General-purpose large language models (LLMs) that rely on in-context learning do not reliably deliver the scientific understanding and performance required for drug discovery tasks. Simply increasing model size or introducing reasoning tokens does not yield significant performance gains. To address this gap, we introduce the MMAI Gym for Science, a one-stop shop molecular data formats and modalities as well as task-specific reasoning, training, and benchmarking recipes designed to teach foundation models the 'language of molecules' in order to solve practical drug discovery problems. We use MMAI Gym to train an efficient Liquid Foundation Model (LFM) for these applications, demonstrating that smaller, purpose-trained foundation models can outperform substantially larger general-purpose or specialist models on molecular benchmarks. Across essential drug discovery tasks - including molecular optimization, ADMET property prediction, retrosynthesis, drug-target activity prediction, and functional group reasoning - the resulting model achieves near specialist-level performance and, in the majority of settings, surpasses larger models, while remaining more efficient and broadly applicable in the domain.

A new study provides a rigorous theoretical and numerical analysis of the accuracy of the method of characteristics (MoC), a key computational tool in nuclear reactor simulations. The results clarify how different source approximations affect error behavior, offering guidance for improving high-fidelity nuclear modeling. Importantly, the research findings serve as a theoretical foundation for code verification of MoC programs.

A new study published in Big Earth Data introduces the Cloud-Aware Mixture-of-Experts Linear Transformer U-Net (CA-MTransU-Net) for precise forest burned area (FBA) detection. By integrating Sentinel-1 SAR and Sentinel-2 optical data, the model overcomes traditional challenges such as cloud contamination and the high computational complexity of standard segmentation models. Achieving the highest mean Intersection-over-Union (mIoU) of 87.00% and an inference speed of 6.26 ms, this architecture significantly outperforms conventional benchmark models, offering a robust and efficient solution for post-fire environmental monitoring.

Meteor impacts may have helped spark life on Earth, creating hot, chemical-rich environments where the first living cells could take shape, according to research integrated by a recent Rutgers University graduate.

“No one knows, from a scientific perspective, how life could have been formed from an early Earth that had no life,” said Shea Cinquemani, who earned her bachelor’s degree in marine biology and fisheries management from the Rutgers School of Environmental and Biological Sciences in May 2025. “How does something come from nothing?” Cinquemani is the lead author of a scientific review, published in the peer-reviewed Journal of Marine Science and Engineering, examining where life may have first formed on Earth. The paper focuses on hydrothermal vents, places where hot, mineral-rich water flows through rock and emerges into surrounding water, creating the chemical conditions and energy gradients needed for complex reactions.

Soil salinization is a growing global threat that reduces crop yields and limits sustainable agriculture, especially in arid and semi-arid regions. A new study shows that carefully designed biochar amendments can significantly improve plant growth and soil health in these challenging environments by reshaping both plant metabolism and soil microbial communities.