Recent advances in machine learning and AI, including Generative AI and LLMs, are disrupting technological innovation, product development, and society as a whole. AI's contribution to technology can come from multiple approaches that require access to large training data sets and clear performance evaluation criteria, ranging from pattern recognition and classification to generative models. Yet, AI has contributed less to fundamental science in part because large data sets of high-quality data for scientific practice and model discovery are more difficult to access. Generative AI, in general, and Large Language Models in particular, may represent an opportunity to augment and accelerate the scientific discovery of fundamental deep science with quantitative models. Here we explore and investigate aspects of an AI-driven, automated, closed-loop approach to scientific discovery, including self-driven hypothesis generation and open-ended autonomous exploration of the hypothesis space. Integrating AI-driven automation into the practice of science would mitigate current problems, including the replication of findings, systematic production of data, and ultimately democratisation of the scientific process. Realising these possibilities requires a vision for augmented AI coupled with a diversity of AI approaches able to deal with fundamental aspects of causality analysis and model discovery while enabling unbiased search across the space of putative explanations. These advances hold the promise to unleash AI's potential for searching and discovering the fundamental structure of our world beyond what human scientists have been able to achieve. Such a vision would push the boundaries of new fundamental science rather than automatize current workflows and instead open doors for technological innovation to tackle some of the greatest challenges facing humanity today.

Motivation:Knowledge graphs (KGs), collectively as a knowledge network, have become critical tools for knowledge discovery in computable and explainable knowledge systems. Due to the semantic and structural complexities of biomedical data, these KGs need to enable dynamic reasoning over large evolving graphs and support fit-for-purpose abstraction. Crucially, this requires establishing standards, preserving provenance and enforcing policy constraints for actionable discovery. Results:A recent meeting of leading scientists discussed the opportunities, challenges, and future directions of a biomedical knowledge network. Here we present six desiderata inspired by the meeting: (i) inference and reasoning in biomedical KGs need domain-centric approaches, (ii) harmonized and accessible standards are required for knowledge graph representation and metadata, (iii) robust validation of biomedical KGs needs multilayered, context-aware approaches that are both rigorous and scalable, (iv) the evolving and synergistic relationship between KGs and large language models is essential in empowering AI-driven biomedical discovery, (v) integrated development environments, public repositories, and governance frameworks are essential for secure and reproducible knowledge graph sharing, and (vi) robust validation, provenance, and ethical governance are critical for trustworthy biomedical KGs. Addressing these key issues will be essential to realize the promises of a biomedical knowledge network in advancing biomedicine.

Coreference resolution in biomedical texts presents unique challenges due to complex domain-specific terminology, high ambiguity in mention forms, and long-distance dependencies between coreferring expressions. In this work, we present a comprehensive evaluation of generative large language models (LLMs) for coreference resolution in the biomedical domain. Using the CRAFT corpus as our benchmark, we assess the LLMs' performance with four prompting experiments that vary in their use of local, contextual enrichment, and domain-specific cues such as abbreviations and entity dictionaries. We benchmark these approaches against a discriminative span-based encoder, SpanBERT, to compare the efficacy of generative versus discriminative methods. Our results demonstrate that while LLMs exhibit strong surface-level coreference capabilities, especially when supplemented with domain-grounding prompts, their performance remains sensitive to long-range context and mentions ambiguity. Notably, the LLaMA 8B and 17B models show superior precision and F1 scores under entity-augmented prompting, highlighting the potential of lightweight prompt engineering for enhancing LLM utility in biomedical NLP tasks.

Despite recent progress, Biomedical Entity Linking (BEL) with large language models (LLMs) remains computationally inefficient and challenging to deploy in practical settings. In this work, we demonstrate that instruction-tuning of open-source generative models can offer an effective solution when applied at the re-ranking stage of the BEL pipeline. We propose a set-wise instruction-tuning formulation that enables fast and accurate candidate selection. Our method demonstrates strong performance on multiple BEL benchmarks, yielding significant improvements in linking accuracy (3

Language models (LMs) have excelled in various broad domains. However, to ensure their safe and effective integration into real-world educational settings, they must demonstrate proficiency in specific, granular areas of knowledge. Existing cloze-style benchmarks, commonly used to evaluate LMs' knowledge, have three major limitations. They: 1) do not cover the educational domain; 2) typically focus on low-complexity, generic knowledge or broad domains, which do not adequately assess the models' knowledge in specific subjects; and 3) often rely on templates that can bias model predictions. Here, we introduce MALAMUTE, a multilingual, template-free, and highly granular probing dataset comprising expert-written, peer-reviewed probes from 71 university-level textbooks across three languages (English, Spanish, and Polish). MALAMUTE is the first education-based cloze-style dataset. It covers eight domains, each with up to 14 subdomains, further broken down into concepts and concept-based prompts, totaling 33,361 university curriculum concepts and 116,887 prompts. MALAMUTE's fine granularity, educational focus, and inclusion of both sentence-level and paragraph-level prompts make it an ideal tool for evaluating LMs' course-related knowledge. Our evaluation of masked and causal LMs on MALAMUTE shows that despite overall proficiency, they have significant gaps in knowledge when examined closely on specific subjects, hindering their safe use in classrooms and underscoring the need for further development.

The evolution of biological neural systems has led to both modularity and sparse coding, which enables energy efficiency and robustness across the diversity of tasks in the lifespan. In contrast, standard neural networks rely on dense, non-specialized architectures, where all model parameters are simultaneously updated to learn multiple tasks, leading to interference. Current sparse neural network approaches aim to alleviate this issue but are hindered by limitations such as 1) trainable gating functions that cause representation collapse, 2) disjoint experts that result in redundant computation and slow learning, and 3) reliance on explicit input or task IDs that limit flexibility and scalability.In this paper we propose Conditionally Overlapping Mixture of ExperTs (COMET), a general deep learning method that addresses these challenges by inducing a modular, sparse architecture with an exponential number of overlapping experts. COMET replaces the trainable gating function used in Sparse Mixture of Experts with a fixed, biologically inspired random projection applied to individual input representations. This design causes the degree of expert overlap to depend on input similarity, so that similar inputs tend to share more parameters. This results in faster learning per update step and improved out-of-sample generalization. We demonstrate the effectiveness of COMET on a range of tasks, including image classification, language modeling, and regression, using several popular deep learning architectures.

Summary:Modern biological research critically depends on public databases. The introduction and propagation of errors within and across databases can lead to wasted resources as scientists are led astray by bad data or have to conduct expensive validation experiments. The emergence of generative artificial intelligence systems threatens to compound this problem owing to the ease with which massive volumes of synthetic data can be generated. We provide an overview of several key issues that occur within the biological data ecosystem and make several recommendations aimed at reducing data errors and their propagation. We specifically highlight the critical importance of improved educational programs aimed at biologists and life scientists that emphasize best practices in data engineering. We also argue for increased theoretical and empirical research on data provenance, error propagation, and on understanding the impact of errors on analytic pipelines. Furthermore, we recommend enhanced funding for the stewardship and maintenance of public biological databases. Availability and implementation:Not applicable.

Scientific progress is driven by the deliberate articulation of what remains unknown. This study investigates the ability of large language models (LLMs) to identify research knowledge gaps in the biomedical literature. We define two categories of knowledge gaps: explicit gaps, clear declarations of missing knowledge; and implicit gaps, context-inferred missing knowledge. While prior work has focused mainly on explicit gap detection, we extend this line of research by addressing the novel task of inferring implicit gaps. We conducted two experiments on almost 1500 documents across four datasets, including a manually annotated corpus of biomedical articles. We benchmarked both closed-weight models (from OpenAI) and open-weight models (Llama and Gemma 2) under paragraph-level and full-paper settings. To address the reasoning of implicit gaps inference, we introduce TABI, a Toulmin-Abductive Bucketed Inference scheme that structures reasoning and buckets inferred conclusion candidates for validation. Our results highlight the robust capability of LLMs in identifying both explicit and implicit knowledge gaps. This is true for both open- and closed-weight models, with larger variants often performing better. This suggests a strong ability of LLMs for systematically identifying candidate knowledge gaps, which can support early-stage research formulation, policymakers, and funding decisions. We also report observed failure modes and outline directions for robust deployment, including domain adaptation, human-in-the-loop verification, and benchmarking across open- and closed-weight models.

Prior work has uncovered a set of common problems in state-of-the-art context-based question answering (QA) systems: a lack of attention to the context when the latter conflicts with a model's parametric knowledge, little robustness to noise, and a lack of consistency with their answers. However, most prior work focus on one or two of those problems in isolation, which makes it difficult to see trends across them. We aim to close this gap, by first outlining a set of -- previously discussed as well as novel -- desiderata for QA models. We then survey relevant analysis and methods papers to provide an overview of the state of the field. The second part of our work presents experiments where we evaluate 15 QA systems on 5 datasets according to all desiderata at once. We find many novel trends, including (1) systems that are less susceptible to noise are not necessarily more consistent with their answers when given irrelevant context; (2) most systems that are more susceptible to noise are more likely to correctly answer according to a context that conflicts with their parametric knowledge; and (3) the combination of conflicting knowledge and noise can reduce system performance by up to 96%. As such, our desiderata help increase our understanding of how these models work and reveal potential avenues for improvements.

The differences between cloze-task language model (LM) probing with 1) expert-made templates and 2) naturally-occurring text have often been overlooked. Here, we evaluate 16 different LMs on 10 probing English datasets - 4 template-based and 6 template-free - in general and biomedical domains to answer the following research questions: (RQ1) Do model rankings differ between the two approaches? (RQ2) Do models' absolute scores differ between the two approaches? (RQ3) Do the answers to RQ1 and RQ2 differ between general and domain-specific models? Our findings are: 1) Template-free and template-based approaches often rank models differently, except for the top domain-specific models. 2) Scores decrease by up to 42% Acc@1 when comparing parallel template-free and template-based prompts. 3) Perplexity is negatively correlated with accuracy in the template-free approach, but, counter-intuitively, they are positively correlated for template-based probing. 4) Models tend to predict the same answers frequently across prompts for template-based probing, which is less common when employing template-free techniques. Code and data can be found here: https://github.com/Shaier/probing_template_based_template_free.git.

Previous work finds that recent long-context language models fail to make equal use of information in the middle of their inputs, preferring pieces of information located at the tail ends which creates an undue bias in situations where we would like models to be equally capable of using different parts of the input. Thus far, the problem has mainly only been considered in settings with single pieces of critical information, leading us to question what happens when multiple necessary pieces of information are spread out over the inputs. Here, we demonstrate the effects of the "lost in the middle" problem in the multi-hop question answering setting – in which multiple reasoning "hops" over disconnected documents are required – and show that performance degrades not only with respect to the distance of information from the edges of the context, but also between pieces of information. Additionally, we experiment with means of alleviating the problem by reducing superfluous document contents through knowledge graph triple extraction and summarization, and prompting models to reason more thoroughly using chain-of-thought prompting.

Advances in high-throughput technologies have enhanced our ability to describe microbial communities as they relate to human health and disease. Alongside the growth in sequencing data has come an influx of resources that synthesize knowledge surrounding microbial traits, functions, and metabolic potential with knowledge of how they may impact host pathways to influence disease phenotypes. These knowledge bases can enable the development of mechanistic explanations that may underlie correlations detected between microbial communities and disease. In this review, we survey existing resources and methodologies for the computational integration of broad classes of microbial and host knowledge. We evaluate these knowledge bases in their access methods, content, and source characteristics. We discuss challenges of the creation and utilization of knowledge bases including inconsistency of nomenclature assignment of taxa and metabolites across sources, whether the biological entities represented are rooted in ontologies or taxonomies, and how the structure and accessibility limit the diversity of applications and user types. We make this information available in a code and data repository at: https://github.com/lozuponelab/knowledge-source-mappings. Addressing these challenges will allow for the development of more effective tools for drawing from abundant knowledge to find new insights into microbial mechanisms in disease by fostering a systematic and unbiased exploration of existing information.

The more science advances, the more questions are asked. This compounding growth can make it difficult to keep up with current research directions. Furthermore, this difficulty is exacerbated for junior researchers who enter fields with already large bases of potentially fruitful research avenues. In this paper, we propose a novel task and a recommender system for research directions, RecSOI, that draws from statements of ignorance (SOIs) found in the research literature. By building researchers’ profiles based on textual elements, RecSOI generates personalized recommendations of potential research directions tailored to their interests. In addition, RecSOI provides context for the recommended SOIs, so that users can quickly evaluate how relevant the research direction is for them. In this paper, we provide an overview of RecSOI’s functioning, implementation, and evaluation, demonstrating its effectiveness in guiding researchers through the vast landscape of potential research directions.

State-of-the-art question answering (QA) models exhibit a variety of social biases (e.g., with respect to sex or race), generally explained by similar issues in their training data. However, what has been overlooked so far is that in the critical domain of biomedicine, any unjustified change in model output due to patient demographics is problematic: it results in the unfair treatment of patients. Selecting only questions on biomedical topics whose answers do not depend on ethnicity, sex, or sexual orientation, we ask the following research questions: (RQ1) Do the answers of QA models change when being provided with irrelevant demographic information? (RQ2) Does the answer of RQ1 differ between knowledge graph (KG)-grounded and text-based QA systems? We find that irrelevant demographic information change up to 15% of the answers of a KG-grounded system and up to 23% of the answers of a text-based system, including changes that affect accuracy. We conclude that unjustified answer changes caused by patient demographics are a frequent phenomenon, which raises fairness concerns and should be paid more attention to.

Knowledge graphs have found broad biomedical applications, providing useful representations of complex knowledge. Although plentiful evidence exists linking the gut microbiome to disease, mechanistic understanding of those relationships remains generally elusive. Here we demonstrate the potential of knowledge graphs to hypothesize plausible mechanistic accounts of host-microbe interactions in disease. To do so, we constructed a knowledge graph of linked microbes, genes and metabolites called MGMLink. Using a semantically constrained shortest path search through the graph and a novel path prioritization methodology based on cosine similarity, we show that this knowledge supports inference of mechanistic hypotheses that explain observed relationships between microbes and disease phenotypes. We discuss specific applications of this methodology in inflammatory bowel disease and Parkinson's disease. This approach enables mechanistic hypotheses surrounding the complex interactions between gut microbes and disease to be generated in a scalable and comprehensive manner.

Motivation:Knowledge graphs have found broad biomedical applications, providing useful representations of complex knowledge. Although plentiful evidence exists linking the gut microbiome to disease, mechanistic understanding of those relationships remains generally elusive. A structured analysis of existing resources is necessary to characterize the resources and methodologies needed to facilitate mechanistic inference.Results:Here we demonstrate the potential of knowledge graphs to hypothesize plausible mechanistic accounts of host-microbe interactions in disease and define the need for semantic constraint in doing so. We constructed a knowledge graph of linked microbes, genes and metabolites called MGMLink, and one of microbial traits, environments, and human pheno-types called KG-microbe-phenio. Using a shortest path search and a pattern based semantically constrained path search through the graphs, we highlight the need for a microbiome-disease resource and semantically informed search methods to enable mechanistic inference.Availability:The software to create MGMLink is openly available at https://github.com/bsantan/MGMLink , and KG-microbe is available at https://github.com/Knowledge-Graph-Hub/kg-microbe and KG-phenio is available at https://github.com/Knowledge-Graph-Hub/kg-phenio .Contact:brook.santangelo@cuanschutz.edu.

We have developed a flexible undergraduate curriculum that leverages the place-based research of environmental microbiomes to increase the number of Indigenous researchers in microbiology, data science and scientific computing. Monitoring Environmental Microbiomes (MEM) provides a curriculum and research framework designed to integrate an Indigenous approach when conducting authentic scientific research and to build interest and confidence at the undergraduate level. MEM has been successfully implemented as a short summer workshop to introduce computing practices in microbiome analysis. Based on self-assessed student knowledge of topics and skills, increased scientific confidence and interest in genomics careers were observed. We propose MEM be incorporated in a scalable course-based research experience for undergraduate institutions, including tribal colleges and universities, community colleges and other minority serving institutions. This coupled curricular and research framework explicitly considers cultural perspectives, access and equity to train a diverse future workforce that is more informed to engage in microbiome research and to translate microbiome science to benefit community and environmental health.

This chapter introduces three exploratory studies that assessed the capabilities of artificial intelligence (AI) through standardised education tests designed for humans. The first two studies, conducted in 2016 and 2021/22, asked experts to evaluate AI's performance on the literacy and numeracy tests of the OECD's Survey of Adult Skills (PIAAC). The third study collected expert judgements of whether AI can solve science questions from the OECD's Programme for International Student Assessment (PISA). The studies aimed to refine the assessment framework for eliciting expert knowledge on AI using established educational assessments. They explored different test formats, response methodologies and rating instructions, along with two distinct assessment approaches. A "behavioural approach" used in the PIAAC studies emphasised smaller expert groups engaging in discussions, and a "mathematical approach" adopted in the PISA study relied more heavily on quantitative data from a larger expert pool. This chapter presents the results of the studies and discusses the advantages and disadvantages of their methodological approaches.

BACKGROUND:Scientific discovery progresses by exploring new and uncharted territory. More specifically, it advances by a process of transforming unknown unknowns first into known unknowns, and then into knowns. Over the last few decades, researchers have developed many knowledge bases to capture and connect the knowns, which has enabled topic exploration and contextualization of experimental results. But recognizing the unknowns is also critical for finding the most pertinent questions and their answers. Prior work on known unknowns has sought to understand them, annotate them, and automate their identification. However, no knowledge-bases yet exist to capture these unknowns, and little work has focused on how scientists might use them to trace a given topic or experimental result in search of open questions and new avenues for exploration. We show here that a knowledge base of unknowns can be connected to ontologically grounded biomedical knowledge to accelerate research in the field of prenatal nutrition. RESULTS:We present the first ignorance-base, a knowledge-base created by combining classifiers to recognize ignorance statements (statements of missing or incomplete knowledge that imply a goal for knowledge) and biomedical concepts over the prenatal nutrition literature. This knowledge-base places biomedical concepts mentioned in the literature in context with the ignorance statements authors have made about them. Using our system, researchers interested in the topic of vitamin D and prenatal health were able to uncover three new avenues for exploration (immune system, respiratory system, and brain development) by searching for concepts enriched in ignorance statements. These were buried among the many standard enriched concepts. Additionally, we used the ignorance-base to enrich concepts connected to a gene list associated with vitamin D and spontaneous preterm birth and found an emerging topic of study (brain development) in an implied field (neuroscience). The researchers could look to the field of neuroscience for potential answers to the ignorance statements. CONCLUSION:Our goal is to help students, researchers, funders, and publishers better understand the state of our collective scientific ignorance (known unknowns) in order to help accelerate research through the continued illumination of and focus on the known unknowns and their respective goals for scientific knowledge.