Patterns最新文献

Microalgal genomes contain a vast "dark proteome"-sequences lacking detectable homology that evade conventional classification tools. We developed LA⁴SR (language modeling with AI for algal amino acid sequence representation), a framework using transformer- and state-space models to classify translated ORFeomes across ten algal phyla. Training on ∼77 million sequences, LA⁴SR achieves near-complete recall, accelerates classification by ∼10,701× relative to BLASTP⁺, and generalizes robustly to unseen sequences using less than 2% of available data. Models trained on synthetic, chimeric (terminal information [TI]-free) sequences maintained high accuracy, demonstrating that internal sequence features alone can drive robust classification. Inference speed and scalability were further enhanced under TI-free settings, supporting rapid annotation of large proteomic datasets. Custom explainability tools revealed interpretable amino acid patterns linked to evolutionary and biophysical features. Designed for accessibility across disciplines, LA⁴SR integrates biological context and computational innovation in parallel, enabling both biologists and data scientists to interrogate the microbial dark proteome.

Fully automated self-driving laboratories promise high-throughput, large-scale scientific discovery by reducing repetitive labor. However, they require deep integration of laboratory knowledge, which is often unstructured, multimodal, and hard to incorporate into current AI systems. This paper introduces the "k-agents" framework, designed to support experimentalists in organizing laboratory knowledge and automating experiments with agents. The framework uses large-language-model-based agents to encapsulate laboratory knowledge, including available operations and methods for analyzing results. To automate experiments, execution agents break multistep procedures into agent-based state machines, interact with other agents to execute steps, and analyze results. These results drive state transitions, enabling closed-loop feedback control. We demonstrate the system on a superconducting quantum processor, where agents autonomously planned and executed experiments for hours, successfully producing and characterizing entangled quantum states at human-level performance. Our knowledge-based agent system opens new possibilities for managing laboratory knowledge and accelerating scientific discovery.

The structure of the protein binding pocket governs the ligand binding affinity by providing crucial intermolecular interactions and spatial compatibility. While existing methods have leveraged these structural insights to advance affinity prediction, they often treat virtual screening and hit-to-lead optimization separately, mainly due to incompatible speed-accuracy requirements. However, these two tasks complement each other, and their integration enables broader chemical exploration while preserving focus on affinity-determining substructures. Here, we present ligand unified affinity (LigUnity), a foundation model for affinity prediction that jointly embeds ligands and pockets into a shared space. In particular, LigUnity learns coarse-grained active/inactive distinction through scaffold discrimination and fine-grained pocket-specific ligand preference through pharmacophore ranking. We demonstrate the effectiveness and versatility of LigUnity on eight benchmarks across six settings. In virtual screening, LigUnity outperforms 24 methods with >50% improvement and demonstrates robust generalization to novel targets. In hit-to-lead optimization, it achieves state-of-the-art performance across split-by-time, split-by-scaffold, and split-by-unit settings, emerging as a cost-efficient alternative to free energy perturbation. We further showcase how LigUnity can be employed in an active learning framework for tyrosine kinase 2 (TYK2) to efficiently find optimal ligands. Collectively, these results establish LigUnity as a versatile foundation model for affinity prediction, offering broad applicability across the drug discovery pipeline.

Edith Luhanga is an assistant research professor at Carnegie Mellon University Africa in Rwanda. Her work focuses on designing and evaluating technologies that leverage artificial intelligence (AI) to promote behavioral change in low-resource communities. She is currently working on digital interventions for maternal health, child nutrition and online safety, and financial inclusion in Rwanda, Tanzania, Kenya, and South Africa. Edith holds a PhD in information science (ubiquitous computing) from the Nara Institute of Science and Technology in Japan and an MSc in advanced computing science and BEng (Hons) in electronic and computer engineering from the University of Nottingham in the UK. In this interview, Edith shares her experience as a human-centric AI researcher, along with her opinions about ethical AI and her thoughts on current technology developments in African communities.

Human brain organoids (HBOs) have emerged as transformative models for neurodevelopment and disease, yet ethical concerns persist regarding their potential to develop consciousness. Since 2020, a growing cohort of neuroscientists and philosophers has dismissed these concerns as unscientific, citing limited structural complexity, absence of bodily integration and environmental interaction, and a prevailing neuroscientific consensus against the feasibility of any, or any near-future, emergence of HBO consciousness, thus challenging any suggested revisions of ethical guidelines and safeguards. We argue that this dismissal is premature. Drawing on neuroscientific benchmarks, comparisons to the developing human brain, contemporary theories of consciousness, and principles of natural developmental progression, we question the basis for selectively excluding consciousness from among HBOs' expanding functional repertoire. We caution against enshrining such skepticism into dogma or using it to defer ethical engagement. Instead, we advocate for proactive, ongoing assessment of the moral implications of advancing HBO capabilities.

Hepatocellular carcinoma (HCC) treatment is challenging due to tumor heterogeneity and patient variability. Current guidelines often overlook individual factors, limiting treatment precision. We developed an integrated framework combining radiomics, deep learning, and large language model (LLM)-based decision agents to generate personalized HCC treatment recommendations. A modified GhostNet incorporating dilated convolutions, channel and spatial attention mechanism (CBAM), and residual channel attention (RCA) modules was trained on MRI to predict pathological markers such as microvascular invasion (MVI), capsule presence, and tumor differentiation. A fusion model integrating radiomics and deep learning enhanced prediction accuracy. Six AI agents processed structured multimodal data and generated individualized treatment strategies, which were evaluated by hepatobiliary surgeons. The fusion model significantly improved prediction accuracy, with MVI and capsule presence reaching 0.8902 and 0.8765, respectively. DeepSeek-R1 achieved the highest clinical relevance score, followed by GPT-4 and Med-PaLM 2. This framework demonstrates the feasibility of AI-assisted, patient-specific HCC decision-making, offering a promising direction for precision oncology.

The integration of large language models (LLMs) into the drug discovery and development field marks a significant paradigm shift, offering novel methodologies for understanding disease mechanisms, facilitating de novo drug discovery, and optimizing clinical trial processes. This review highlights the expanding role of LLMs in revolutionizing various stages of the drug development pipeline. We investigate how these advanced computational models can uncover target-disease linkage, interpret complex biomedical data, enhance drug molecule design, predict drug efficacy and safety profiles, and facilitate clinical trial processes. In this paper, we aim to provide a comprehensive overview for researchers and practitioners in computational biology, pharmacology, and AI4Science by offering insights into the potential transformative impact of LLMs on drug discovery and development.

In vitro fertilization (IVF) has significantly advanced the treatment of infertility, yet success rates remain modest due to its complexity and reliance on clinical experience. Recent advances in artificial intelligence (AI) offer promising tools to support decision-making throughout the IVF process. This review summarizes current applications of AI in IVF by organizing studies according to the data modality they use, including structured health records, biomedical images, and omics data. For each modality, we describe representative tasks, model performance, and key methodological progress. We also examine the potential of emerging AI approaches, such as multi-modal learning and large language models. In addition, we acknowledge ongoing challenges, including limited model generalizability, data bias, and the need for clinically validated, transparent AI systems. While the integration of AI into IVF is promising, its success will depend on rigorous validation, ethical safeguards, and interdisciplinary efforts to ensure safe and equitable implementation.

The rise of data-driven scientific investigations has made research data management (RDM) essential for good scientific practice. Implementing RDM is a complex challenge for research communities, infrastructures, and host organizations. Generic RDM guidelines often do not address practical questions, and disciplinary best practices can be overwhelming without proper context. Once guidelines are established, expanding their reach and keeping them up to date is challenging. The RDMkit is an open community-led resource designed as a gateway to reach the wealth of RDM knowledge, tools, training, and resources in life sciences. The RDMkit provides best-practice guidelines on common RDM tasks expected of data stewards and researchers, specific data management challenges and solutions from life science domains, and tool assemblies showcasing holistic solutions to support the research data life cycle. Built on a reusable open infrastructure, the RDMkit allows organizations to create their own guidelines using it as a blueprint.

We propose the theories of pragmatism and functionalism to differentiate between artificial neural networks (ANNs) and biological neural networks (BNNs). While ANNs emulate some cell structures and function approximation mechanisms, questions remain about their ability to emulate intelligent behavior observed in BNNs. We propose that relying solely on biological structures suitable for function approximation may overlook pivotal aspects of ANNs' development, limiting their potential to emulate robust intelligence. Specifically, we investigate the role of ion channels in biological neurons and the randomness they introduce. This randomness seems to be vital for spike generation, although it is not directly related to function approximation. We conclude that structures, which do not directly contribute to function approximation, play a significant role in controlled activity, such as behavior, and should be integrated more into the controlled activity of artificial systems.