explainable ai

Unified and explainable molecular representation learning for imperfectly annotated data from the hypergraph view

Molecular representation learning (MRL) has shown promise in accelerating drug development by predicting chemical properties. However, imperfectly annotation among datasets pose challenges in model design and explainability. In this work, we formulate molecules and corresponding properties as a hypergraph, extracting three key relationships: among properties, molecule-to-property, and among molecules, and developed a unified and explainable multi-task MRL framework, OmniMol. It integrates a task-related meta-information encoder and a task-routed mixture of experts (t-MoE) backbone to capture correlations among properties and produce task-adaptive outputs. To capture underlying physical principles among molecules, we implement an innovative SE(3)-encoder for physical symmetry, applying equilibrium conformation supervision, recursive geometry updates, and scale-invariant message passing to facilitate learning-based conformational relaxation. OmniMol achieves state-of-the-art performance in properties prediction, reaches top performance in chirality-aware tasks, demonstrates explainability for all three relations, and shows effective performance in practical applications. Our code is available in our https://github.com/bowenwang77/OmniMol public repository. AI models for drug discovery often struggle with real-world, incomplete data. Here, the authors present OmniMol, a framework using hypergraphs to improve predictions of molecular properties, addressing challenges of imperfect data annotation and enhancing model explainability.

Token-Mol 1.0: tokenized drug design with large language models

The integration of large language models (LLMs) into drug design is gaining momentum; however, existing approaches often struggle to effectively incorporate three-dimensional molecular structures. Here, we present Token-Mol, a token-only 3D drug design model that encodes both 2D and 3D structural information, along with molecular properties, into discrete tokens. Built on a transformer decoder and trained with causal masking, Token-Mol introduces a Gaussian cross-entropy loss function tailored for regression tasks, enabling superior performance across multiple downstream applications. The model surpasses existing methods, improving molecular conformation generation by over 10% and 20% across two datasets, while outperforming token-only models by 30% in property prediction. In pocket-based molecular generation, it enhances drug-likeness and synthetic accessibility by approximately 11% and 14%, respectively. Notably, Token-Mol operates 35 times faster than expert diffusion models. In real-world validation, it improves success rates and, when combined with reinforcement learning, further optimizes affinity and drug-likeness, advancing AI-driven drug discovery. In this work the authors present Token-Mol, a token-only 3D drug design model, which deploys the Gaussian cross-entropy (GCE) loss function for regression tasks. It exhibits superior performance in molecular conformation generation, property prediction, and pocket-based generation, thus opening up new avenues for drug design.

HiCLR: Knowledge-Induced Hierarchical Contrastive Learning with Retrosynthesis Prediction Yields a Reaction Foundation Model

Reaction representation learning is of paramount importance for adopting deep-learning-based chemistry modeling to solve real-world tasks such as synthesis planning. Most prevailing models are prestrained by self-supervised objectives that rely solely on the chemical structure information. Since structurally similar reactions could possess entirely distinct properties (e.g., reaction yields) and the synthesis-related tasks are highly heterogeneous, there are inherent limitations in constructing a foundational reaction model within the existing approaches. To tackle this limitation, we propose HiCLR, a knowledge-induced hierarchical contrastive learning framework for chemical reactions, by introducing relational inductive bias to forge chemically meaningful and generally applicable reaction fingerprints. Critically, the pretraining scheme combining both retrosynthesis prediction and contrastive loss enables HiCLR to tackle generation-based and understanding-based tasks simultaneously. Comprehensive experiments demonstrate that HiCLR successfully organizes the reaction space into hierarchical global semantic clusters, aligned well with prior knowledge. Consequently, HiCLR is the first foundation model that can be broadly applied to various synthesis-related tasks, and it achieves state-of-the-art performance in reaction classification, reaction condition recommendation, reaction yield prediction, synthesis planning, and even molecular property prediction. HiCLR demonstrates clear benefits in incorporating domain knowledge to guide the learning of neural networks, expediting AI-driven advancements in chemistry.

AI-driven PROTAC design overcomes oncogenic resilience by eliminating the CLIP1-LTK fusion protein.

The discovery of CAP-Gly domain-containing linker protein 1(CLIP1)-Leukocyte tyrosine kinase (LTK) as an oncogenic fusion reveals a unique dependency not only on LTK kinase activity but also on CLIP1-mediated multimerization, a noncatalytic function that drives oncogenic signaling. While this fusion is currently targeted with anaplastic lymphoma kinase inhibitors, their exclusive focus on kinase inhibition leaves the scaffolding function intact, necessitating a complete protein clearance strategy. Here, we report the AI-guided development of a first-in-class proteolysis-targeting chimera (PROTAC) designed to selectively degrade the CLIP1-LTK fusion protein. By integrating deep learning models for ternary complex prediction with structure-based molecular optimization, we designed DCL05, an orally bioavailable degrader of CLIP1-LTK fusion protein, achieving picomolar degradation potency (DC50 = 40 pM) and robust antitumor activity. DCL05 consistently outperformed existing kinase inhibitors across a broad spectrum of LTK resistance-associated mutations, both in vitro and in vivo. Collectively, our study explores resistance-associated contexts of LTK and establishes a structure-guided PROTAC development pipeline, providing a promising therapeutic strategy for overcoming acquired resistance in kinase-driven cancers.

k-jinma/dogo-tutor: dogo-tutor v0.0.2 - Initial Release

This is the initial release of "dogo-tutor," an AI tutor designed for coding education. This release is created to archive the repository and obtain a DOI via Zenodo. Key Features Provides step-by-step hints (problem clarification, strategy, logic structuring, and partial code) rather than direct answers. Designed to offer deeper hints only when students explain "what they tried and what happened" in their own words. Automatically generates review notes and short quizzes (/review) based on the last 24 hours of dialogue history, focusing on concepts the student struggled with. Features a self-reflection tool (/log) that allows students to objectively track their dependency on deep hints. Employs a safe guidance design that strictly references URLs from actual lecture materials and official documentation specified by the instructor. Recent Changes Revamped the README for users and added an English summary. Adjusted the icon design for the VS Code extension. Updated various configuration files (e.g., package.json) for public release.

Neuro-Symbolic Reasoning via Temporal Logic Embedding

This paper introduces a novel approach to neuro-symbolic reasoning by embedding temporal logic rules directly into the weights of a neural network. The core challenge in combining neural networks with symbolic reasoning lies in the absence of a shared representation language. Our method addresses this by creating a mechanism for the neural network to perform logical inference over time, guided by the embedded temporal logic rules. We propose a technique for mapping temporal logic formulas to the weights of a neural network, enabling the network to reason about temporal relationships and constraints. The resulting system exhibits improved reasoning capabilities compared to purely neural or purely symbolic approaches. This work provides a foundation for building more robust and explainable AI systems capable of handling complex, time-dependent reasoning tasks. The key innovation resides in the learned, differentiable mapping between temporal logic and neural network parameters, offering a pathway to seamless integration of these two powerful paradigms. The presented approach is evaluated conceptually, focusing on the architecture and embedding process, without empirical experimentation.

Explainable AI-Driven Synthetic Data Generation

This paper explores the application of explainable artificial intelligence (XAI) techniques to the generation of synthetic data. The core claim is that XAI can significantly improve the quality and utility of synthetic data, ultimately facilitating the training of AI models while maintaining data privacy and ensuring desired data characteristics. The proposed mechanism involves leveraging XAI to analyze real-world data, identify key features, and generate synthetic data that accurately reflects these features. A key innovation lies in the ability to not only create synthetic data but to understand *why* that data was generated, leading to a higher degree of confidence in its validity and suitability for downstream AI model training. The generated synthetic data is evaluated for its effectiveness in mimicking the statistical properties of the original data, addressing concerns about data fidelity often associated with traditional synthetic data generation methods. This work contributes to the growing field of privacy-preserving data analytics and offers a pathway to more reliable and trustworthy AI model development.

naganjaneyulu75/NeuroCareIoT: NeuroCareIoT v1.0.0

NeuroCareIoT v1.0.0 Initial release of NeuroCareIoT — Reliable Multimodal Edge-Based Alzheimer's Home Monitoring Using NeuroFuseNet. Features SafeFallNet for IMU-based fall detection WanderSenseNet for indoor localization and wandering-risk assessment DailyRoutineNet for temporal routine-deviation detection VitalRhythmNet for physiological anomaly assessment NeuroFuseNet for reliability-aware multimodal fusion Probability calibration using Platt Scaling and Temperature Scaling Predictive uncertainty using Monte-Carlo dropout Uncertainty-gated alerting Context-aware alert policies Modality reliability scoring Personalized resident baselines Drift-aware adaptation Explainable AI support Controlled synchronized multimodal replay ONNX edge-model export support Edge benchmarking utilities Ablation and statistical evaluation support Public Datasets This implementation supports: UP-Fall UJIIndoorLoc CASAS Aruba PPG-DaLiA Edge Deployment The framework supports deployment-oriented experiments using: Raspberry Pi 4B NVIDIA Jetson Nano ONNX Runtime TensorRT-compatible inference Important Note NeuroCareIoT is a research and experimental framework. The public datasets are modality-specific and were not collected as a naturally synchronized Alzheimer's cohort. System-level multimodal evaluation therefore uses controlled synchronized replay. This software is not a certified medical device and is not intended for autonomous clinical or emergency decision-making.

Explainable Artificial Intelligence (AI) Modeling of Co-Exposure Synergies in Tobacco Farming Health Risk Assessment with Epidemiological Analysis

HOW DO INFORMATICS TEACHERS PERCEIVE THE CHANGING PURPOSES AND PRACTICES OF INFORMATICS EDUCATION IN THE ERA OF GENERATIVE AI?

AbstractGenAI is increasingly capable of performing activities central to Informatics learning, includinggenerating, explaining, modifying, and debugging code, raising questions about what studentsneed to learn and how teachers' roles may change. This study explored Informatics teachers'perceptions of the changing purposes and practices of Informatics education in the GenAI era.An exploratory qualitative design was employed with 12 Informatics teachers from a collegespecializing in Informatics education. Data were collected through semi-structured interviewsand analysed using reflexive thematic analysis. Three interconnected themes were identified.First, teachers perceived a shift in emphasis from producing correct computational outputstoward understanding, explaining, evaluating, and modifying both independently produced andAI-generated solutions, while continuing to regard foundational programming knowledge asessential. Second, teachers described their roles as increasingly involving the evaluation,contextualization, and mediation of AI-supported learning rather than primarily providinginformation and solutions. Third, participants negotiated a contextual boundary between AI asassistance and AI as substitution, particularly when evaluating whether successful taskperformance represented genuine student competence. The findings suggest that GenAI does notsimply introduce a new instructional tool but challenges established assumptions aboutcomputational competence, pedagogical expertise, and evidence of learning. Informaticseducation may therefore need to balance independent computational competence with criticaljudgement and purposeful human-AI collaboration.

Pathway to achieve sustainable value through task-AI fit: the critical roles of green knowledge management capabilities, competitive advantage and financial resilience

Purpose This study aims to investigate how artificial intelligence (AI)-based technologies are associated with sustainable value intentions in green-oriented organisations operating under circular economy principles in China. Drawing on an adapted view of task-technology fit and the resource-based view (RBV), this study examines how technology and task characteristics are associated with Task-AI Fit, and how this fit is associated with green knowledge management (KM) capabilities, competitive advantage and financial resilience, which serve as key enablers of sustainable value intentions. Design/methodology/approach A survey was conducted with 373 representatives from Chinese organisations committed to circular-economy goals. Using covariance-based SEM and statistical tools, including SPSS Statistics and SPSS AMOS, the study tested the hypothesised paths, the structural model and mediation and moderation effects. It also examined the association between perceived task-AI Fit and internal capabilities, and their subsequent relationship with sustainable value intentions, while assessing the moderating role of environmental uncertainty. Findings The findings show that perceived task-AI Fit is significantly associated with Green KM capabilities, competitive advantage and financial resilience. Specifically, the Green KM capabilities show the strongest association with intentions to realise sustainable value. Environmental uncertainty significantly moderates only the relationship between Green KM capabilities and sustainable value intentions. Research limitations/implications By proposing a Task-AI Fit-driven RBV, the study highlights that the perceived Task-AI Fit is associated with internal capabilities that can channel sustainable intentions in the face of environmental uncertainty. Practical implications Organisations must prioritise, adapt and practice AI characteristics and features to strengthen AI-task alignment and strategically develop sustainability-oriented knowledge routines and value realisation. Social implications Future work can examine cross-cultural aspects across regulatory and cultural contexts, adopt a longitudinal design and include objective sustainability performance indicators to identify which potential attributes fail to materialise and to track the development of AI-driven capabilities. Originality/value This research extends the understanding of AI-enabled sustainability transitions by integrating task-technology alignment with resource-based capabilities in the context of environmental uncertainty. It highlights the central role of green KM practices in explaining sustainable value intentions in circular economy-oriented firms.

Using Artificial Intelligence to Achieve Inclusive and Effective Education

Artificial intelligence, as one of the transformative technologies of the contemporary era, has considerable capacity to respond to learners’ individual differences, reduce barriers to access to education, and improve the quality of the teaching–learning process. The present study was conducted with the aim of explaining how artificial intelligence can be used to achieve inclusive and effective education. The research method is library-based and analytical; accordingly, the data were collected through the study, classification, and analysis of documents, scientific reports, and reputable peer-reviewed articles in the fields of artificial intelligence in education, inclusive education, and learning technologies. The findings indicate that AI-based tools, including adaptive learning systems, intelligent tutors, learning analytics, speech-to-text technologies, text-to-speech technologies, and assistive technologies, can align educational content, learning pace, feedback, and assessment methods with learners’ diverse needs, abilities, and circumstances. These capabilities can provide more equitable opportunities for participation in education, particularly for students with special educational needs, learners in under-resourced areas, and individuals with linguistic or cultural differences. However, the real effectiveness and inclusiveness of these technologies depend on observing principles such as the protection of personal data, algorithmic transparency, the mitigation of bias, equal access to digital infrastructure, and the preservation of the teacher’s central role (Miao & Holmes, 2023; Zawacki-Richter et al., 2019). It can therefore be concluded that artificial intelligence, when used responsibly and through a human-centered approach, can move beyond being merely a technological tool and become a means of strengthening educational equity, personalizing learning, and increasing the effectiveness of education.