Genomics

Zhida He, Xia Hu, Baichen Le et al.

This work highlights a critical gap in current safeguards for frontier LLMs, demonstrating that text-level protections are insufficient against emerging biological risks, which is important for AI safety researchers and policymakers.

Nicolas Huynh, Krzysztof Kacprzyk, Ryan Sheridan et al.

For genomic researchers needing interpretable models, DEFT provides a novel method that combines the interpretability of decision trees with the expressivity of deep learning, addressing the bottleneck of tree depth in sequence analysis.

Tianyu Liu, Wangjie Zheng, Rui Yang et al.

For clinicians diagnosing rare diseases, Hygieia reduces diagnostic delays and workload while improving accuracy, validated with real-world cases from Yale and Duke-NUS.

Han Zhang, Guo-Hua Yuan, Chaohao Yuan et al.

This work provides a generative cellular world model for in silico simulation of cell states and perturbation responses, addressing the need for virtual cells in biological discovery and perturbation screening.

Jiayuan Ding, Jianhui Lin, Ziyang Miao et al.

This work addresses the need for privacy-preserving and structurally-aware foundation models in single-cell genomics, enabling collaborative training across institutions without sharing raw data.

Xuefei Julie Wang, Hao Cui, Michael P. Brenner et al.

This work addresses the problem of systematic exploration in automated scientific coding for researchers using LLMs, offering a method to avoid local optima and unproductive loops.

Maciej Sypetkowski, Joanna Krawczyk, Łukasz Smoliński et al.

For biologists and computational researchers, OmicsLM bridges the gap between quantitative omics data and natural-language reasoning, enabling more interpretable and flexible analysis of transcriptomic data.

Daria Ledneva, Mikhail Nuridinov, Denis Kuznetsov

Provides a standardized evaluation framework for the genomic ML community to enable principled model comparison and selection.

Yusen Hou, Weicai Long, Haitao Hu et al.

This addresses the need for better tools in microbiology and biotechnology by assessing LLMs' potential for genomic interpretation, though it is incremental as it focuses on benchmarking rather than a new model.

Junda Ying, Yuxuan Wang, Bowen Yang et al.

For computational biologists studying cellular differentiation and lineage branching, USB provides a rigorous microscopic interpretation of birth-death events, addressing a key limitation of existing continuous optimal transport methods.

Yuhao Sun, Zekun Wu, Zixun Huang et al.

This work addresses the limitation of first-order dynamics in optimal transport for generative modeling and single-cell omics, which often fail to capture regulatory momentum and time-delayed responses, providing a more expressive model for these fields.

Xingzhong Zhao, Ziqian Xie, Islam et al.

For bioinformaticians analyzing large phenotype panels (e.g., from imaging or representation learning), TorchGWAS makes large-scale GWAS screening practical where existing tools are too slow.

Kamila Szewczyk, Sven Rahmann

This addresses the problem of efficient genomic data storage and access for bioinformatics researchers, offering incremental improvements in speed and compression.

Shiv Sundram

For computational genomics researchers, this work incrementally improves the efficiency of genome alignment pipelines by synthesizing existing optimizations.

Silas Ruhrberg Estévez, Nicolas Huynh, Tennison Liu et al.

This work provides an incremental improvement for biologists inferring cellular trajectories from scRNA-seq data, by incorporating cell-cell communication into the alignment process.

Yanan Li, Christina Yi Jin, Yuan Jin et al.

For researchers building multimodal LLMs, this work provides a theoretically grounded and empirically validated alternative to modular architectures, though it is demonstrated only on a specific biological domain.

Jonathan Liu, Kia Ghods

This work addresses the challenge of designing cell-type-specific regulatory elements for genomics applications, representing a novel method for a known bottleneck.

Yi Duan, Zhao Yang, Jiwei Zhu et al.

Provides interpretable and accurate prediction of DNA regulatory activity for biologists studying gene expression, with explicit mechanistic explanations.

Lei Huang, Chuan Qiu, Kuan-Jui Su et al.

This provides a scalable and robust solution for genotype imputation in genomic studies, addressing ancestry bias and rare-variant accuracy limitations, though it is incremental as it adapts existing transformer methods to this domain.

Yuheng Liang, Lucy Chuo, Ahmadreza Argha et al.

This highlights a critical problem for cancer therapy, as accurate pre-treatment prediction is needed to address patient resistance, but the findings are incremental, showing current models lack robustness.