AI for science & health

AI applied to biology, chemistry, physics, medicine, climate and other sciences

67 articles · page 1 of 2

AI for science & health

Cancer AI models follow false statements about their own drug resistance findings

The authors evaluated ten language models on 1,464 cancer drug resistance cases from five expert-curated databases. They found that models often recommended therapies despite curated evidence of resistance. More strikingly, when models were told they had concluded a drug is effective (although they had indicated resistance), they endorsed it at much higher rates, revealing a vulnerability to fabricated conclusions.

AI for science & health

Instructor-designed AI boosts learning in 19-university trial

In a pre-registered randomized controlled trial across 19 universities and three semesters, researchers tested whether AI designed by frontline instructors can improve learning. They found that instructor-crafted AI innovations boosted student performance on proctored exams by 0.32 standard deviations, a substantial effect compared to typical education interventions. Student engagement and satisfaction remained unchanged.

AI for science & health

AI agents nearly triple success rate in solving crystal structures from powder X-ray diffraction

The authors formalize powder X-ray diffraction (pXRD) structure determination as a verifiable agentic search and introduce pXRD-Bench, a 200-problem benchmark. They show that adapting an LLM agent with a coding harness, crystallographic software, and verifier-informed training raises exact structure recovery from 27% to 69%. For experimental multiphase patterns, a reflexive evolution framework that learns from self-generated synthetic problems improves profile agreement and can recover a phase whose structure is absent from the reference library.

AI for science & health

Simple equations rival neural networks for predicting monsoon rainfall

Ferretti et al. used symbolic distillation, a two-step process combining neural networks and symbolic regression, to derive closed-form equations for South Asian monsoon precipitation from 21 years of reanalysis data. Their best equation explains 1.5 times as much precipitation variance as a standard buoyancy diagnostic and recovers about 85% of a neural network's skill, while using roughly a thousand times fewer parameters. The resulting formulas treat moisture and instability as competing controls rather than additive contributions, with heaviest rain when the lower free troposphere is humid.

AI for science & health

Physics-informed transport enables faster, more accurate inverse problem solving

The authors propose AdjointFlow, a generative framework for scientific inverse problems that injects known physics directly into the transport process rather than relying on a Gaussian reference distribution. By using an adjoint-based initialization and a noise-free velocity formulation, the method produces more accurate reconstructions than conditional diffusion models and achieves inference up to five orders of magnitude faster than plug-and-play methods.

AI for science & health

AI estimates global precipitation more accurately, especially during extreme events

Shalev and colleagues trained an AI model to estimate precipitation worldwide from atmospheric analyses and satellite imagery, using sparse spaceborne radar measurements as training targets. They report improvements over five established precipitation products, with the largest gains at extreme intensities, and show that the resulting estimates also improve precipitation forecasts when used to train a weather model.

AI for science & health

AI agents find cheaper quantum programs and mathematically prove their correctness

Shirobokov and colleagues used teams of AI agents to discover quantum programs and prove both their correctness and resource bounds in Lean, a software system for checking mathematical proofs. The authors report improvements over prior baselines ranging from 1.41-fold for parallel search to 290-fold for one arithmetic operation, although their scientific applications remain beyond projected early quantum hardware.

AI for science & health

AI forecasting model adapts to sensor changes, guarantees uncertainty at new locations

The authors propose OmniST, a topology-invariant spatiotemporal forecaster that does not depend on the number of sensor nodes. Composed with inductive split-conformal calibration, it provides distribution-free prediction intervals with coverage guarantees at unseen locations. On real air quality, weather, and traffic data, it achieves competitive accuracy and degrades minimally under sensor dropout.

AI for science & health

Satellite data improves AI species maps in remote areas by 51%

The authors trained a neural network called OCELOT on 2 billion species occurrence records from GBIF and satellite foundation embeddings to predict relative species occurrence. Satellite features improved geographic extrapolation by 51% over models using only coordinates, while matching or exceeding existing baselines on standard interpolation benchmarks. The model handles 100,000 taxa and shares statistical strength across related species to benefit rare ones.

AI for science & health

Structured AI chatbots prepare patients for clinic visits better than open-ended ones

Researchers developed an LLM-based agent that leads patients through a structured workflow: gather context, set expectations, provide health information, suggest questions, and summarize. In a three-arm randomized study across 802 US adults on the Google Health app, the agent significantly outperformed a minimally prompted control chatbot. Both independent clinician reviews and patient surveys showed consistent improvements, indicating that structured AI guidance can safely enhance pre-visit preparation.

AI for science & health

Language models rival trained baselines on molecular properties from a few examples

The authors compared seven frontier language models with eight classical baselines on four molecular property regression tasks, using in-context examples and a self-reported uncertainty. On three datasets verified to be absent from pre-training data, the best language model outperformed every baseline at 30 examples, though the advantage shrank with more context. The models' self-reported uncertainty ranked prediction errors about as well as the baseline uncertainties, with Spearman correlations up to 0.46.

AI for science & health

Transformer-like networks predict electron densities accurately enough to skip self-consistent calculations

Jeheon Woo introduces machine learning models, SwinUNETR-V2 and NeXtUNETR, based on Transformer-inspired architectures that predict electron densities with high accuracy. These models enable bypassing the iterative self-consistent field (SCF) calculation in density functional theory (DFT), achieving median energy errors as low as 0.036 meV and accurate forces for geometry optimization and molecular dynamics.

AI for science & health

On-the-fly GPU mesh preprocessing adds just 2% overhead to physics AI training

The authors present PhysicsNeMo-Mesh, a GPU-native mesh module for PyTorch that stores simulation cases with full semantic structure (named boundaries, field ranks). On one GB200 GPU, mesh operations run 6.2–1,068× faster than VTK with 72 threads, and loading a whole case is 8–88× faster. Overlapped with training, this preprocessing costs only 2.1% of throughput, enabling one dataset to serve multiple models without repeated curation.

AI for science & health

AI world models solve tasks without faithfully capturing how environments change

The authors introduce faithfulness as an equivariance condition between environment transitions and learned latent dynamics. They propose latent rollout consistency (LRC) and post-hoc state decoding to evaluate action-conditioned latent prediction. Across three state-of-the-art latent world models on simple control environments, they find that high task reward does not correspond to faithful latent dynamics, especially when learning from pixels.

AI for science & health

AI model estimates treatment effects faster without retraining for each dataset

The authors introduce PICLE-PFN, the first foundation model for longitudinal causal effect estimation that supports both static and dynamic treatment policies. The model represents target policies as in-context demonstrations, is pretrained on synthetic structural causal models, and predicts conditional average potential outcomes directly. On benchmarks it matches or outperforms conventional estimators while being about 100 to 5000 times faster at deployment.

AI for science & health

AI models can reuse learned dynamics across tasks and physical conditions

The authors formalize action learning as recovering a stochastic action from trajectory observations. They introduce LEAP, which learns a correction term relative to a tractable reference process, yielding a differentiable path-space objective that captures both local dynamics and nonadditive temporal dependencies. The learned action can be reused for path ranking, mechanism steering, and constrained optimization across different physical conditions.

AI for science & health

Memory-augmented Mamba branch improves GraphCast forecast error by 14 percent

The authors augment a frozen GraphCast backbone with a trainable Mamba correction branch that retains temporal memory across forecast steps. After training only the branch on ERA5 data from 2015–2021, they report a 14.42% reduction in the normalized weighted MSE over ten-day forecasts and a 21.66% reduction at the 240-hour lead. At day 10, RMSE decreases by 7–14% across all evaluated variable groups.

AI for science & health

Probabilistic AI forecasts retain skill for record-breaking weather extremes

Van Velthoven et al. evaluate four probabilistic AI weather models against the physics-based IFS-ENS ensemble for forecasting record-breaking temperatures and wind over land in 2022. The authors find that AI models outperform IFS-ENS for heat and cold records at nearly all lead times and record depths, with the advantage strengthening for deeper heat records. However, all systems show reduced skill for wind and as lead time increases.

AI for science & health

Predicting synthesis conditions enables resource-aware materials planning

Noh et al. propose CondMSP, a framework that extends materials synthesis planning from recipe prediction to execution-aware planning. By jointly predicting synthesis conditions and using them as process context for operation sequences and as constraints for resource allocation, they achieve a reference-recipe recovery rate of 32.9% under a budget of three furnace runs (compared to 27.9% for baselines) and 40.0% at 30 trials (compared to 34.7%).

AI for science & health

AI models excel at cyclone track prediction but struggle with intensity and uncertainty

Chalumeau et al. benchmarked three AI weather models—GenCast, FourCastNet 3, and FGN—against ECMWF's operational ensemble (IFS-ENS) on 178 tropical cyclones. FGN achieved better probabilistic track scores than IFS-ENS for 90% of 2025 storms, a clear improvement in location forecasting. However, all models underestimated peak wind speeds by 10–15% and showed overconfident ensemble spreads, indicating that the gains are concentrated in spatial structure rather than intensity.

AI for science & health

Orchestrating coding agents broadens autonomous computational research beyond optimization

Lange and colleagues present The Agentic Scientist, a pipeline that divides computational research into hypothesis generation, planning, experimentation, verification, analysis and manuscript writing. Across detailed case studies and 70 research campaigns, the authors report that coordinated coding agents produced testable findings and research artifacts, while orchestration and evidence preservation helped prevent the process from collapsing into simple performance optimization.

AI for science & health

Lean proof assistant generates step-level labels to train natural-language proof verifiers.

The authors introduce LeanPRM, a framework that uses the Lean proof assistant to generate step-level correctness labels for natural-language proofs without human annotation. A model writes paired natural and formal proofs in one pass; Lean checks the formal steps, and a second model filters out failures caused solely by formal syntax. Training a Qwen3-4B verifier on the resulting 152K labeled steps raises exact step-label accuracy from 48.2% to 74.7% on a held-out benchmark, outperforming verifiers trained on human labels from PRM800K.

AI for science & health

Mixing simulation fidelities lowers data generation costs for neural PDE surrogates

Setinek et al. formalize neural surrogate training as a function of both data quantity and numerical fidelity under a fixed data generation compute budget. Across over 7,000 training runs, they find that optimal multi-fidelity mixtures consistently match or beat single-fidelity training at equal compute, with savings up to 6.3x. The results hold across two distinct error sources and four surrogate architectures.

AI for science & health

Competition challenges AI models to optimize HVAC control across unfamiliar buildings

The authors introduce a competition built on the Smart Buildings Control Suite, in which participants train HVAC control policies on procedurally generated simulated buildings and are evaluated on held-out buildings and climates. The goal is to determine whether a building foundation model can transfer across buildings with minimal reconfiguration. The authors argue that even a 5% reduction in building emissions could cut total US emissions by roughly 1%.

AI for science & health

Natural language can steer all-atom protein design across diverse objectives

Lee and colleagues developed Prompteina, a model that jointly generates protein sequences and all-atom structures conditioned on natural-language descriptions of desired properties. Across structural-control and binder-design benchmarks, it generally outperformed models requiring specialized input formats, and nine of 89 experimentally tested designs bound a previously unaddressed target.

AI for science & health

Pretraining on synthetic primitive patterns enables robust zero-shot multivariate time series forecasting

The authors introduce Timer-M1, a multivariate time series foundation model pretrained on a pipeline that synthesizes elementary temporal patterns (trend, seasonality, regime changes) and cross-variate relations. By assembling real and generated series into forecasting episodes with varied variate roles, the model learns to generalize across domains and task configurations. On three large-scale benchmarks, Timer-M1 ranks first on FEV and TIME and second on GIFT-Eval among recent time series foundation models.

AI for science & health

Scaling law for physiological waveforms favors more training data over larger models

The authors train over one hundred physiological waveform foundation models (20M to 2.1B parameters) on up to 36.3M hours of data and derive a scaling law that predicts downstream clinical performance. They find that compute-optimal allocation scales pretraining hours much faster than model size, and that larger models use data more efficiently, reducing both waveform and labeling requirements for a given performance target.