Bay Institute

Spatial transcriptomics (ST) enables joint profiling of gene expression and spatial positions, thereby revealing spatially resolved biological functions. However, many existing ST analysis methods often fail to explicitly quantify the belief and uncertainty in decisions caused by noisy ST data, making it difficult to handle spots of varying quality in a fine-grained manner. In addition, domain identification is a fundamental and critical task in ST, but commonly used models that separate expression learning and clustering often struggle to learn cluster-friendly latent representations effectively. To address these issues, we propose PREST, a prototype-based evidence-aware integration framework for ST data. PREST performs multi-scale representation learning with fine-grained attention fusion and introduces learnable class prototypes to quantify belief and uncertainty in model decisions. We aim to align overall belief scores with latent semantic information to enhance uncertainty quantification and prototype learning, thereby promoting the learning of clustering-friendly representations. PREST further integrates an uncertainty-aware reconstruction module and spatial regularization to reduce overfitting to unreliable spots and promote denoised, discriminative representations. Extensive experiments on several benchmark datasets validate the effectiveness and superiority of our proposed PREST across various downstream tasks.

Can Multimodal Large Language Models (MLLMs) discern confused objects that are visually present but audio-absent? To study this, we introduce a new benchmark, AV-ConfuseBench, which simulates an “Audio-Visual Confusion” scene by modifying the corresponding sound of an object in the video, e.g., mute the sounding object and ask MLLMs “Is there a/an {muted-object} sound”. Experimental results reveal that MLLMs, such as Qwen2.5-Omni and Gemini 2.5, struggle to discriminate non-existent audio due to visually dominated reasoning. Motivated by this observation, we introduce RL-CoMM, a Reinforcement Learning-based Collaborative Multi-MLLM that is built upon the Qwen2.5-Omni foundation. RL-CoMM includes two stages: 1) To alleviate visually dominated ambiguities, we introduce an external model, a Large Audio Language Model (LALM), as the reference model to generate audio-only reasoning. Then, we design a Step-wise Reasoning Reward function that enables MLLMs to self-improve audio-visual reasoning with the audio-only reference. 2) To ensure an accurate answer prediction, we introduce Answer-centered Confidence Optimization to reduce the uncertainty of potential heterogeneous reasoning differences. Extensive experiments on audio-visual question answering and audio-visual hallucination show that RL-CoMM improves accuracy by 10~30% over the baseline model with limited training data.

Stentor, the genus of large trumpet-shaped ciliates, is well-known for its complex morphology and striking behaviors. Members of this genus are distributed throughout the world in a wide and diverse pool of freshwater ecosystems. Recently, the molecular phylogeny of Stentor has been explored through comparison of 18 S small subunit (SSU) ribosomal DNA (rDNA) sequences, clarifying several previously mischaracterized species and species complexes. However, despite their wide distribution, to-date, only about a dozen species of Stentor have been described and verified by phylogenetic means. Here, we introduce the discovery of a new species within genus Stentor: Stentor stipatus spec. nov., so named for their distinctive cytosolic dark pigmented granules which surround the macronucleus and are also present cortically alongside cortically-distributed green microalgae. We present morphological, phylogenetic, ecological, and behavioral characterizations of these cells. Phylogenetic analysis of S. stipatus spec. nov. by comparison of SSU rDNA sequence suggests it is a distinct species from its closest relative, S. amethystinus. We demonstrate that S. stipatus spec. nov. is capable of habituation in response to repeated mechanical stimulation. Further, S. stipatus spec. nov. exhibits strongly directed positive phototaxis, like its relative S. pyriformis, but with a distinct action spectrum from both S. coeruleus and S. pyriformis. Finally, S. stipatus phototaxis response strength varies in a consistent pattern throughout the day, providing evidence of potential circadian regulation. This work expands the current understanding of the ecological distribution of and behavioral features present within genus Stentor.

The homeobox transcription factor (TF) superfamily includes the WUSCHEL-RELATED HOMEOBOX (WOX) family, which plays a critical role in adaptive plant growth. Specifically, WOX regulates stem growth in plants, with stems serving as the structural framework for laticifers in Euphorbia hirta. However, the number of WOX gene family members in the E. hirta genome has not been reported. In this study, we identified 14 EhWOX genes in E. hirta and characterized their physicochemical properties, chromosomal locations, phylogenetic relationships, conserved motifs, gene structures, promoter cis elements, gene ontology (GO) enrichment, tissue-specific expression patterns, and subcellular localization. Chromosomal mapping indicated their distribution across nine chromosomes. Phylogenetic analysis classified these genes into three evolutionary clades. Promoter cis-element analysis identified abundant light-responsive, hormone-responsive, and stress-responsive elements. GO enrichment suggested their broad involvement in diverse biological processes. Additionally, RNA-seq revealed high expression levels of EhWOX4-6 and EhWOX14 in stems. Furthermore, RT-qPCR confirmed tissue-specific expression in stems. Moreover, experimental evidence confirmed the subcellular localization and autoactivation capability of some WOX proteins that may be involved in stem development. Overall, this study provides a comprehensive characterization of the candidate EhWOX genes and provides a foundational resource for future functional investigations into their possible roles in stem and laticifer biology.

Rice yield-related traits are regulated by numerous quantitative trait loci (QTLs). Pyramiding these QTLs offers a promising strategy for crop yield improvement; however, this approach is constrained by the limited understanding of the interactions among these QTLs. In this study, we demonstrated that the QTL DTH8 significantly increases branch numbers, panicle length, and number of total grains per panicle in rice. Notably, the DTH89311 allele exhibited a more pronounced effect on enhancing yield-related traits compared to DTH8Nip. Furthermore, QTL pyramiding involving SD1 and DTH8 indicated that the regulation of panicle architecture by DTH8 partially depends on SD1. Genetic analysis revealed an interaction between these two QTL/genes with respect to panicle-related traits. Dual-luciferase and ChIP-PCR analyses suggested that DTH8 might regulate the expression of SD1, thereby governing branch numbers, panicle length, and number of total grains per panicle. Additionally, our findings highlight the importance of avoiding the simultaneous pyramiding of sd1 and DTH89311 in practical breeding programs due to their genetic and molecular interactions.

Methyl farnesoate (MF) is a sesquiterpenoid hormone that controls a variety of physiological processes in crustaceans, including morphogenesis, development, reproduction, and molting. MF action is mediated by a transcriptional signaling cascade consisting of Methoprene-tolerant (Met), Steroid receptor coactivator (Src), Krüppel homolog 1 (Kr-h1), and Ecdysone response gene 93 (E93) transcription factors (TFs), and transcriptional co-regulators CREB-binding protein (CBP) and C-terminal-binding protein (CtBP). Phylogenetic and sequence analyses revealed that these genes were highly conserved across pancrustacean species. Met and Src were characterized as basic helix-loop-helix, Period (Per)-Aryl Hydrocarbon Nuclear Translocator (ARNT)-Single-minded (Sim) protein (bHLH-PAS) TFs; Kr-h1 was characterized as a C2H2 zinc finger TF with seven zinc finger motifs; E93 was characterized as a helix-turn-helix, pipsqueak (HTH_Psq) TF. CBP was identified by several zinc finger-binding regions with Transcription Adaptor Zinc Finger 1 and 2, Really Interesting New Gene, Plant homeodomain, and Z-type zinc finger domains; the Kinase-inducible Domain Interacting-transcription factor docking site; the Bromodomain-acetylated lysine recognition and binding site; the histone acetyltransferase domain; and a C-terminal CREB-binding region containing a nuclear receptor co-activator-binding domain. CtBP had a dehydrogenase domain with arginine-glutamate-histidine catalytic triad. 81 Met contigs, 45 Src contigs, 136 Kr-h1 contigs, 66 E93 contigs, 60 CBP contigs, and 172 CtBP contigs were identified across pancrustacean taxa, including decapod crustaceans. Bioinformatic identification and annotation of these TFs and co-regulators in brachyuran Y-organ (YO) transcriptomes suggests that MF signaling influences YO ecdysteroidogenesis; functional tests in the YO are needed to establish causality.

Abstract Background Endoscopic activity indices for ulcerative colitis (UC) demonstrate limited accuracy, particularly in acute severe UC (ASUC). Artificial intelligence-based scoring (AI) may enhance this. We investigated whether the DovaVision UC AI tool could detect differences in endoscopic response rates between personalised and standard infliximab (IFX) dosing in ASUC patients enrolled in the prospective randomised TITRATE trial, that were previously not detected with human assessment. Methods In TITRATE, adult IFX-naive steroid-refractory ASUC patients were randomised 1:1 to standard dosing (SD) or personalised dosing (PD) with IFX. After an initial 5 mg/kg IFX infusion, patients in the SD group received 5 mg/kg IFX at week 2 and 6. In the PD arm, additional 5 mg/kg IFX infusions were administered guided by a Bayesian pharmacokinetic algorithm (iDose™) aiming to achieve prospectively defined target IFX serum concentrations until day 42. Endoscopies were recorded at baseline and at week 6. Definitions of endoscopic response were ≥2-point improvement in UC endoscopic index of severity (UCEIS) and ≥1-point improvement in Mayo endoscopic subscore (MES) at week 6 compared to baseline. Patients with missing endoscopic data were classified as non-responders. Endoscopic activity was previously scored on blinded videos by 2 independent human experts, with a third for adjudication (1). Here, we applied the DovaVision UC AI tool to the blinded videos. This scored the UCEIS and MES on individual frames and aggregated results to yield final procedure-level scores. Results In total, 48 patients were randomised to PD (n = 23) or SD (n = 25) (Table 1). Human UCEIS or MES assessments could not distinguish differences in response rates (UCEIS: 13 (56.5%) in the PD and 11 (44.0%) in the SD group (p = 0.540); MES: 14 (60.8%) in PD and 10 (40.0%) in SD group (p = 0.250)) (1).Using AI-assessed UCEIS, 27/48 (56.2%) patients achieved endoscopic response, comprising 18/23 (78.3%) in the PD and 9/25 (36.0%) in the SD group (p = 0.004). Using AI-assessed MES, 24/48 patients (50.0%) responded (16/23 (69.6%) in PD and 8/25 (32.0%) in SD group (p = 0.020)) (Figure 1). Conclusion In ASUC patients treated with IFX, the AI tool was able to detect endoscopic response at 6 weeks of IFX treatment, regardless of the endoscopic activity index that was used. The AI tool could also distinguish response rate differences between personalised and standard dosing of IFX. In the original study, human UCEIS or MES assessments could not distinguish such differences. This is the first IBD study in which AI was able to distinguish treatment effects between different treatment strategies, demonstrating its potential to fundamentally improve clinical trial methodology.

This study examines the paradox of Nigerianisation in Nigeria’s oil and private sectors, where indigenous ownership and local content policies were expected to strengthen national capacity, improve worker welfare, and align employment conditions with international best practice. Drawing on the ILO decent work framework and a mixed-methods design, the study compares wages, welfare packages, and employment security across Nigerian-owned oil firms, IOC/IOC-linked service companies, and Nigerian-owned private enterprises, supported by interviews with workers, HR managers, and union officials. The Findings indicate a marked erosion of welfare provisions, especially medical insurance, pensions, and workers allowances, alongside wage suppression, graduate devaluation, and widespread casualization through short-term and outsourced contracts. The study further highlights a widening gap between executive wealth accumulation and workforce vulnerability, deepening class fragmentation and weakening social cohesion. In many cases, employment in high-revenue sectors functions as underemployment: workers remain in precarious jobs due to unemployment pressures rather than decent work incentives. The study concludes that Nigerianisation has often produced ownership transfer without labour responsibility. It recommends redefining Nigerianisation metrics to include enforceable welfare standards, stronger labour regulation, and incentive structures tied to decent work compliance, to restore dignity, productivity, and inclusive development.

Abstract This paper explores the lack of coordination on drug clinical trials in the Greater Bay Area, encompassing Guangdong, Hong Kong, and Macao. Despite growing regional cooperation in the field of biomedicine and biopharmaceuticals, coordinated drug clinical trial governance across the three jurisdictions remains largely disjointed, with separate approval systems, ethics review processes, and deeper obstacles such as compliance burdens and talent-related issues. Drawing lessons from the European Union’s experience, the paper argues that harmonization and simplification of regulatory standards can enhance coordination efficiency and effectiveness. It recommends an approach in which the three regions develop a mutually agreeable framework while aligning with top-level design and strategic priorities set by Mainland, as well as international standards. This coordinated model could pave the way for the Greater Bay Area to become a leading hub for internationally recognized high-quality clinical research.