The core idea is to pair large language model agents, the same class of AI systems that power tools like Claude or ChatGPT, with automated laboratory infrastructure capable of executing hundreds to thousands of experiments in parallel. Instead of relying on the published literature as its primary source of inspiration, the system will mine deidentified human data and molecular information from human cell and tissue samples to generate hypotheses about how cancer works. Those hypotheses will first be evaluated in computer modeling experiments using virtual tumors, AI models constructed from data on how real human cells respond when they are perturbed. Only after an idea survives simulation will it advance to physical testing on robotics platforms that manipulate human cells or organoids, the miniature three-dimensional tumors grown from patients’ own cells that have become a cornerstone of modern precision medicine.

Dr. Elemento argues that this architecture addresses two chronic weaknesses in how biomedical science is currently done. Previous efforts to build an AI scientist have concentrated on agents that read the scientific literature and propose hypotheses based on what has already been written. He considers those efforts valuable but fundamentally limited, because the literature is a secondhand and sometimes incomplete record of human biology, and an AI trained only on it can mostly recombine existing knowledge rather than surface genuinely new mechanisms. The second weakness is the traditional reliance on animal models tested one hypothesis at a time, a slow and expensive process whose results do not always translate to human patients. By grounding hypothesis generation in human data and testing many ideas simultaneously in human-derived systems, the discovery engine is designed to close both gaps at once.

The workflow he envisions is cyclical and self-correcting. AI agents propose biological mechanisms, the robotics platform tests them in parallel, and every round of experimental results is fed back into the AI, which learns from the outcomes and decides where the next experiments matter most. Virtual tumors serve as an intermediate filter, allowing the team to test an idea in simulation before committing laboratory resources to it. This iterative loop is what separates the project from simple prediction tools. As Dr. Elemento put it, prediction is not discovery; the goal is a system that proposes, tests and revises biological mechanisms the way scientists do, but at a scale and speed no human laboratory could match.

Human scientists remain firmly in the loop. Researchers will supervise the AI tools by setting research objectives, rejecting spurious hypotheses, and blocking experiments that are too costly or unlikely to succeed. They will also ensure that all work adheres to strict safety and ethical standards, a critical consideration for a system empowered to design and execute its own experiments on human-derived material. Dr. Elemento emphasized that the award will support the creation of an AI system to drive research from the initial data analysis through the validation stage, with human scientists’ oversight at every step, framing the technology as an amplifier of scientific judgment rather than a replacement for it.

The choice of lung cancer as the first target reflects the scale of unmet need. Patients with lung cancer currently face an overall five-year survival rate of fewer than 3 in 10, according to the American Cancer Society, and early-stage disease remains particularly difficult to manage. Dr. Elemento’s team will investigate why only some early lung lesions progress to invasive cancer, with particular attention to the hazy spots known as ground-glass opacities that appear on CT scans. By identifying the molecular and cellular changes that drive progression in these lesions, and by pinpointing where along that trajectory intervention might halt it, the researchers hope to open new therapeutic windows at the earliest and most treatable stages of the disease. As Dr. Elemento noted, there is a real need to discover new mechanisms that can be targeted pharmacologically, especially in the early stages of lung cancer.

The Englander Institute’s readiness for this challenge is no accident. Under Dr. Elemento’s leadership over the past five years, the institute has made major investments in data systems, data collection, robotics, and the development of disease models across many cancer types. Among these assets is a collection of more than 300 patient-derived organoid models spanning more than a dozen cancer types, a living library that gives the AI system an immediate and diverse testing ground. The institute has also assembled a very large dataset of information on the molecular and cellular features of lung cancer, which Dr. Elemento says now makes it possible to put everything together and let AI run the scientific process, with human oversight, in the hope of making major discoveries along the way.

Beyond lung cancer, the long-term ambition is to expand the platform to other types of cancer and, more broadly, to transform how science itself is conducted. Dr. Elemento envisions customizing AI agents for specific scientific roles, with separate agents specialized for hypothesis generation, critique, experimental design, or data analysis, each complementing the expertise of human scientists. Such specialized agents could analyze large, complex datasets that exceed human analytical capacity and enable the testing of personalized treatments tailored to individual tumors. He also wants to democratize access to these AI tools so that researchers without formal computer training can leverage them to advance and accelerate their own discoveries, spreading the benefits of automated experimentation far beyond institutions with deep computational resources.

The philosophical ambition underlying the project may be its most striking feature. For decades, Dr. Elemento observed, scientists have upgraded their instruments while keeping the same scientific method; this award, in his view, makes it possible to change the method itself. By entrusting part of the hypothesis-generation and testing cycle to AI agents that learn continuously from experimental outcomes, the program aims to establish a new AI-powered paradigm for research that makes large language models work for every scientist. If the discovery engine succeeds, the payoff could extend well beyond a single disease: a reusable, self-improving framework for biological discovery in which machines propose, robots test, and human researchers steer, all grounded in human data and validated in human-derived models. For a field where the path from observation to validated mechanism often takes years, compressing that cycle could reshape how the next generation of cancer therapies is found.

Subject of Research: An AI-driven discovery engine combining large language model agents, virtual tumor modeling and robotic organoid experiments to accelerate lung cancer research

Article Title: Dr. Olivier Elemento wins NIH Director’s Pioneer Award

Article References: Dr. Olivier Elemento wins NIH Director’s Pioneer Award. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: NIH Director's Pioneer Award, Olivier Elemento, artificial intelligence, large language models, lung cancer, organoids, robotics, precision medicine, Weill Cornell Medicine, virtual tumors, high-risk high-reward research, cancer discovery

The therapeutic concept at the heart of the new grant is one that Weinberger calls a neoplasm emergent oncolytic virus, or neovirus, a name that nods both to the existing class of tumor-busting viruses that selectively infect cancer cells and, playfully, to the film The Matrix. Oncolytic viruses are not new in principle; researchers have long sought viruses that preferentially replicate inside and destroy malignant cells while sparing healthy tissue. What distinguishes the neovirus concept is its dynamism. Rather than deploying a fixed, static agent, the therapy is engineered to undergo its own evolutionary changes, so that as a tumor accumulates mutations to escape attack, the therapeutic agent can counter-adapt. The work is still at an early stage, but Weinberger’s laboratory has already demonstrated proof of concept for a therapy capable of evolving alongside a tumor.

Weinberger’s target diseases reflect the most stubborn failures of modern oncology: pancreatic cancer, glioblastoma, and acute myeloid leukemia. Each of these cancers carries devastating survival statistics, and each is driven in part by extraordinary genetic diversity within the tumor itself. That heterogeneity is precisely what makes them so dangerous. A conventional drug, however potent, applies uniform selective pressure across billions of malignant cells, and among that vast population there is almost always a subset carrying mutations that confer resistance. Those survivors repopulate the tumor, and the treatment that once worked becomes useless. It is a pattern that repeats across cancer medicine and that Weinberger regards as a fundamental flaw in how therapies are designed.

His critique of that flaw is deceptively simple. Cancers, like viruses, are dynamic, evolving biological systems, yet for more than a century medicine has attacked them with static, chemistry-based technologies. The insight first struck Weinberger as a young graduate student studying the biophysics of HIV, when he became consumed by what he saw as a fundamental mismatch between viruses and the therapies built to treat them. The rational solution, he concluded, was to build a therapy that could itself mutate, so that it could not be circumvented by mutation. Instead of the drug being outmaneuvered, the therapy would evolve together with its biological target, locked in a continuous chase.

The intellectual framework behind this idea comes from evolutionary biology, specifically the Red Queen Hypothesis, named for the character in Lewis Carroll’s Through the Looking-Glass who tells Alice that it takes all the running you can do to keep in the same place. In a co-evolutionary arms race, a pathogen continually counter-evolves even as its host evolves resistance and attempts to escape, leaving both parties running in place. Weinberger aims to impose that same dynamic on cancer. Rather than a tumor escaping a fixed drug, the therapy would run after the tumor as the tumor tries to run away, establishing a scenario in which neither side can gain a permanent advantage but the disease can no longer simply outpace treatment.

The road to this point has been long and littered with failure. For more than twenty years, Weinberger recalls, colleagues dismissed the idea in polite and sophisticated ways, telling him the concept did not exist, that he would not find it, and that he should stop working on it. When he won his first Pioneer Award in 2013, he spent years engineering roughly 150,000 therapeutic candidates, and every one of them failed. The turning point came unexpectedly: a postdoctoral researcher in his lab brought him an image of the amputated genetic material the team had pursued for two decades, and it had evolved by itself in a flask at the back of the laboratory. That accidental foothold led to pre-clinical trials that behaved exactly as Weinberger had imagined twenty years earlier, protecting cells from HIV infection.

That HIV therapy is now in the final stages of FDA review for clinical trials, and the concept has already proven adaptable. In 2020, Weinberger’s team produced a COVID-19 therapeutic candidate based on the same evolutionary principles in roughly eighteen months. That speed matters for the cancer pivot. Because the group now has a working precedent for a therapeutic that can track the rapid evolution of HIV, Weinberger believes the transition to oncology can proceed far faster than the decades the original concept required. The same design logic, he argues, can be used to rapidly build a new class of therapeutics across multiple domains of medicine, including the deadliest cancers.

To carry the work from laboratory to patients, Weinberger moved to the Miller School in 2025, drawn by its Phase 1 clinical trial unit for experimental therapeutics. He describes Sylvester as a place designed to propel therapeutic concepts into the clinic, and he notes that clinical colleagues there have indicated that when the medical need is dire enough, they are prepared to move directly to testing in patients rather than delay a potentially life-saving medicine. That bench-to-bedside integration is paired with a demographic reality: the region’s rapidly expanding aging population represents the dominant risk factor for cancer, placing the new therapy’s potential patient population close at hand.

The stakes of the project are matched by its uncertainty, and Weinberger is candid about both. He acknowledges that it is not perfectly sane to follow an idea for decades while it keeps failing, but he argues that when success means a medicine that could help millions of people, persistence is the only reasonable course. If the reward is significant enough, he says, for example overcoming pancreatic cancer, it is worth the possibility of failure. With the Pioneer Award now funding the effort, an evolving-therapy precedent already validated against HIV, and a clinical infrastructure built for experimental treatments, Weinberger and his team are positioned to test whether the Red Queen can finally be made to run on the side of medicine, chasing tumors that have always, until now, run faster.

Subject of Research: Development of an evolving oncolytic virus cancer therapy to counter tumor mutation-driven drug resistance

Article Title: NIH honors Sylvester scientist developing cancer therapy that can evolve to outsmart tumors

Article References: NIH honors Sylvester scientist developing cancer therapy that can evolve to outsmart tumors. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: NIH Director's Pioneer Award, Leor Weinberger, oncolytic virus, cancer therapy, drug resistance, tumor evolution, pancreatic cancer, glioblastoma, acute myeloid leukemia, Red Queen Hypothesis, Sylvester Comprehensive Cancer Center, High-Risk High-Reward Research

As the head of the Computational Neurobiology Laboratory at Salk, Sejnowski’s latest project is poised to delve deeply into the intricate workings of neural circuits involved in working memory. With the NIH award funding $3.5 million over the next five years, his team intends to revolutionize our understanding by leveraging advanced computational models to dissect the dynamical activity of neuronal networks. These models aim to uncover principles that govern how information is temporarily encoded and maintained in the brain, a cornerstone process critical for decision-making and reasoning.

Working memory, a fundamental cognitive system, enables the transient storage and manipulation of information necessary for complex tasks. Despite its significance, the precise neural mechanisms underlying its robustness and capacity limitations remain enigmatic. Sejnowski’s approach utilizes state-of-the-art techniques in computational neuroscience, combining biophysically realistic neural network models with machine learning algorithms to simulate and predict the behavior of large-scale brain circuits. Such technological sophistication promises to elucidate how synaptic plasticity and circuit dynamics coalesce to sustain memory traces in the prefrontal cortex and hippocampus.

This research direction extends beyond theoretical exploration by aiming to address practical concerns related to neuropsychiatric conditions. Memory impairments are pervasive across multiple disorders, including schizophrenia, traumatic brain injury (TBI), and post-traumatic stress disorder (PTSD). By decoding the neural code of working memory with unprecedented granularity, Sejnowski’s team aspires to identify biomarkers and potential therapeutic targets that could alleviate cognitive deficits associated with these afflictions. The translational impact of this project cannot be overstated as it offers hope for interventions that restore normal cognitive function.

Throughout his distinguished career, Sejnowski has consistently pioneered transformative technologies and ideas in neuroscience. His earlier contributions profoundly shaped the study of neuroeconomics, neuroanatomy, neurophysiology, cognitive psychology, and artificial intelligence. A landmark achievement was his 1985 invention of the Boltzmann machine alongside Geoffrey Hinton, PhD, a quantum leap for neural network learning algorithms. The Boltzmann machine represents the first algorithm capable of learning internal representations in multilayer networks, providing a biologically plausible framework that continues to influence the development of AI to this day.

Beyond algorithmic advances, Sejnowski developed NETtalk, an early neural network system that mimicked human speech synthesis by learning to convert written text into phonetic output. This pioneering work laid a critical foundation for the field of deep learning and current technologies like ChatGPT, which simulate language and cognition through complex neural architectures. His dual focus on biological inspiration and computational implementation epitomizes the fusion of neuroscience and artificial intelligence.

Sejnowski’s influence extends deeply into neuroimaging advancements and the exploration of behaviorally relevant neural systems. His recent innovation involves a novel method for precisely measuring synaptic strength and plasticity, vital parameters underpinning learning and memory formation. By quantifying how synapses encode information and undergo activity-dependent changes, this technique sheds light on the synaptic basis of cognitive decline observed in aging and neurodegenerative diseases. Such insights are critical for developing strategies to preserve cognitive function across the lifespan.

Recognition of Sejnowski’s scientific excellence is reflected in the numerous prestigious awards he has garnered over the years. Most notably, he was honored with the 2024 Brain Prize and the 2022 Gruber Prize in Neuroscience, reflecting his outstanding contributions to understanding brain function. These accolades underscore his role as a luminary whose work bridges multiple disciplines and drives progress in the neurosciences.

A testament to his standing in the scientific community, Sejnowski holds positions in several elite academies, including the United States National Academy of Sciences, National Academy of Medicine, National Academy of Engineering, and National Academy of Inventors. He is also a Fellow of the Royal Society in the United Kingdom and a member of the American Philosophical Society, highlighting his global influence as a thought leader and innovator.

The Salk Institute, where Sejnowski conducts his research, is renowned for its relentless pursuit of groundbreaking knowledge in biological sciences. Founded by Jonas Salk, the developer of the first effective polio vaccine, the institute embodies a culture of bold scientific exploration. Within its walls, scientists continually push the boundaries of understanding in neuroscience, cancer biology, aging, immunobiology, and computational biology, among other fields. Sejnowski’s work fits seamlessly into this mission, combining innovation with impact.

Sejnowski’s ongoing commitment to deciphering the biological basis of cognition, learning, and memory signifies a new chapter in brain science. As computational tools grow increasingly sophisticated, melding detailed biophysical models with AI-driven analytics, his laboratory stands at the forefront of unraveling the complexities of neural information processing. This work holds profound implications not only for fundamental neuroscience but also for the development of targeted treatments for cognitive impairments.

With the NIH Director’s Pioneer Award supporting this ambitious endeavor, Terrence Sejnowski’s visionary research will continue to illuminate the neural architecture of working memory. His innovative approach exemplifies the power of interdisciplinary science to tackle some of the most challenging questions in brain function and dysfunction. The coming years promise remarkable advances from his lab that could redefine therapeutic strategies for mental health disorders and enhance our comprehension of the human mind.

Subject of Research: Neural circuit dynamics of working memory and computational modeling of cognitive processes in health and disease.

Article Title: NIH Director’s Pioneer Award Enables Terrence Sejnowski’s Cutting-Edge Computational Exploration of Working Memory Circuitry

News Publication Date: October 13, 2025

Web References:
– https://www.salk.edu/scientist/terrence-sejnowski/
– https://commonfund.nih.gov/pioneer
– https://www.salk.edu/news-release/upgrading-brain-storage-quantifying-how-much-information-our-synapses-can-hold/
– https://www.salk.edu/news-release/salk-professor-terrence-sejnowski-wins-brain-prize/
– https://www.salk.edu/news-release/salk-institutes-terrence-sejnowski-awarded-gruber-prize/
– https://www.salk.edu/news-release/terrence-sejnowski-elected-to-the-royal-society-and-the-american-philosophical-society/

Image Credits: Salk Institute

Keywords: Life sciences, Neuroscience, Cognitive neuroscience, Computational neuroscience, Memory, Cognition, Cognitive psychology, Psychological science, Social sciences