CRISPR-Cas systems work by pairing an RNA guide with a DNA target and then using a nuclease domain to cut. Most widely used technologies rely on natural Cas proteins shaped by evolution. But whether AI can generate functional equivalents with novel sequence features, rather than near replicas of known enzymes, has been a difficult question.

Petr Skopintsev and colleagues tackled that challenge using a strategy built around ESM Inverse Folding (ESM-IF1). Instead of allowing the model to drift toward sequences close to training references, the team introduced evolution-informed residue constraints designed to keep key functional elements in place while still permitting large sequence divergence.

Their starting point was TnpB, a minimal CRISPR-Cas12-like nuclease. The researchers designed new variants they call SynTnpBs, which are engineered to remain RNA-guided and catalytically active despite being non-natural. This “minimal nuclease” context is especially important because multi-domain architectures can be fragile—small changes can destroy activity.

After design, the synthetic nucleases were screened across bacterial cells, plant cells, and human cells. Many of the AI-generated enzymes retained or exceeded the performance of the natural TnpB across these distinct environments, supporting the idea that the designs were not merely theoretical but biochemically robust.

To understand how these divergent proteins work, the authors turned to cryo-electron microscopy (cryo-EM). They determined structures for the most divergent SynTnpB variants, providing the first experimentally resolved pictures of AI-designed RNA-guided nucleases.

Across multiple conformations, cryo-EM revealed new stabilizing interactions at the RNA-DNA interface. Rather than matching the natural enzyme’s exact contact patterns, the engineered nucleases appear to solve the same recognition problem through alternative molecular geometry—an outcome consistent with the promise of protein design guided by structure.

Together, the study demonstrates a practical path for creating active genome-editing enzymes that do not simply imitate nature’s sequences. By combining inverse folding with evolution-informed constraints and validating function in real cellular systems, the work suggests that AI-driven design can generate new classes of editing tools.

Subject of Research: AI-designed RNA-guided nucleases for CRISPR-based genome editing
Article Title: Structure and evolution-guided design of minimal RNA-guided nucleases
News Publication Date: 16-Jul-2026
Web References: http://dx.doi.org/10.1126/science.aed6123
References: 10.1126/science.aed6123
Image Credits: Not provided

Keywords: artificial intelligence, protein design, CRISPR, Cas12-like nuclease, ESM Inverse Folding, ESM-IF1, RNA-DNA interface, cryo-EM, genome editing, TnpB, SynTnpBs

The Center for Data Science and Artificial Intelligence at Sanford Burnham Prebys is a nexus where experts from AI, statistics, genetics, and molecular biology collaborate intensively to extract meaningful insights from vast, raw, and often previously untapped biological data. By employing sophisticated computational tools and advanced machine learning algorithms, the center aims to reveal intricate patterns and relationships within genomic, proteomic, and phenotypic datasets that were previously inaccessible through conventional research methodologies. This cross-disciplinary initiative is instrumental in fostering innovation that could lead to novel diagnostic, therapeutic, and prognostic strategies for a variety of diseases.

A hallmark of the center’s recent achievements includes the development of an advanced computational biology tool designed to automate and standardize genome sequencing analysis. This tool represents a monumental step forward in computational genomics, allowing scientists to analyze the entirety of genomes from multiple patient samples, animal models, or cell cultures within a single experimental framework. By streamlining the analytical pipeline and enhancing reproducibility, the center enables researchers to interrogate genetic variations and molecular interactions with unprecedented precision and scale, facilitating rapid identification of disease biomarkers and therapeutic targets.

Andrew Viterbi expressed profound confidence in the leadership and scientific vision of Sanford Burnham Prebys, highlighting the transformative potential of AI in biomedical research. His investment aims to empower the center to harness emergent computational technologies to generate novel insights and accelerate the pace of discovery in human health. The endowed chair reflects Viterbi’s enduring commitment to innovation and his belief in the catalytic power of collaboration across computer science and life sciences to address pressing biomedical challenges.

The inaugural Andrew and Erna Viterbi Distinguished Chair will be held by Dr. Yuk-Lap (Kevin) Yip, an esteemed leader in computational biology and bioinformatics. Dr. Yip’s work epitomizes the fusion of high-level data science and biological inquiry. Utilizing machine learning models and advanced statistical frameworks, his research dissects complex biological phenomena, facilitating breakthroughs in understanding disease mechanisms and propelling precision medicine initiatives. His leadership is vital to directing the center’s multidisciplinary endeavors toward impactful translational outcomes.

This philanthropic initiative by Dr. Viterbi exemplifies how visionary support can accelerate technological convergence in biomedical research. His legacy includes the invention of the Viterbi Algorithm, a seminal dynamic programming method instrumental in decoding sequences of hidden states from observed data, which has far-reaching applications in signal processing and communications. Bringing this legacy into the biomedical domain, Viterbi’s gift symbolizes a bridge between engineering excellence and transformative health science.

Dr. Viterbi’s contribution arrives at a pivotal moment when AI is reshaping the landscape of medical research, diagnostics, and therapeutics. The Center for Data Science and AI is leveraging deep learning, natural language processing, and probabilistic modeling to interrogate multi-dimensional biomedical data at scale. These computational approaches enable decoding of complex disease signatures and prediction of clinical outcomes, heralding a new era of data-driven medicine that promises to personalize healthcare and improve patient prognoses.

Sanford Burnham Prebys is celebrating five decades of groundbreaking biomedical research, and this newly endowed chair marks a significant milestone in expanding its capabilities in AI-driven science. The institute’s diverse research portfolio encompasses cancer, neuroscience, aging, cardiovascular, metabolic, and liver diseases, supported by sophisticated drug discovery platforms and global collaborations. By integrating data science and AI, the institute enhances its capacity to translate fundamental biological discoveries into meaningful clinical applications.

The institute’s commitment to training the next generation of scientists through its graduate school complements its research ambitions. By fostering an environment where computational methods are seamlessly integrated into biological education, the institute ensures a sustainable pipeline of scientists equipped to tackle future challenges at the interface of data science and medicine. This culture of collaboration and education underpins the transformative potential encapsulated in the Viterbi endowment.

David Brenner, MD, president and CEO of Sanford Burnham Prebys, lauded Andrew Viterbi’s gift as a catalyst for redefining how biology and disease are understood in the 21st century. By enabling large-scale, AI-powered discoveries, the endowed chair is expected to accelerate the identification of novel molecular mechanisms underlying disease, streamline drug development, and ultimately, enhance patient care. This partnership exemplifies the power of philanthropy in driving scientific innovation and societal impact.

The endowed chair will continue to attract world-class talent to Sanford Burnham Prebys, reinforcing its status as a leader in biomedical research powered by AI. The center’s unique integration of computational and experimental biology leverages cutting-edge technology to address complex health challenges, from decoding cancer genomes to mapping neuronal networks, illustrating the vast potential of AI to transform medicine.

In sum, Andrew and Erna Viterbi’s remarkable philanthropic vision seeds a future where artificial intelligence and data science are central pillars in biomedical discovery. Their legacy will not only enrich Sanford Burnham Prebys but will resonate throughout the scientific community, inspiring new approaches that harness the full power of computational innovation to unravel the mysteries of human health and disease.

Subject of Research: Computational biology, Artificial intelligence in biomedical research
Article Title: Andrew Viterbi Endows Distinguished Chair to Propel AI-Driven Biomedical Research at Sanford Burnham Prebys
News Publication Date: Not specified
Web References:

• Center for Data Science and Artificial Intelligence: https://sbpdiscovery.org/research/centers/center-for-data-science-and-artificial-intelligence/
• Press release on computational biology tool: https://sbpdiscovery.org/press/new-computational-biology-tool-automates-and-standardizes-genome-sequencing-analysis/
• Sanford Burnham Prebys: http://sbpdiscovery.org/
  Image Credits: Sanford Burnham Prebys
  Keywords: Philanthropy, Artificial intelligence, Computational biology, Biomedical research, Data science, Genome sequencing, Machine learning, Precision medicine

The new multimodal language model combines advancements in artificial intelligence and molecular biology, making it possible to interpret complex RNA datasets with high accuracy. By harnessing the vast potential of deep learning, the model offers a sophisticated framework to decode the nuances of RNA expression profiles. This integration of technology and biology sets a benchmark for future research, paving the way for enhanced patient outcomes through personalized treatment strategies. As the field of liquid biopsy continues to evolve, the ability to analyze and interpret RNA biomarkers will significantly impact the early detection of tumors, enabling timely interventions.

Carcinogenesis is a highly complex process, and tumors are characterized by their dynamic evolution in response to various internal and external stimuli. The researchers’ model addresses this complexity by simulating the biological context surrounding circulating RNA, thus enabling the extraction of invaluable information related to tumor heterogeneity and treatment response. The ability to analyze RNA at different stages of cancer progression empowers oncologists with a deeper understanding of individual tumors’ behavior. This personalized approach risks changing the landscape of cancer treatment, allowing therapies to be tailored to patients based on their unique molecular profiles.

A key component of this multimodal model is its ability to analyze heterogeneous RNA populations derived from various sources, including tumor cells and the surrounding microenvironment. Traditional methods of RNA sequencing often overlook the intricate intercellular communications that occur within the tumor ecosystem. By leveraging a more holistic perspective, this model enhances the resolution at which cancer genomics can be assessed, ultimately refining therapeutic targets. This insight could lead to a more precise identification of actionable mutations, significantly improving patient stratification and therapeutic decision-making.

As researchers delve deeper into RNA’s role in cancer progression, the importance of data interpretation becomes paramount. The multimodal language model not only processes RNA sequences but also incorporates contextual knowledge that aids in understanding the biological implications of these sequences in real-time. For instance, the model can predict the likelihood of oncogenic changes based on specific RNA profiles, enabling early detection of potential malignancies. This predictive capability represents a substantial leap forward in oncological diagnostics, enhancing the clinician’s arsenal in combating cancer in its infancy.

Moreover, the model is designed to handle the vast complexities inherent in liquid biopsy data. Given the abundance of RNA molecules that are present in bodily fluids, it is crucial to distinguish between meaningful biomarkers and background noise. This sophisticated model effectively filters out irrelevant signals, thereby increasing the accuracy of diagnostic predictions. By systematically refining the process of biomarker discovery, researchers can swiftly identify the most impactful RNA sequences linked to cancer, facilitating their integration into clinical settings.

The implications of this research extend far beyond the realm of cancer diagnostics. Similar methodologies could be adapted to investigate various diseases where RNA plays a crucial role, such as neurological disorders, infectious diseases, and genetic conditions. The versatility of the multimodal approach fosters a deeper understanding of disease dynamics, thereby propelling advancements in personalized medicine across multiple medical disciplines. As the scientific community uncovers new connections between RNA profiles and health outcomes, the need for comprehensive models that encompass all aspects of RNA biology becomes increasingly critical.

Another noteworthy aspect of the study is the model’s capability to adapt to emerging data. As the landscape of RNA research continues to evolve, new biomarkers and genetic variations will become apparent. The model’s inherent flexibility allows it to integrate these discoveries, ensuring that its predictive accuracy remains relevant and reliable. This adaptability positions the model as a valuable tool not only for current research but also for future explorations into the molecular underpinnings of health and disease.

The researchers envision that widespread implementation of this multimodal language model could potentially democratize access to advanced diagnostics. By reducing the reliance on traditional biopsy techniques, patients could benefit from quicker, less invasive testing methods. This shift toward non-invasive diagnostics could also lead to increased screening rates, enabling early detection of cancers that might otherwise go unnoticed until they reach advanced stages. Therefore, this research could have far-reaching implications for public health, ultimately leading to improved survival rates and a better quality of life for individuals battling cancer.

In conclusion, the development of a multimodal cell-free RNA language model represents a significant advancement in the field of liquid biopsy and precision medicine. By integrating advanced computational techniques with a deep understanding of molecular biology, this research sets the stage for transformative changes in cancer diagnostics. As researchers continue to refine this model and explore its applications in various clinical settings, the hope is that such innovations will lead to a brighter future in cancer treatment, characterized by early detection, personalized therapies, and improved patient outcomes.

This groundbreaking study serves as a testament to the power of interdisciplinary collaboration, bridging together experts from different fields to tackle the pressing challenges posed by cancer. As we look to the future, the potential applications of this model will shape the next generation of diagnostic technologies, fundamentally altering how we approach disease detection and management in the years to come.

Subject of Research: Cell-free RNA language model for liquid biopsy applications

Article Title: A multimodal cell-free RNA language model for liquid biopsy applications

Article References:

Karimzadeh, M., Sababi, A.M., Momen-Roknabadi, A. et al. A multimodal cell-free RNA language model for liquid biopsy applications.
Nat Mach Intell (2025). https://doi.org/10.1038/s42256-025-01148-x

Image Credits: AI Generated

DOI: https://doi.org/10.1038/s42256-025-01148-x

Keywords: Liquid biopsy, RNA, multimodal language model, cancer detection, personalized medicine

Parkinson’s disease is characterized by the progressive loss of dopaminergic neurons and the accumulation of alpha-synuclein aggregates, leading to debilitating motor and non-motor symptoms. Despite considerable research efforts, the molecular underpinnings driving disease progression remain elusive, primarily due to the complexity of neuronal networks and cellular diversity within affected brain regions. Traditional bulk transcriptomic analyses lack the resolution needed to disentangle this complexity, often masking subtle yet critical gene expression changes occurring in specific cell populations. Addressing this limitation, the research team applied state-of-the-art neural network algorithms to single-nuclei RNA sequencing data, enabling the extraction of cell-type–specific gene expression patterns with extraordinary precision.

The methodology employed leverages deep learning architectures adept at recognizing intricate patterns within vast datasets, surpassing the capabilities of conventional bioinformatic tools. Through this approach, the researchers dissected the transcriptomic profiles of individual nuclei isolated from post-mortem brain tissue of Parkinson’s patients and matched controls. This granular data facilitated the identification of novel gene signatures, including those implicated in neuronal vulnerability, glial dysregulation, and synaptic remodeling — processes integral to Parkinson’s pathology but previously underappreciated due to the limitations of less granular techniques.

Crucially, the neural network’s predictions uncovered unique molecular signatures in glial cells, such as astrocytes and microglia, highlighting their hitherto unrecognized roles in disease progression. These findings align with mounting evidence that neuroinflammation and glial dysfunction are not merely secondary effects but active contributors to PD pathogenesis. By pinpointing gene expression patterns specific to these cell types, the study provides compelling grounds to reconsider therapeutic strategies, potentially redirecting focus to modulating glial activity in the Parkinsonian brain.

Another remarkable outcome was the identification of differential gene expression linked to mitochondrial pathways and oxidative stress responses, which have long been associated with neurodegeneration. The neural network analysis uncovered hitherto unknown players within these pathways that might serve as early biomarkers or therapeutic targets. Identifying such molecular markers at the single-nucleus level offers a more nuanced temporal and spatial understanding of disease onset and progression, which is critical for the development of precision medicine approaches.

Moreover, synaptic genes exhibited altered expression patterns across multiple neuronal subtypes, suggesting that synaptic dysfunction is a pervasive feature in PD. The study’s findings implicate synapse-specific molecular disruptions that could contribute to both motor symptoms and cognitive decline seen in Parkinson’s patients. This granularity is pivotal because it delineates distinct molecular cascades that might be selectively targeted to preserve synaptic integrity, thereby slowing or halting symptom progression.

The study’s application of neural networks also illuminated cellular heterogeneity within the substantia nigra, the brain region most severely impacted by Parkinson’s. By stratifying the expression profiles of dopaminergic neurons and their subpopulations, the analysis revealed distinct vulnerability markers, shedding light on why certain neuronal subsets succumb earlier or more severely than others. This insight is crucial for developing targeted neuroprotective strategies that could selectively bolster the resilience of these vulnerable neuronal populations.

Importantly, the research emphasizes the transformative potential of integrating computational intelligence with high-resolution molecular data in neurodegenerative disease research. The successful deployment of neural networks to dissect single-nuclei transcriptomes represents a quantum leap, moving beyond descriptive biology towards predictive and mechanistic insights. Such technological synergy accelerates the identification of candidate genes and pathways, guiding experimental validation and therapeutic development with unprecedented efficiency.

With these compelling findings, the authors advocate for broader incorporation of neural network–assisted analyses in future Parkinson’s research and beyond. The approach is scalable and adaptable, suitable for exploring other neurodegenerative diseases characterized by cellular complexity and heterogeneity, such as Alzheimer’s and ALS. By enhancing the resolution at which disease biology is understood, neural networks promise to uncover universal and disease-specific molecular signatures that could revolutionize diagnostics, prognostics, and treatment paradigms.

While the promises are vast, the research also underscores challenges inherent in data complexity, variability in human brain tissue samples, and the need for robust computational models trained across diverse datasets. Addressing these hurdles will require multidisciplinary collaborations among neurologists, computational biologists, and data scientists, fostering an ecosystem where artificial intelligence seamlessly integrates with clinical and experimental neuroscience.

Looking ahead, this pioneering study sets a precedent for the application of neural networks in precise cellular characterization within pathological contexts. The ability to decode gene expression landscapes at single-nucleus resolution empowers researchers to untangle the labyrinthine networks that govern neuronal health and disease. Beyond Parkinson’s, these insights herald a new era in neuroscience where machine learning augments human expertise to unlock the mysteries of brain disorders that have long confounded scientific inquiry.

Furthermore, the practical implications of this work extend to biomarker discovery and personalized medicine. By defining clear genetic signatures associated with distinct cellular dysfunctions in PD, clinicians may better stratify patients based on molecular profiles, enabling tailored therapeutic regimens. Early detection of these molecular changes through minimally invasive techniques could revolutionize patient outcomes, transforming Parkinson’s disease from a progressively debilitating disorder to a manageable condition.

In summary, the fusion of neural network analytics with single-nuclei transcriptomics marks a milestone in neurodegenerative disease research. This innovative study not only deepens our mechanistic understanding of Parkinson’s disease but also opens transformative paths towards targeted therapies and precision diagnostics. As artificial intelligence continues to evolve and integrate with biomedical science, the vision of conquering complex neurological diseases appears increasingly within reach, promising new hope for millions affected worldwide.

Subject of Research: Parkinson’s disease gene signatures identified through single-nuclei transcriptomics using neural networks.

Article Title: Neural networks reveal novel gene signatures in Parkinson disease from single-nuclei transcriptomes.

Article References:
Fiorini, M.R., Li, J., Fon, E.A. et al. Neural networks reveal novel gene signatures in Parkinson disease from single-nuclei transcriptomes. npj Parkinsons Dis. 11, 304 (2025). https://doi.org/10.1038/s41531-025-01147-0

Image Credits: AI Generated