For decades, the quest to identify reliable biomarkers circulating freely in bodily fluids has been hampered by considerable analytical and interpretative challenges. Cell-free RNA, fragments of RNA shed by cells into the bloodstream and other biofluids, encapsulates a treasure trove of biological information reflective of an individual’s health status and disease progression. However, the complexity of cfRNA transcripts, their low abundance, and the biological noise inherent to such data have posed formidable obstacles to their effective utilization in clinical diagnostics.

This new study pioneers a systematic evaluation of state-of-the-art large language models, typically employed in natural language processing tasks, for their ability to digest vast volumes of cfRNA sequencing data and autonomously identify candidate biomarkers. By leveraging the intrinsic pattern recognition and semantic understanding capabilities of LLMs, the research team aimed to transcend conventional algorithmic pipelines that often rely on rudimentary feature extraction and handcrafted rules, which can miss subtle but critical molecular signatures.

The team orchestrated an exhaustive benchmarking framework encompassing multiple LLM architectures trained on diverse cfRNA datasets encompassing various disease states, including oncological, neurodegenerative, and inflammatory disorders. This approach allowed them to dissect how different model configurations and training paradigms influenced biomarker detection sensitivity, specificity, and robustness. Performance was compared against gold-standard biomarker discovery methodologies established in molecular biology and bioinformatics.

One notable technical revelation was the LLMs’ capacity to contextualize cfRNA sequences beyond mere nucleotide composition, integrating secondary structure information, transcript isoform variability, and even epitranscriptomic modifications into their predictive models. This unprecedented depth of interpretation allowed the AI to pinpoint diagnostic signatures that remain elusive to traditional algorithms, particularly in heterogeneous sample cohorts where signal dilution is problematic.

Furthermore, the researchers deployed advanced interpretability techniques borrowed from explainable AI to elucidate how these language models formulate their predictions, thereby providing crucial insights into cfRNA pathological relevance. These findings enhance the clinical trustworthiness and adoption potential of AI-driven biomarker discovery, addressing a key bottleneck that has historically prevented machine learning methods from being fully embraced by medical practitioners.

Importantly, the study underscores the scalability and adaptability of LLM-based biomarker workflows. By fine-tuning pre-trained models on modestly sized domain-specific cfRNA datasets, the approach facilitates rapid deployment across multiple disease contexts without the prerequisite for extensive retraining. This adaptability transforms the biomarker discovery pipeline from a painstaking, manual endeavor into an agile, automated process with the power to accelerate diagnostic innovation at an unprecedented pace.

The implications for patient care are profound. Early and accurate detection of diseases through blood-based cfRNA biomarkers can enable earlier interventions, better prognostic assessments, and more personalized therapeutic regimens. By sharply reducing dependence on invasive tissue biopsies or complex imaging, this AI-powered paradigm promises to improve patient comfort, accessibility, and monitoring frequency.

The researchers also highlight how this inter-disciplinary fusion prompts a reevaluation of how biological datasets are curated and annotated. Incorporating contextual metadata and harmonizing nomenclature between molecular biology and computational linguistics are critical to optimizing LLM training. This study sets a new standard for cross-domain collaboration between data scientists, clinicians, and molecular researchers, fostering a virtuous cycle of data quality improvement and model advancement.

While the potential of LLMs in cfRNA biomarker discovery is vividly demonstrated, the authors candidly discuss prevailing challenges. Chief among these is the need for comprehensive, high-fidelity ground truth datasets to validate AI-predicted biomarkers in prospective clinical trials. Additionally, questions of model bias, overfitting to training data, and generalizability across diverse populations require ongoing vigilant scrutiny and methodological refinements.

Looking forward, the study envisions a future where LLMs become an integral component of diagnostic laboratories, seamlessly embedded within clinical decision-support systems. Coupled with advances in portable sequencing technologies and real-time data streaming, the fusion of AI with cfRNA analysis could enable dynamic health monitoring platforms capable of anticipating disease flares or treatment responses.

Moreover, this paradigm holds promise beyond diagnostics, potentially guiding drug target discovery and unraveling complex regulatory networks underpinning human diseases. Harnessing the nuanced language understanding abilities of LLMs to decode the transcriptomic ‘language’ of cfRNA epitomizes a bold step towards truly integrative, systems-level biology.

In essence, the comprehensive benchmarking study by Gaudio and colleagues illuminates the transformative potential of leveraging cutting-edge large language models in the quest for next-generation cfRNA diagnostic biomarkers. By forging a new path that bridges AI and molecular diagnostics, this work not only accelerates biomarker discovery but also sets the stage for innovative healthcare solutions that are less invasive, more accurate, and profoundly responsive to individual patient contexts. As the medical community embraces these insights, we stand on the cusp of a new era where AI deciphers biological complexity with an unprecedented fluency—rewriting the future of medicine.

Subject of Research:
The evaluation of large language models for their application in discovering diagnostic biomarkers from cell-free RNA data.

Article Title:
Benchmarking large language models for cell-free RNA diagnostic biomarker discovery.

Article References:
Gaudio, H.A., Bliss, A., Loy, C.J. et al. Benchmarking large language models for cell-free RNA diagnostic biomarker discovery. Nat Commun (2026). https://doi.org/10.1038/s41467-026-74077-x

Image Credits: AI Generated

Conventional genetic assessments have long operated within a binary diagnostic framework—classifying individuals as either affected or unaffected by certain diseases. However, this categorical approach inadequately captures the complexities inherent in many common conditions such as hypertension, diabetes, and various forms of cancer, where phenotypic expression can span a spectrum of severity and onset. Addressing this limitation, the Mount Sinai team employed advanced machine learning algorithms to quantify disease expression continuously, thereby providing a more refined and clinically actionable insight into penetrance. This approach transcends simplistic yes/no verdicts, offering patients and healthcare providers a dynamic and scalable risk assessment tool.

At the core of this innovation is the integration of over one million electronic health records (EHRs), which furnish the AI models with an unprecedented depth of longitudinal clinical data. Variables such as lipid profiles, complete blood counts, and markers of renal function—parameters routinely collected in clinical practice—serve as real-world physiological indicators that enrich the model’s predictive capacity. By harmonizing these diverse data streams, the AI system calculates an individualized penetrance score ranging from 0 to 1, wherein values nearing unity denote a higher probability that a particular genetic variant will precipitate disease, and values closer to zero suggest negligible or absent risk.

Senior author Dr. Ron Do, Charles Bronfman Professor in Personalized Medicine, articulates the transformative potential of this approach: “Our goal was to move beyond binary interpretations that often leave patients and clinicians uncertain about the real-world implications of genetic test results. By harnessing artificial intelligence alongside routinely available clinical laboratory data, we can now deliver more precise estimates of disease risk for patients harboring specific variants, particularly those that are rare or previously uncharacterized.” This paradigm shift promises to enhance clinical decision-making by facilitating personalized risk stratification grounded in empirical evidence rather than theoretical assumptions.

The study’s development of the “ML penetrance” score entailed rigorous data curation and algorithmic training across ten prevalent diseases. The spectrum of diseases was carefully chosen to encompass conditions with heterogenous presentation and variable genetic etiology, ensuring robust applicability of the model. When applied to over 1,600 rare genetic variants, the AI revealed unexpected patterns: some variants formerly deemed of “uncertain significance” exhibited clear associations with disease phenotypes, while others previously implicated as pathogenic showed minimal effect in population-level clinical data. These findings underscore the critical importance of leveraging large-scale, real-world datasets to revisit and refine the pathogenicity classification of genetic variants.

Lead study author Dr. Iain S. Forrest emphasizes the clinical utility of these findings, cautioning that while the AI tool is not intended to supplant physician judgment, it offers an invaluable adjunct in ambiguous cases. For instance, in carriers of variants linked to Lynch syndrome—a hereditary cancer predisposition syndrome—the penetrance score could prompt timely screening interventions in high-risk individuals, thereby preventing cancer development or enabling early detection. Conversely, a low-risk score might spare patients from unnecessary surveillance and the anxiety associated with overdiagnosis. This precision-guided approach fosters a balance between proactive care and avoidance of overtreatment.

Moreover, the investigators are expanding the scope of their model to incorporate additional diseases and a broader array of genetic alterations, including structural variants and complex haplotypes. A critical future direction involves validating the predictive accuracy longitudinally by monitoring whether individuals with high penetrance scores indeed manifest disease and assessing the impact of early clinical interventions prompted by AI-based risk assessment. Such longitudinal studies will be pivotal in solidifying the clinical integration of AI-driven penetrance estimation.

Beyond the algorithmic innovation, this research exemplifies the fruitful synergy achievable through the confluence of genomics, clinical informatics, and artificial intelligence. Mount Sinai’s Windreich Department of AI and Human Health, under the leadership of Dr. Girish N. Nadkarni, who is internationally recognized for his expertise in ethical AI deployment in healthcare, played an instrumental role in driving this interdisciplinary endeavor. The department’s commitment to responsible AI research ensures that technologies like the ML penetrance model are developed with rigorous attention to clinical applicability, patient safety, and ethical considerations.

This work also benefits from Mount Sinai’s partnership with the Hasso Plattner Institute for Digital Health, a unique collaboration between the Mount Sinai Health System and the Hasso Plattner Institute for Digital Engineering in Germany. Their combined expertise in biomedical informatics, machine learning, and digital engineering accelerates the translation of computational breakthroughs into practical clinical tools, fostering scalable innovations geared toward improving health outcomes globally.

The broader institutional context is equally significant. The Icahn School of Medicine at Mount Sinai, one of the preeminent academic medical centers in the United States, boasts extensive expertise in translational research and clinical care. Its integration within a large, diverse health system provides unparalleled access to rich clinical datasets, enabling the development of data-driven approaches such as the ML penetrance model on a population scale. This infrastructure is essential for validating AI models across heterogeneous patient populations and ensuring their generalizability and equity.

In an era when the volume of genetic testing continues to surge, yielding a vast number of rare and ambiguous variants awaiting clinical interpretation, the integration of AI-driven penetrance estimation represents a crucial advancement. This methodology has the potential to demystify genetic risk, foster precision interventions, and ultimately improve patient outcomes through data-driven personalization. As genetic medicine moves toward this more refined, continuous risk assessment paradigm, patients and clinicians alike stand to gain clarity amidst the complexity of genomic information.

The study, titled “Machine learning-based penetrance of genetic variants,” signifies a landmark step in moving beyond traditional genetics into an era where machine learning and comprehensive clinical data converge to illuminate the nuanced realities of disease risk. By equipping healthcare providers with probabilistic tools grounded in rigorous data analysis, this research heralds a future where genetic information is no longer a source of uncertainty but a guiding beacon for tailored medical care.

Subject of Research: People

Article Title: Machine learning-based penetrance of genetic variants

News Publication Date: 28-Aug-2025

Web References: https://ai.mssm.edu/

References: Forrest IS, Vy HMT, Rocheleau G, Jordan DM, Petrazzini BO, Nadkarni GN, Cho JH, Ganapathi M, Huang K-L, Chung WK, Do R. Machine learning-based penetrance of genetic variants. Science. 2025 Aug 28.

Keywords: Genetic algorithms, Machine learning, Genetic penetrance, Precision medicine, Electronic health records, Rare genetic variants

Recent technological advances have expanded the horizons of imaging genomics beyond traditional single-modality approaches. Modern investigations harness multi-modal imaging data—including computed tomography (CT), magnetic resonance imaging (MRI), X-rays, and ultrasound—alongside a spectrum of molecular data, spanning genomics, transcriptomics, and proteomics. This integrative approach offers unprecedented insights into the molecular architecture of diseases, particularly across complex pathologies such as cancer and cardiovascular disorders. By superimposing imaging phenotypes with multi-omic molecular profiles, researchers are beginning to unravel the pathophysiological mechanisms at a resolution previously unattainable.

Central to the future trajectory of imaging genomics is the advancement of computational frameworks. The past few years have witnessed the emergence of large-scale foundational models, leveraging deep learning architectures and increasingly powerful computational resources. These models show exceptional promise in decoding the high-dimensional, multimodal data intrinsic to imaging genomics. Nonetheless, a major technical bottleneck remains: the absence of a robust, unified foundation model capable of seamlessly integrating cross-scale imaging and omics information. Challenges include harmonizing data across disparately scaled modalities, achieving interpretability in cross-modal analyses, and meeting the formidable demands on computing power.

One of the most exciting frontiers lies in the union of imaging genomics with precision medicine. Imaging genomics complements the phenotypic limitations inherent in electronic medical records by providing detailed molecular and structural disease characterizations. However, current clinical translation efforts are hampered by several factors. Most studies rely on retrospective, cross-sectional data, lacking the longitudinal dimension necessary for tracking disease progression and therapeutic response over time. Furthermore, existing analyses predominantly validate known diagnostic or treatment paradigms, rather than discovering novel biomarkers and therapeutic targets through integrative image-omics correlations.

Recent advancements in cross-modal translation techniques create a paradigm shift in how imaging data might inform omic profiles and vice versa. This burgeoning cross-talk facilitates not only the identification of prognostic biomarkers but also the rational design of targeted therapies. A systematic framework encompassing cross-organ and cross-disease associations stands to radically enhance our understanding of disease etiology. By embedding principles of biological connectivity and multi-organ pathophysiological pathways, imaging genomics is positioned to provide comprehensive disease atlases that elucidate early disease onset, progression trajectories, and multisystem interactions.

The roadmap for the coming decade envisions a transformative shift in imaging genomics from retrospective data validation to integrative systems biology modeling. Such modeling paradigms will utilize interpretative deep learning and large language models to generate interpretable, multimodal disease representations. These will underpin biomarker discovery and the identification of novel therapeutic targets, ultimately empowering clinicians to deliver precise, individualized medical interventions. The integration of these technologies promises to bridge the conceptual gap between molecular biology and clinical applicability.

As Dr. Xiao Ping Cen from the University of Chinese Academy of Sciences highlights, the evolution of imaging genomics will elevate the field from isolated correlation studies to holistic systems-level insights. The increased accessibility to global data collaboration networks, combined with advances in artificial intelligence, positions imaging genomics as a cornerstone of future diagnosis and treatment. These innovations are expected to culminate in clinical decision-making tools capable of tailoring therapy plans to the unique genetic and phenotypic profiles of each patient.

The challenges inherent in this transition are non-trivial. New algorithms must overcome the complexities inherent to multi-modality data heterogeneity, as well as the interpretability crisis characteristic of many “black-box” AI models. Additionally, computational infrastructures will need to scale efficiently to manage the massive datasets generated by high-throughput sequencing and advanced imaging platforms. Researchers increasingly emphasize model transparency and explainability to foster clinical trust and regulatory acceptance.

A significant portion of future research will also focus on longitudinal data integration. By capturing temporal changes in imaging and omic profiles, scientists can delineate disease progression pathways, identify early markers of therapeutic resistance, and optimize intervention timing. The incorporation of longitudinal analyses introduces dynamic modeling capabilities that can predict future outcomes and simulate intervention effects, advancing imaging genomics beyond static snapshots to predictive, real-world clinical utility.

The broad applicability of imaging genomics extends across diverse disease frameworks. In oncology, the correlation of tumor imaging phenotypes with mutational landscapes paves the way for non-invasive tumor characterization and personalized treatment planning. In cardiovascular medicine, imaging-genomic associations promise improved stratification of atherosclerotic risk and tailored management protocols. The integration of multi-organ data sets enables holistic patient profiling, accounting for systemic factors influencing disease manifestation and treatment response.

Ultimately, imaging genomics encapsulates the synergistic potential of advanced imaging technologies, high-throughput omics, and state-of-the-art artificial intelligence methods. As we progress into an era defined by precision medicine, the capacity to interpret complex biological data within a unified, clinically actionable framework becomes paramount. The burgeoning field of imaging genomics offers a visionary path forward—a confluence where biology, technology, and medicine coalesce to drive transformative healthcare outcomes.

Subject of Research: Imaging Genomics and Multimodal Data Integration in Precision Medicine

Article Title: Roadmap for Imaging Genomics in the Next Decade

Web References:
http://dx.doi.org/10.1016/j.scib.2025.04.058

Image Credits: Created with Advanced Deep Learning and Large Language Models Frameworks

Keywords: Imaging Genomics, Radiogenomics, Deep Learning, Large Language Models, Multimodal Data Integration, Precision Medicine, Biomarker Discovery, Systems Biology, Cross-Modal Analysis, Disease Atlas