Principles2Plan: LLM-Guided System for Operationalising Ethical Principles into Plans

By: Tammy Zhong · Yang Song · Maurice Pagnucco

10 December 2025 at 13:00

arXiv:2512.08536v1 Announce Type: new Abstract: Ethical awareness is critical for robots operating in human environments, yet existing automated planning tools provide little support. Manually specifying ethical rules is labour-intensive and highly context-specific. We present Principles2Plan, an interactive research prototype demonstrating how a human and a Large Language Model (LLM) can collaborate to produce context-sensitive ethical rules and guide automated planning. A domain expert provides the planning domain, problem details, and relevant high-level principles such as beneficence and privacy. The system generates operationalisable ethical rules consistent with these principles, which the user can review, prioritise, and supply to a planner to produce ethically-informed plans. To our knowledge, no prior system supports users in generating principle-grounded rules for classical planning contexts. Principles2Plan showcases the potential of human-LLM collaboration for making ethical automated planning more practical and feasible.

ToolMind Technical Report: A Large-Scale, Reasoning-Enhanced Tool-Use Dataset

cs.AI, q-bio.NC updates on arXiv.org

By: Chen Yang · Ran Le · Yun Xing · Zhenwei An · Zongchao Chen · Wayne Xin Zhao · Yang Song · Tao Zhang

8 December 2025 at 13:00

arXiv:2511.15718v2 Announce Type: replace Abstract: Large Language Model (LLM) agents have developed rapidly in recent years to solve complex real-world problems using external tools. However, the scarcity of high-quality trajectories still hinders the development of stronger LLM agents. Most existing works on multi-turn dialogue synthesis validate correctness only at the trajectory level, which may overlook turn-level errors that can propagate during training and degrade model performance. To address these limitations, we introduce ToolMind, a large-scale, high-quality tool-agentic dataset with 160k synthetic data instances generated using over 20k tools and 200k augmented open-source data instances. Our data synthesis pipeline first constructs a function graph based on parameter correlations and then uses a multi-agent framework to simulate realistic user-assistant-tool interactions. Beyond trajectory-level validation, we employ fine-grained turn-level filtering to remove erroneous or suboptimal steps, ensuring that only high-quality reasoning traces are retained. This approach mitigates error amplification during training while preserving self-corrective reasoning signals essential for robust tool-use learning. Models fine-tuned on ToolMind show significant improvements over baselines on several benchmarks.

Cell-free epigenomes enhanced fragmentomics-based model for early detection of lung cancer

Omics In Lung

By: Yadong Wang · Qiang Guo · Zhicheng Huang · Liyang Song · Fei Zhao · Tiantian Gu · Zhe Feng · Haibo Wang · Bowen Li · Daoyun Wang · Bin Zhou · Chao Guo · Yuan Xu · Yang Song · Zhibo Zheng · Zhongxing Bing · Haochen Li · Xiaoqing Yu · Ka Luk Fung · Heqing Xu · Jianhong Shi · Meng Chen · Shuai Hong · Haoxuan Jin · Shiyuan Tong · Sibo Zhu · Chen Zhu · Jinlei Song · Jing Liu · Shanqing Li · Hefei Li · Xueguang Sun · Naixin Liang

5 February 2025 at 19:00

Clin Transl Med. 2025 Feb;15(2):e70225. doi: 10.1002/ctm2.70225.

ABSTRACT

BACKGROUND: Lung cancer is a leading cause of cancer mortality, highlighting the need for innovative non-invasive early detection methods. Although cell-free DNA (cfDNA) analysis shows promise, its sensitivity in early-stage lung cancer patients remains a challenge. This study aimed to integrate insights from epigenetic modifications and fragmentomic features of cfDNA using machine learning to develop a more accurate lung cancer detection model.

METHODS: To address this issue, a multi-centre prospective cohort study was conducted, with participants harbouring suspicious malignant lung nodules and healthy volunteers recruited from two clinical centres. Plasma cfDNA was analysed for its epigenetic and fragmentomic profiles using chromatin immunoprecipitation sequencing, reduced representation bisulphite sequencing and low-pass whole-genome sequencing. Machine learning algorithms were then employed to integrate the multi-omics data, aiding in the development of a precise lung cancer detection model.

RESULTS: Cancer-related changes in cfDNA fragmentomics were significantly enriched in specific genes marked by cell-free epigenomes. A total of 609 genes were identified, and the corresponding cfDNA fragmentomic features were utilised to construct the ensemble model. This model achieved a sensitivity of 90.4% and a specificity of 83.1%, with an AUC of 0.94 in the independent validation set. Notably, the model demonstrated exceptional sensitivity for stage I lung cancer cases, achieving 95.1%. It also showed remarkable performance in detecting minimally invasive adenocarcinoma, with a sensitivity of 96.2%, highlighting its potential for early detection in clinical settings.

CONCLUSIONS: With feature selection guided by multiple epigenetic sequencing approaches, the cfDNA fragmentomics-based machine learning model demonstrated outstanding performance in the independent validation cohort. These findings highlight its potential as an effective non-invasive strategy for the early detection of lung cancer.

KEYPOINTS: Our study elucidated the regulatory relationships between epigenetic modifications and their effects on fragmentomic features. Identifying epigenetically regulated genes provided a critical foundation for developing the cfDNA fragmentomics-based machine learning model. The model demonstrated exceptional clinical performance, highlighting its substantial potential for translational application in clinical practice.

PMID:39909829 | PMC:PMC11798665 | DOI:10.1002/ctm2.70225

Reading view