AI is reshaping scientific research and innovation. Scientists can leverage AI to generate, summarize, mix, and analyze scientific data. AI models can find patterns in data that human scientists have missed, find connections between seemingly unrelated fields and phenomena, and even propose latest hypotheses to be tested.
An AI co-scientist is a collaborative, multi-agent AI system designed to help human researchers in generating, reviewing, and refining novel scientific hypotheses, research proposals, and experimental plans. It’s a virtual scientific partner that leverages advanced reasoning, interdisciplinary knowledge synthesis, and iterative feedback to speed up scientific discovery. It’s able to designing experiments, analyzing data, and testing ends in partnership with human experts and enabling rigorous, reproducible research.
Large language models (LLMs) are customized using knowledge and data from each text and non-text sources. The co-scientist uses this data to generate latest hypotheses and run simulations to check and validate these ideas. Human-in-the-loop collaboration is important.
This post explores how NVIDIA is powering these AI co-scientists. It showcases two agents being developed by Los Alamos National Laboratories (LANL) to deal with two of the hardest challenges in science today: inertial confinement fusion (ICF) hypothesis generation and cancer treatment.
AI co-scientist for ICF hypothesis generation
LANL and NVIDIA are collaborating on a multiphase process to develop a co-scientist for ICF fusion hypothesis generation.
Fusion is the method that powers the celebrities. Achieving energy generation through fusion on Earth is one among the best scientific challenges. Inertial confinement fusion (ICF) is a process that achieves nuclear fusion by rapidly compressing and heating a tiny pellet of fuel using intense energy sources like lasers, causing the nuclei to fuse and release energy. ICF can also be used to grasp the exotic properties of matter, similar to those present in the inside of Jupiter, and for national security purposes.
Because ICF is a highly coupled multiphysics, non-linear problem, predictability of large-scale codes stays a significant scientific challenge. This complexity arises because ICF requires simulation of several physical phenomena that may interact in unpredictable ways and operate across vastly different spatial and temporal scales. Results from experiments at large laser facilities can deviate from predictions attributable to changes in initial conditions or selection of target-design parameters. To speed up understanding and progress, it is important to leverage all available tools—including AI.
In the primary phase of the method, LANL is leveraging open source NVIDIA NeMo framework libraries, including:
- NeMo Curator for data curation
- NeMo 2.0 for continual pretraining and fine-tuning
- NeMo RL for reinforcement learning of the Llama Nemotron Super 1.5 model. It should turn out to be a more domain-aware reasoning model that could be used as the idea for constructing a trusted AI co-scientist
Figure 2 shows the steps involved in transforming Llama Nemotron Super 1.5 right into a reasoning model for ICF physics. The steps encompass preparing datasets for domain adapted pretraining (DAPT), supervised fine-tuning (SFT), and reasoning traces using open access documents from public datasets, CORE, arXiv and OSTI.gov covering physics and ICF.
To confirm that the model is becoming knowledgeable on ICF, academic and custom benchmarks are used, including questions generated by subject material experts.
The last word goal of this work is to unravel among the most difficult problems in fusion research, including improving the performance of ongoing ICF implosion experiments on the National Ignition Facility and the OMEGA laser. This involves developing and benchmarking scientific concepts against computational simulations and physical experiments.
By refining designs and integrating feedback from experimental results, the AI co-scientist will provide insights that can inform latest experiments at current and the following generation ICF facilities. This permits progress toward more efficient and reliable fusion energy solutions, while also addressing key questions referring to fundamental properties of matter and national security.
AI co-scientist for cancer treatment
Targeted alpha therapy (TAT) generally is a highly effective treatment against cancer when precisely delivered. The radioactive atoms emit energetic alpha particles that destroy nearby cancer cells. Nonetheless, imprecise targeting may cause these powerful emissions to wreck healthy tissues, resulting in unintended unintended effects.
To attenuate such collateral damage, TAT relies on specialized chelator molecules that bind and transport the radioactive atoms to tumor sites. Designing effective chelators that remain stable and selective inside complex biological environments stays a significant research challenge.
Since the metals utilized in TAT have large radii, only a few molecules are known to reliably bind to them. This limits researchers’ ability to use data-driven approaches in designing latest and improved therapeutic agents.
LANL is constructing an agentic AI discovery platform that mixes generative AI and simulation in a single workflow to discover latest and improved chelator molecules. By helping to rapidly search vast chemical spaces, this research is paving the best way for safer, more practical, and more targeted therapies.
AI plays a central role in answering fundamental design questions which are involved, similar to “What makes a great molecule?” and “Which molecules fit that behavior?” To facilitate the method, LANL has adapted the NVIDIA Llama Nemotron Super 1.5 and GenMol models to give attention to molecular discovery and optimization.
Workflow overview
On this workflow, the agent leverages Llama Nemotron Super 1.5 for hypothesis generation. Hypothesis generation works by prompting the LLM with an outline of the issue and an inventory of the hypotheses it has tested before. The LLM then identifies probably the most promising hypothesis for the following iteration of the invention loop, given its base knowledge and the assessment of prior hypotheses.
GenMol is then used to generate a set of molecules to check the hypotheses. GenMol produces molecules that resemble known drugs and could be tuned to satisfy scientific criteria, similar to the traits listed within the LLM hypothesis, based on prompts or scientists’ design requirements.
The method then moves to construct chemical complexes between the chelator and the radioactive atom, using Architector.
Next, the workflow shifts to give attention to computational modeling on the LANL NVIDIA-powered supercomputer, Venado. 3D molecular structures are designed using high-performance quantum simulations to predict key chemical properties.
This simulation data is finally used to evaluate the validity of the hypothesis proposed by the LLM, informing its next decisions. Each tools are packaged as NIMs that help to robotically select one of the best performance settings. With accelerated computing, scientists close the loop between hypotheses and generated data to rapidly adapt and cycle through further rounds of design and simulation.
Using this workflow, the LANL and NVIDIA joint team has already discovered molecules with improved binding energetics for Actinium atoms. The hypothesis-driven design accelerates the identification of one of the best molecules and highlights the properties that make them especially useful. This approach enables researchers to adapt the design process for more practical collaboration with AI, supporting further refinement of molecular candidates.
This work marks the start of a transformative effort to design latest molecules with real-world applications. The impact has the potential to be far-reaching, as chelators are good for cancer therapies and in addition excellent for rapid treatment of poisoning and efficient purification of metals, and other chemical applications.
Moving forward, the main target might be on evaluating its feasibility, integration with delivery systems, and potential safety implications.
“With NVIDIA, Los Alamos National Laboratory is pioneering the design and deployment of AI co-scientists in research,” said Mark Chadwick, Associate Laboratory Director for Simulation, Computing, and Theory. “These co-scientists enable rapid hypothesis generation and validation across complex disciplines. We’re combining domain knowledge with a mix of AI capabilities from NVIDIA to construct co-scientists which are purpose-built for our mission to tackle a few of humanity’s grandest challenges.”
This research used the Perlmutter supercomputer resources on the National Energy Research Scientific Computing Center (NERSC), a U.S. Department of Energy Office of Science User Facility.
Start constructing AI co-scientists
Leveraging AI for scientific discovery helps to speed up critical assessments, shorten development cycles, and unlock deeper scientific insights faster than ever before. To learn more about this work, join NVIDIA at SC25 for the LANL Reasoning Model for Fusion and Agentic AI for Molecular Discovery talks on the NVIDIA booth. To start constructing your personal AI co-scientist, explore NVIDIA NeMo and Nemotron.
Acknowledgments
Due to Ping Yang, Danny Perez, Logan Augustine, Pascal Grosset, Jiyoung Lee, Thomas Summers, Michael Taylor, Radha Bahukutumbi, and David D Meyerhofer for his or her contributions.
