natural and artificial selection to suit their living conditions and meet human needs. In arid and semi-arid regions, they provide valuable resources through economically important traits such as meat, milk, and wool production, contributing significantly to both economy and ecology. However, research on gene function, adaptive evolution, and molecular mechanisms underlying complex traits in Bactrian camels remains limited, constraining the full utilization of their breeding potential and conservation efforts.To address this gap, we performed whole-genome resequencing (WGS) on 35 domesticated Bactrian camels from the Chinese Sunite local population and integrated this data with publicly available WGS data from 112 Asian Bactrian camels. Our comprehensive analysis revealed significant genetic differences and genomic features between wild and domesticated Bactrian camels. We further integrated 27 epigenomic (ATAC-seq) and transcriptomic (RNA-seq) datasets from eight biologically important tissues of four Bactrian camels, along with skeletal muscle scRNA-seq and scATAC data. This allowed us to identify tissue-specific genes, regulatory elements, and genes under positive selection.Our integrated analysis identified target genes and their upstream regulatory elements, including potential regulatory transcription factors, with a focus on positively selected regions in Bactrian camel muscle tissue. We constructed regulatory pathways from key SNPs to regulatory elements to genes, and for the first time, identified regulatory elements in major tissues of the Bactrian camel genome. Our results also summarized the single-cell transcriptional landscape of Bactrian camel skeletal muscle, revealing cellular heterogeneity and genetic expression patterns across two developmental stages.This comprehensive regulatory element map provides a valuable resource for genetic and genomic studies in Bactrian camels. It reveals key regulatory networks, important genes, and regulatory elements influencing muscle development, as well as potentially significant functional variants affecting meat production traits. These findings will provide crucial data support for future genetic improvement and breeding efforts in Bactrian camels.

genetics↗

bioRxiv2025

Lactate metabolism antagonizes antiviral innate immunity bypromoting VDAC1-lactylation-mediated mtDNA release

Foot-and-mouth disease (FMD) is an acute, highly contagious disease caused by the foot-and-mouth disease virus (FMDV). Due to its zoonotic nature, this disease not only significantly impacts livestock productivity but also poses a serious threat to human health. FMDV exhibits distinct host specificity, mainly infecting cloven-hoofed animals such as cattle, sheep, pigs, and deer, while perissodactyl animals like horses and donkeys show natural resistance. Elucidating the interspecies differences in response to FMDV infection is essential for understanding the pathogenesis of the virus. In this study, we report that FMDV infection in horse epithelial cells induces the production of lactate, which modulates the function of VDAC1 protein through lactylation, facilitating the release of mtDNA from mitochondria into the cytoplasm. This process activates the cGAS-STING signaling pathway, which stimulates interferon-{beta} expression and suppresses FMDV replication. Our findings provide new insights into the mechanism behind horse resistance to FMDV and lay a theoretical foundation for further research concerning the differential pathogenesis of FMDV infection across species.

cell biology↗

bioRxiv2025

In situ generated vaccine-like pyroptosome for personalized cancer immunotherapy

The efficacy of in situ cancer vaccination has been hampered by poor spatiotemporal orchestration of multiple key steps of cancer-immunity cycle in most tumours and systemic toxicity related to therapeutic strategies. Here, we report a systemic injectable and pyroptosis-enabled nanoadjuvant (SPEN) that precisely evokes the secretion of vaccine-like pyroptosome in tumour area for eliciting robust anti-tumour immunity. SPEN induces vigorous immunogenic pyroptosis, triggering the efficient release of tumour antigen-rich pyroptosomes, DAMPs, and proinflammatory cytokines. Upon the activatable release of a TLR7/8 agonist into pyroptosome, the generated pyroptosome functions as in situ cancer vaccine for cooperatively activating the cancer-immunity cycle while avoiding systemic toxicity. This in situ vaccine boosts both innate and adaptive immune response, facilitating the eradication of primary tumour and long-lasting cancer prevention. Our findings provide new insights into the rational design of pyroptosis-inducing nanomedicines for boosting the cancer-immunity cycle, thus advancing personalized cancer immunotherapy.

cancer biology↗

bioRxiv2025

Generative latent diffusion language modeling yields anti-infective synthetic peptides

Generative artificial intelligence (AI) offers a powerful avenue for peptide design, yet this process remains challenging due to the vast sequence space, complex structure-activity relationships, and the need to balance antimicrobial potency with low toxicity. Traditional approaches often rely on trial-and-error screening and fail to efficiently navigate the immense diversity of potential sequences. Here, we introduce AMP-Diffusion, a novel latent diffusion model fine-tuned on antimicrobial peptide (AMP) sequences using embeddings from protein language models. By systematically exploring sequence space, AMP-Diffusion enables the rapid discovery of promising antibiotic candidates. We generated 50,000 candidate sequences, which were subsequently filtered and ranked using our APEX predictor model. From these, 46 top candidates were synthesized and experimentally validated. The resulting AMP-Diffusion peptides demonstrated broad-spectrum antibacterial activity, targeting clinically relevant pathogens--including multidrug-resistant strains--while exhibiting low cytotoxicity in human cell assays. Mechanistic studies revealed bacterial killing via membrane permeabilization and depolarization, and the peptides showed favorable physicochemical profiles. In preclinical mouse models of infection, lead peptides effectively reduced bacterial burdens, displaying efficacy comparable to polymyxin B and levofloxacin, with no detectable adverse effects. This study highlights the potential of AMP-Diffusion as a robust generative platform for designing novel antibiotics and bioactive peptides, offering a promising strategy to address the escalating challenge of antimicrobial resistance.

microbiology↗

bioRxiv2025

Improving functional protein generation via foundation model-derived latent space likelihood optimization

A variety of deep generative models have been adopted to perform de novo functional protein generation. Compared to 3D protein design, sequence-based generation methods, which aim to generate amino acid sequences with desired functions, remain a major approach for functional protein generation due to the abundance and quality of protein sequence data, as well as the relatively low modeling complexity for training. Although these models are typically trained to match protein sequences from the training data, exact matching of every amino acid is not always essential. Certain amino acid changes (e.g., mismatches, insertions, and deletions) may not necessarily lead to functional changes. This suggests that maximizing the training data likelihood beyond the amino acid sequence space could yield better generative models. Pre-trained protein large language models (PLMs) like ESM2 can encode protein sequences into a latent space, potentially serving as functional validators. We propose training functional protein sequence generative models by simultaneously optimizing the likelihood of training data in both the amino acid sequence space and the latent space derived from a PLM. This training scheme can also be viewed as a knowledge distillation approach that dynamically re-weights samples during training. We applied our method to train GPT- like models (i.e., autoregressive transformers) for antimicrobial peptide (AMP) and malate dehydrogenase (MDH) generation tasks. Computational experiments confirmed that our method outperformed various deep generative models (e.g., generative adversarial net, variational autoencoder, and GPT model without the proposed training strategy) on these tasks, demonstrating the effectiveness of our multi-likelihood optimization strategy.

bioinformatics↗

bioRxiv2024

Design of multimodal antibiotics against intracellular infections using deep learning

The rise of antimicrobial resistance has rendered many treatments ineffective, posing serious public health challenges. Intracellular infections are particularly difficult to treat since conventional antibiotics fail to neutralize pathogens hidden within human cells. However, designing molecules that penetrate human cells with preserved antimicrobial activity has historically been a major challenge. Here, we introduce ApexDuo, a multimodal artificial intelligence (AI) model for generating peptides with both cell- penetrating and antimicrobial properties. From a library of 50 million AI-generated compounds, we selected and characterized several candidates. Our lead, Turingcin-46, penetrated mammalian cells and reduced intracellular Staphylococcus aureus and Listeria monocytogenes. In mouse models of skin abscess and peritonitis, Turingcin-46 reduced bacterial loads by up to two orders of magnitude. In sum, ApexDuo generated multimodal peptide antibiotics, opening new avenues for designing molecules with more than one function.

synthetic biology↗

bioRxiv2024

A generative artificial intelligence approach for antibiotic optimization

Antimicrobial resistance (AMR) poses a critical global health threat, underscoring the urgent need for innovative antibiotic discovery strategies. While recent advances in peptide design have yielded numerous antimicrobial agents, optimizing these molecules experimentally remains challenging due to unpredictable and resource-intensive trial-and-error approaches. Here, we introduce APEX Generative Optimization (APEXGO), a generative artificial intelligence (AI) framework that integrates a transformer-based variational autoencoder with Bayesian optimization to design and optimize antimicrobial peptides. Unlike traditional supervised learning approaches that screen fixed databases of existing molecules, APEXGO generates entirely novel peptide sequences through arbitrary modifications of template peptides, representing a paradigm shift in peptide design and antibiotic discovery. Our framework introduces a new peptide variational autoencoder with design and diversity constraints to maintain similarity to specific templates while enabling sequence innovation. This work represents the first in vitro and in vivo experimental validation of generative Bayesian optimization in any setting. Using ten de-extinct peptides as templates, APEXGO generated optimized derivatives with enhanced antimicrobial properties. We synthesized 100 of these optimized peptides and conducted comprehensive in vitro characterizations, including assessments of antimicrobial activity, mechanism of action, secondary structure, and cytotoxicity. Notably, APEXGO achieved an outstanding 85% ground-truth experimental hit rate and a 72% success rate in enhancing antimicrobial activity against clinically relevant Gram-negative pathogens, outperforming previously reported methods for antibiotic discovery and optimization. In preclinical mouse models of Acinetobacter baumannii infection, several AI-optimized molecules--most notably derivatives of mammuthusin-3 and mylodonin-2--exhibited potent anti-infective activity comparable to or exceeding that of polymyxin B, a widely used last-resort antibiotic. These findings highlight the potential of APEXGO as a novel generative AI approach for peptide design and antibiotic optimization, offering a powerful tool to accelerate antibiotic discovery and address the escalating challenge of AMR.

microbiology↗

bioRxiv2024

Discovery of antibiotics in the archaeome using deep learning

Antimicrobial resistance (AMR) is one of the greatest threats facing humanity, making the need for new antibiotics more critical than ever. While most antibiotics have traditionally been derived from bacteria and fungi, archaea--a distinct and underexplored domain of life--offer a largely untapped reservoir for antibiotic discovery. In this study, we leveraged deep learning to systematically explore the archaeome, uncovering promising new candidates for combating AMR. By mining 233 archaeal proteomes, we identified 12,623 molecules with potential antimicrobial activity. These newly discovered peptide compounds, termed archaeasins, exhibit unique compositional features that differentiate them from traditional antimicrobial peptides, including a distinct amino acid profile. We synthesized 80 archaeasins, 93% of which demonstrated antimicrobial activity in vitro. Notably, in vivo validation identified archaeasin-73 as a lead candidate, significantly reducing bacterial loads in mouse infection models, with effectiveness comparable to established antibiotics like polymyxin B. Our findings highlight the immense potential of archaea as a resource for developing next-generation antibiotics.

microbiology↗

bioRxiv2023

Molecular de-extinction of antibiotics enabled by deep learning

Molecular de-extinction is an emerging field that aims to resurrect molecules to solve present-day problems such as antibiotic resistance. Here, we introduce a deep learning approach called Antibiotic Peptide de-Extinction (APEX) to mine the proteomes of all available extinct organisms (the "extinctome") searching for encrypted peptide (EP) antibiotics. APEX mined a total of 10,311,899 EPs and identified 37,176 sequences predicted to have broad-spectrum antimicrobial activity, 11,035 of which were not found in extant organisms. Chemical synthesis and experimental validation yielded archaic EPs (AEPs) with activity against dangerous bacterial pathogens. Most peptides killed bacteria by depolarizing their cytoplasmic membrane, contrary to known antimicrobial peptides, which target the outer membrane. Notably, lead peptides, including those derived from the woolly mammoth, ancient sea cow, giant sloth, and extinct giant elk, exhibited anti-infective activity in preclinical mouse models. We propose molecular de-extinction, accelerated by deep learning, as a framework for discovering therapeutic molecules.

bioengineering↗

bioRxiv2023

A novel bacterial signal transduction system specifically senses and responds to peptidoglycan damage

The peptidoglycan (PG) layer is a mesh-like structure within the cell envelope essential for maintenance of cell shape and resistance to osmotic stress, and therefore is a primary target of many important and widely used antibiotics, such as {beta}-lactams. In Gram-negative bacteria, while signal transduction systems that monitor the state of the inner- and outer-membranes have been extensively studied and well understood, much less is known about how cells sense and respond to PG-targeting stresses. Here we show that a novel bacterial two-component system (PghKR) from Shewanella oneidensis is capable of sensing and responding to PG damage. This system is specifically activated in cells exposed to various PG-targeting antibiotics or carrying a defect in PG synthesis, resulting in induced expression of blaA and relV, which encode a {beta}-lactamase conferring resistance to {beta}-lactams and a small ppGpp synthetase responsible for antibiotic tolerance, respectively. Intriguingly, the PghKR homologs are widespread among several classes of Proteobacteria and the periplasmic domain of sensor kinase PghK contains a family 9 carbohydrate-binding module that is required for signal perception, implying that the signals could be the glycan fragments of PG. Overall, our results provide critical insights into the regulation of PG homeostasis in Gram-negative bacteria.

microbiology↗

bioRxiv2022

Adipokine C1q/Tumor necrosis factor-related protein 3 (CTRP3) Attenuates Intestinal Inflammation via Histone Deacetylase Sirtuin 1 (SIRT1)/NF-κB Signaling

BACKGROUND & AIMSThe adipokine C1q/tumor necrosis factor-related protein 3 (CTRP3) has anti-inflammatory effects in several non-intestinal disorders. Although CTRP3 is reduced in the serum of patients with inflammatory bowel disease (IBD), its function in IBD has not been established. We aimed to elucidate the function of CTRP3 and related molecular mechanisms in intestinal inflammation using a colitis model of genetically-modified CTRP3 mice and intestinal epithelial tissue from patients with Crohns disease (CD), one of the two main forms of IBD. METHODSCTRP3 knockout (KO) and overexpressing transgenic (Tg) mice along with their corresponding wild-type (WT) littermates were subjected to drinking water containing dextran sulfate sodium (DSS) for 6-10 days to induce acute colitis. Mouse colitis symptoms and histological data were analyzed. CTRP3-mediated signaling was examined in the intestinal tissue of mice and patients with CD. RESULTSCTRP3 mRNA and protein were detected in murine and human intestinal epithelial cells, as well as in murine intestinal smooth muscle cells and mesenteric fat. In DSS-induced acute colitis models, CTRP3 KO mice developed more severe colitis than their WT littermates, while CTRP3 overexpressing Tg mice developed less severe colitis than their WT littermates. In both water- and DSS-treated CTRP3 KO mice, reduced CTRP3 levels correlated with decreased levels of Sirtuin 1 (SIRT1), a histone deacetylase, increased levels of phosphorylated nuclear factor kappa B (NF-{kappa}B) subunit p65, resulting in increased expression of pro-inflammatory cytokines tumor necrosis factor- (TNF-) and interleukin 6 (IL-6). The results from CTRP3 Tg mice mirrored those from CTRP3 KO mice in most respects. This CTRP3/SIRT1/NF-{kappa}B relationship was also observed in the intestinal epithelial tissue of patients with active and inactive CD. CONCLUSIONSCTRP3 expression levels correlate negatively with intestinal inflammation in mouse colitis models and CD patients. CTRP3 attenuates intestinal inflammation via SIRT1/NF-{kappa}B signaling to suppress pro-inflammatory cytokines in mouse colitis models and patients with IBD. The manipulation of CTRP3 signaling, including through the use of SIRT1 agonists, may offer translational potential in the treatment of IBD. WHAT YOU NEED TO KNOWO_ST_ABSBACKGROUND AND CONTEXTC_ST_ABSAdipokine C1q/tumor necrosis factor-related protein 3 (CTRP3) is a novel adipokine with known non-intestinal anti-inflammatory effects. CTRP3 is reduced in the serum of patients with inflammatory bowel disease (IBD). However, little is known about whether and how CTRP3 influences intestinal inflammation in IBD. NEW FINDINGSCTRP3 mRNA and protein were detected in murine and human intestinal epithelial cells, as well as in murine intestinal smooth muscle cells and mesenteric fat. CTRP3 deletion was associated with more severe acute dextran sulfate sodium (DSS)-induced colitis, while CTRP3 overexpression was associated with less severe colitis. In both mice and humans, reduced CTRP3 levels correlated with reduced levels of the histone deacetylase Sirtuin 1 (SIRT1), resulting in the up-regulation of phosphorylated nuclear factor-kappa B (NF-{kappa}B) p65 and pro-inflammatory cytokine synthesis. LIMITATIONSThis study was performed using genetically modified mice and human tissue samples. An acute DSS-induced colitis model was used; additional mouse colitis models designed to mimic other aspects of IBD will be examined in future studies. The specific source of the secreted CTRP3 protein which influences intestinal inflammation is yet to be identified. The use of recombinant CTRP3 protein supplementation and SIRT1 agonists to mitigate intestinal inflammation also requires further study. IMPACTCTRP3 is a novel anti-inflammatory adipokine that attenuates intestinal inflammation in colitis mouse models and intestinal epithelial tissue of patients with IBD. CTRP3 attenuates intestinal inflammation by activating SIRT1, which suppresses the pro-inflammatory transcriptional activity of phosphorylated NF-{kappa}B p65. CTRP3 and SIRT1 agonists have potential as novel IBD drug targets.

pathology↗

bioRxiv2021

MoTSE: an interpretable task similarity estimator for small molecular property prediction tasks

Understanding the molecular properties (e.g., physical, chemical or physiological characteristics and biological activities) of small molecules plays essential roles in biomedical researches. The accumulating amount of datasets has enabled the development of data-driven computational methods, especially the machine learning based methods, to address the molecular property prediction tasks. Due to the high cost of obtaining experimental labels, the datasets of individual tasks generally contain limited amount of data, which inspired the application of transfer learning to boost the performance of the molecular property prediction tasks. Our analyses revealed that simultaneously considering similar tasks, rather than randomly chosen ones, can significantly improve the performance of transfer learning in this field. To provide accurate estimation of task similarity, we proposed an effective and interpretable computational tool, named Molecular Tasks Similarity Estimator (MoTSE). By extracting task-related local and global knowledge from pretrained graph neural networks (GNNs), MoTSE projects individual tasks into a latent space and measures the distance between the embedded vectors to derive the task similarity estimation and thus enhance the molecular prediction results. We have validated that the task similarity estimated by MoTSE can serve as a useful guidance to design a more accurate transfer learning strategy for molecular property prediction. Experimental results showed that such a strategy greatly outperformed baseline methods including training from scratch and multitask learning. Moreover, MoTSE can provide interpretability for the estimated task similarity, through visualizing the important loci in the molecules attributed by the attribution method employed in MoTSE. In summary, MoTSE can provide an accurate method for estimating the molecular property task similarity for effective transfer learning, with good interpretability for the learned chemical or biological insights underlying the intrinsic principles of the task similarity.

bioinformatics↗