AI Applications in Scientific Software

A nice review article "Transforming Science with Large Language Models: A Survey on AI-assisted Scientific Discovery, Experimentation, Content Generation, and Evaluation" covers the scope of tools and approaches for how AI can support science. Some of areas the paper covers: (link in comments) 🔎 Literature search and summarization. Traditional academic search engines rely on keyword-based retrieval, but AI-powered tools such as Elicit and SciSpace enhance search efficiency with semantic analysis, summarization, and citation graph-based recommendations. These tools help researchers sift through vast scientific literature quickly and extract key insights, reducing the time required to identify relevant studies. 💡 Hypothesis generation and idea formation. AI models are being used to analyze scientific literature, extract key themes, and generate novel research hypotheses. Some approaches integrate structured knowledge graphs to ground hypotheses in existing scientific knowledge, reducing the risk of hallucinations. AI-generated hypotheses are evaluated for novelty, relevance, significance, and verifiability, with mixed results depending on domain expertise. 🧪 Scientific experimentation. AI systems are increasingly used to design experiments, execute simulations, and analyze results. Multi-agent frameworks, tree search algorithms, and iterative refinement methods help automate complex workflows. Some AI tools assist in hyperparameter tuning, experiment planning, and even code execution, accelerating the research process. 📊 Data analysis and hypothesis validation. AI-driven tools process vast datasets, identify patterns, and validate hypotheses across disciplines. Benchmarks like SciMON (NLP), TOMATO-Chem (chemistry), and LLM4BioHypoGen (medicine) provide structured datasets for AI-assisted discovery. However, issues like data biases, incomplete records, and privacy concerns remain key challenges. ✍️ Scientific content generation. LLMs help draft papers, generate abstracts, suggest citations, and create scientific figures. Tools like AutomaTikZ convert equations into LaTeX, while AI writing assistants improve clarity. Despite these benefits, risks of AI-generated misinformation, plagiarism, and loss of human creativity raise ethical concerns. 📝 Peer review process. Automated review tools analyze papers, flag inconsistencies, and verify claims. AI-based meta-review generators assist in assessing manuscript quality, potentially reducing bias and improving efficiency. However, AI struggles with nuanced judgment and may reinforce biases in training data. ⚖️ Ethical concerns. AI-assisted scientific workflows pose risks, such as bias in hypothesis generation, lack of transparency in automated experiments, and potential reinforcement of dominant research paradigms while neglecting novel ideas. There are also concerns about the overreliance on AI for critical scientific tasks, potentially compromising research integrity and human oversight.

As AI+Science went more mainstream in 2025, our team’s seminal contributions to the field are getting wide recognition. Here are top professional contributions and achievements for 2025. 1. Medical Imaging: We applied Neural Operators as a universal AI scheme that can handle any subsampling scheme and can do zero-shot super-resolution and field of view, without the need for any retraining. We applied it to a range of modalities such as MRI, CT, ultrasound, and photo-acoustic imaging. 2. AI Weather and Climate Models: I led the creation of the first AI-based weather model FourCastNet, built on Neural Operators, back in 2021. This year we announced FourCastNet 3, the fastest AI based model to provide calibrated probabilistic answers, crucial for extreme weather events. This also serves as backbone for state of art AI-based climate models . 3. De-Novo Inverse Design of Physical Devices: We were able to design new devices that were previously out of reach in challenging systems such as gate design in quantum dots, controlling quantum systems and non-linear photonics, using Fourier Neural Operator (FNO). 4. Scientific Modeling: FNOs achieved modeling of bio-realistic neurons, quantum dynamics and black holes with significant speedups while maintaining fidelity. 5. Millennium prize in fluid dynamics: We developed high-precision physics-informed neural networks (PINN) to solve a key step in computer-assisted proofs of singularities. 6. Physics-informed chemistry: Orbitall is the first universal quantum-chemical AI model that can handle at any spin, charge and external fields, and can extrapolate to larger molecules than those in training data. Nucleusdiff improves structure-based drug design by generating physically plausible molecules that maintain proper atomic spacing. We were able to beat previously known natural and engineered enzymes in functionality and versatility using protein-language models (genSLM). 7. Neural Operator foundations: We developed a unified framework to convert many popular neural networks like convolutional and graph neural networks, transformers etc to Neural Operators. We improved generalization to different geometries and scales. We developed FunDPS, a diffusion based inverse problem solver on function spaces. It is a resolution-agnostic unified framework for both forward and inverse PDEs. We also established limitations of hybrid learning that combine numerical solvers with learned closures and superiority of operator learning. 8. Verified Learning in LLMs: We released LeanDojo v2, LeanAgent and LeanProgress for theorem proving. 9. TIME 100 Impact Award and IEEE Kiyo Tomiyasu Award. 10. Group members Zongyi Li and Miguel Liu-Schiaffini winning best graduate and undergraduate research at Caltech commencement for work on Neural Operators and alum Zhuoran Qiao winning the Tianqiao and Chrissy Chen Institute AI+Science prize.

How do materials fail, and how can we design stronger, tougher, and more resilient ones? Published in #PNAS, our physics-aware AI model integrates advanced reasoning, rational thinking, and strategic planning capabilities models with the ability to write and execute code, perform atomistic simulations to solicit new physics data from “first principles”, and conduct visual analysis of graphed results and molecular mechanisms. By employing a multiagent strategy, these capabilities are combined into an intelligent system designed to solve complex scientific analysis and design tasks, as applied here to alloy design and discovery. This is significant because our model overcomes the limitations of traditional data-driven approaches by integrating diverse AI capabilities—reasoning, simulations, and multimodal analysis—into a collaborative system, enabling autonomous, adaptive, and efficient solutions to complex, multiobjective materials design problems that were previously slow, expert-dependent, and domain-specific. Wonderful work by my postdoc Alireza Ghafarollahi! Background: The design of new alloys is a multiscale problem that requires a holistic approach that involves retrieving relevant knowledge, applying advanced computational methods, conducting experimental validations, and analyzing the results, a process that is typically slow and reserved for human experts. Machine learning can help accelerate this process, for instance, through the use of deep surrogate models that connect structural and chemical features to material properties, or vice versa. However, existing data-driven models often target specific material objectives, offering limited flexibility to integrate out-of-domain knowledge and cannot adapt to new, unforeseen challenges. Our model overcomes these limitations by leveraging the distinct capabilities of multiple AI agents that collaborate autonomously within a dynamic environment to solve complex materials design tasks. The proposed physics-aware generative AI platform, AtomAgents, synergizes the intelligence of LLMs and the dynamic collaboration among AI agents with expertise in various domains, incl. knowledge retrieval, multimodal data integration, physics-based simulations, and comprehensive results analysis across modalities. The concerted effort of the multiagent system allows for addressing complex materials design problems, as demonstrated by examples that include autonomously designing metallic alloys with enhanced properties compared to their pure counterparts. We demonstrate accurate prediction of key characteristics across alloys and highlight the crucial role of solid solution alloying to steer the development of alloys. Paper: https://lnkd.in/enusweMf Code: https://lnkd.in/eWv2eKwS MIT Schwarzman College of Computing MIT Civil and Environmental Engineering MIT Department of Mechanical Engineering (MechE) MIT Industrial Liaison Program MIT School of Engineering

AI Meets Physics 🚀 Machine Learning is transforming physics - from predicting quantum behavior to simulating complex systems like climate and fluid flow. 📌 Key Applications: - Predictive Modeling for quantum mechanics and chaotic systems - Simulation & Analysis in fluid dynamics and climate science - Discovering Physical Laws using symbolic regression - Material Science innovations via property prediction - Quantum Computing optimization with neural networks 🧠 Popular Models in Use: - MLPs for general regressions - CNNs for image-based phase detection - RNNs for time-dependent physical processes - GANs for synthetic data generation - Encoder-Decoder models for forecasting & solving differential equations - Physics-Informed Neural Networks (PINNs) for integrating physics into ML ⚖️ Benefits vs Challenges ✅ High accuracy ✅ Speed and adaptability ✅ New scientific insights ❌ Black-box nature ❌ Heavy data/computation needs ❌ Risk of overfitting As AI continues to evolve, its role in physics is no longer optional—it’s becoming foundational. 🚀

🚀 Scientific Machine Learning: The Revolution of Computational Science with AI In recent years, we have seen impressive advances in Machine Learning (ML), but when it comes to scientific and engineering problems, a critical challenge remains: limited data and complex physical models. This is where Scientific Machine Learning (SciML) comes in—a field that combines machine learning with physics-based modeling to create more robust, interpretable, and efficient solutions. 🔹 Why isn’t traditional ML enough? Neural networks and statistical models are great at detecting patterns in large datasets, but many scientific phenomena have limited data or follow fundamental laws, such as the Navier-Stokes equations in fluid dynamics or Schrödinger’s equation in quantum mechanics. Training a purely data-driven model, without physical knowledge, can lead to inaccurate or physically inconsistent predictions. 🔹 What makes SciML different? SciML bridges data-driven models with partial differential equations (PDEs), physical laws, and structural knowledge, creating hybrid approaches that are more reliable. A classic example is Physics-Informed Neural Networks (PINNs), which embed differential equations directly into the loss function of the neural network. This allows solving complex simulation problems with high accuracy, even when data is scarce. 🔹 Real-world applications where SciML is already transforming science: ✅ Climate & Environment: Hybrid deep learning + atmospheric equations improve climate predictions. ✅ Engineering & Physics: Neural networks accelerate computational simulations in structural mechanics and fluid dynamics. ✅ Healthcare & Biotechnology: Simulations of molecular interactions for drug discovery. ✅ Energy & Sustainability: Optimized modeling of nuclear reactors and next-generation batteries. 🔹 Challenges and the future of SciML We still face issues such as high computational costs, training stability, and the pursuit of more interpretable models. However, as we continue to integrate deep learning with scientific principles, the potential of SciML to transform multiple fields is immense. 💡 Have you heard about Scientific Machine Learning before? If you work with computational physics, modeling, or applied machine learning, this is one of the most promising fields to explore! 🚀 #SciML #MachineLearning #AI #PhysicsInformed #DeepLearning #ComputationalScience

Today, we release a preprint describing a new AI system built with Gemini, designed to help scientists write empirical software. Unlike conventional software, empirical software is optimized to maximize a predefined quality score. Our system can hypothesize new methods, implement them as code, and validate performance by iterating through thousands of code variants. AI-powered empirical software has the potential of accelerating scientific discovery. Here is how it works (also on the visual graphic): ➡️ The system takes a "scorable task" as input, which includes a problem description, a scoring metric, and data for training and evaluation. ➡️ It generates research ideas, and an LLM implements these ideas as executable code in a sandbox. ➡️ Using a tree search algorithm, it creates a tree of software candidates to iteratively improve the quality score. ➡️ This process allows for exhaustive solution searches at an unprecedented scale, identifying high-quality solutions quickly. We rigorously tested our system on six challenging and diverse benchmarks and demonstrated its effectiveness. The outputs of our system are verifiable, interpretable, and reproducible. The top solutions to each benchmark problem are openly available. We look forward to taking this research through full peer-review. This new ability for AI systems to devise and implement novel solutions highlights AI’s capacity to help accelerate scientific innovation and discovery. The role of AI is evolving from a lab assistant to a collaborator that can transform the speed and scale of research. Read  the blog: https://lnkd.in/dPCZCCHS the preprint: https://lnkd.in/dQqfq8yg

The Rise of the AI Scientist Sam Altman recently predicted that within a year, AI will solve problems beyond human teams' reach — and we may see the first "AI Scientists" discovering new knowledge. That future is already here. FutureHouse just launched AI science agents that outperform human PhDs in research tasks: Crow - serves as a general research assistant Falcon- conducts lightning-fast literature reviews across full scientific papers Owl - identifies research gaps ripe for discovery Phoenix- designs chemistry and biology experiments These agents already surpass humans in precision, speed, and recall when analyzing scientific literature. Behind the scenes, more agents are training for hypothesis generation, protein engineering, and data analysis. We're not just getting AI help with science AI is starting to do the science. The Human Question What happens to the PhD when machines generate hypotheses? What does peer review look like when AI designs the experiments? Who gets credit for AI-driven discoveries? The answer isn't replacement, it's evolution Scientists become orchestrators, creative directors managing AI research networks. PhD programs may shift from "years of manual research" to "mastering scientific AI workflows." The possibilities are staggering: - Speed: Breakthroughs in days, not years - Access: Democratized top-tier research capabilities - Ambition: Tacklin previously impossible problems But critical questions remain: Can we trust AI findings? Who's accountable when AI fails? Will these tools serve everyone — or just tech giants? We're witnessing the biggest shift in knowledge creation since the scientific method itself. The next Nobel Prize might go to a team where AI did the heavy lifting. Small labs powered by agents might outperform entire university departments. This isn't the future of science. This is today. The question isn't whether AI will transform research — it's whether we'll guide that transformation thoughtfully.

I've presented our AI Integration Framework -- My Work | "With Me" Work | "For Me" Work -- a number of times recently and see it being an unlock in helping anyone, in any role, imagine how to partner with a digital collaborator. Having just wrapped up a call about bringing #AI to research scientists in pharma, here is output my #AIIntegration Analyzer generated right on the call as the #AIMap for Pharma Research Scientists. ### Role Overview - Pharmaceutical scientists in a collaborative research environment aim to design, conduct, and interpret experiments to discover and optimize new drugs. Their work spans molecular modeling, clinical trial design, lab testing, and regulatory strategy. AI presents transformative opportunities to speed up data analysis, simulate outcomes, and support complex decision-making while preserving human-led insight and ethical judgment. "My Work" – Human Exclusive Tasks Ethical Oversight of Trials: Interpreting ethical dilemmas in clinical trial design or patient treatment requires empathy, context sensitivity, and moral reasoning. Creative Hypothesis Generation: Scientists generate novel hypotheses based on gaps, intuition, and pattern-breaking thinking—something AI still cannot replicate well. Stakeholder Collaboration and Communication: Presenting findings to regulators, peers, or funding agencies demands persuasion, contextual framing, and relationship-building. "With Me" Work – AI Collaboration Opportunities Drug Discovery Simulations: AI can simulate molecular interactions at scale, identifying potential candidates faster than traditional trial-and-error approaches. Scientific Literature Review: AI tools can quickly summarize recent findings, highlight contradictions, and suggest areas of unexplored potential. Clinical Trial Design Optimization: AI can propose inclusion/exclusion criteria or simulate trial outcomes to help design better, more efficient studies. Data Visualization and Pattern Recognition: AI helps uncover trends across large datasets—gene expressions, patient responses, or assay results—guiding deeper human analysis. Drafting Grant Proposals and Protocols: AI can create first drafts of documents, enabling scientists to focus on refining arguments and adding critical insights. "For Me" Work – AI Automation Potential Data Entry and Preprocessing: Cleaning, labeling, and structuring lab data for analysis is time-consuming and error-prone—perfect for automation. Routine Report Generation: Weekly experiment summaries or compliance documentation can be automated with templates and data inputs. Lab Inventory Monitoring: AI can track chemical usage, alert shortages, and auto-order supplies based on trends and usage patterns. Conclusion - In pharma research collaborations, AI is a force multiplier. Scientists remain essential for guiding research, making ethical judgments, and interpreting results, while AI can dramatically speed up analysis, documentation, and design iterations.

From ChatGPT to ScienceGPT: AI is now learning the languages of physics, chemistry, biology, geology, and even nuclear science! A new preprint by Ameya D. Jagtap et al. offers a comprehensive review of foundation models across natural science domains. Beyond well-known chemistry models like ChemBERTa, MoLFormer, and MatterGen, it highlights breakthrough models across diverse fields, such as Aurora (climate forecasting), SeisT (earthquake detection), scFoundation (single-cell biology), RT-1 (robotics), and POSEIDON (solving partial differential equations). This broad overview reveals some key insights: 🔹Transformer architectures dominate across all disciplines 🔹Data scarcity remains the biggest bottleneck, not computing power 🔹Physics-informed models don't automatically outperform data-only approaches 🔹Domain-specific models may be more practical than universal ones (at least for now) 🔹Cross-domain transfer learning is still limited The field stands at an inflection point, shifting from narrow, task-specific tools toward AI systems that internalize scientific principles. While true universal scientific intelligence remains aspirational, we're steadily laying the groundwork. 📄 On Scientific Foundation Models: Rigorous Definitions, Key Applications, and a Survey, SSRN, August 27, 2025 🔗 https://lnkd.in/eMerEAXC

🔬 Why AI for Basic Research May Be More Impactful Than Clinical AI Clinical AI gets the headlines—diagnostic algorithms, treatment recommendations, automated radiology, digital pathology. But I think the AI tools with the broadest impact so far have been the ones powering basic research, and I am convinced this trend will continue and expand. Consider AlphaFold. According to DeepMind's five-year retrospective (https://lnkd.in/eMrBbVVe), over 3 million researchers across 190+ countries have used it. It's cited in 35,000+ papers. The 2024 Nobel Prize in Chemistry recognized its solution to a 50-year grand challenge. Research linked to AlphaFold is twice as likely to be cited in clinical articles and patents. 🔓 A race to democratize structural biology is underway. AlphaFold3 isn't fully open source, so Mohammed AlQuraishi's lab at Columbia built OpenFold (24 partners, six global pharma firms, https://openfold.io/), and MIT's Jameel Clinic released Boltz—predicting structure and binding affinity in seconds rather than hours. 🧬 David Baker —co-recipient of the 2024 Nobel Prize in Chemistry— and his Institute for Protein Design, University of Washington created RFdiffusion—a generative AI that works like image generators such as OpenAI's DALL-E or Google's Nano Banana, but for molecular structures (https://lnkd.in/eebviD5q). It designs novel proteins, including functional antibodies, entirely in silico—no animal immunization or library screening required. The software is free for both nonprofit and commercial research. 🧪 Microsoft's Skala may follow a similar path. For 60 years, density functional theory (DFT)—the workhorse of computational chemistry—has been limited by approximations 3-30x less accurate than needed for reliable predictions. Skala, a deep-learning model, achieves chemical accuracy (~1 kcal/mol) at a fraction of the computational cost (https://lnkd.in/e_beMVkQ). Thousands of scientists use DFT to simulate molecules and materials; Skala could shift the balance from lab experiments to computational design. 💡 Many other AI tools are emerging for basic research—foundation models for single-cell biology, perturbation prediction, genome editing design. It's too early to tell how impactful they'll be. But I think AlphaFold's lesson is clear: the most transformative AI solves hard problems that become infrastructure for everyone else. In a recent Nature Biotechnology comment (https://lnkd.in/evgr-Hkq), colleagues and I explored one example—how these tools are now essential for tackling "undruggable" cancer targets, about 85% of disease-relevant proteins.