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.
Jialu Wu, Yiheng Zhu, Xiaorui Wang et al.· JACS Au· 0 citations
Retrosynthesis planning is a crucial task in organic synthesis, and deep-learning methods have enhanced and accelerated this process. With the advancement of the emergence of large language models, the demand for data is rapidly increasing. However, available retrosynthesis data are limited to only millions. Therefore, we pioneer the utilization of the template-based algorithm to generate chemical reaction data, resulting in the production of over 10 billion reaction datapoints. A generative pretrained transformer model is subsequently developed for template-free retrosynthesis planning by pre-training on 10 billion generated data. Inspired by the strategies of large language models, we introduce reinforcement learning to capture the relationships among products, reactants, and templates more accurately. Experiments demonstrate that our model achieves state-of-the-art performance on the benchmark, with a Top-1 accuracy of 63.4%, substantially outperforming previous models. Computer-aided synthesis-planning methods have significantly assisted synthesis planning. In this work, the authors present RSGPT, a generative model pre-trained on ten billion data points, achieving state-of-the-art performance for synthesis planning
Yafeng Deng, Xinda Zhao, Hanyu Sun et al.· Nature Communications· 17 citations· ⚡2
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.
Shicheng Chen, Haiting Duan, S. Zhong et al.· Proceedings of the National...· 0 citations
Enzymatic reactions play an emerging role in a broad spectrum of scientific and industrial applications. The inherent complexity of enzymes, such as their substrate specificity, conformational flexibility, and the vast diversity of reactions involved, poses substantial challenges for the advanced computational prediction of enzymatic reactions with desirable accuracy. Moreover, existing approaches are mostly tailored for a specific sub-task, such as substrate prediction or binding site annotation, which limits their applicability. In this study, we introduce ERAM, a task-agnostic multimodal learning framework capable of addressing a broad range of downstream applications with both accuracy and efficiency. ERAM aligns pre-trained molecular representations from Protein Language Model with the knowledge of enzyme catalysis by modeling enzymatic reactions as multi-relational data. In enzyme retrieval tasks, ERAM achieves an improvement of 28.31% in mean average precision compared with the state-of-the-art (SOTA) method, CREEP. In substrate prediction tasks, ERAM outperforms the SOTA method ESP, achieving average improvements of 35.53% and 22.97% in Matthews correlation coefficient across two datasets. Additionally, ERAM exhibits commendable interpretability by assigning higher attention weights to binding sites, resulting in lower false-positive rates (42.36%) and higher overlap scores (70.59%) in the unsupervised binding site prediction task compared to RXNAA Mapper. By learning embeddings of substrates, enzymes, and products within a unified knowledge graph latent space, ERAM demonstrates its potential as a versatile and effective tool for enzyme catalysis research.
Multi-target drugs hold great promise for treating complex diseases, yet existing methodologies predominantly rely on ligand-based approaches, which lack sufficient biological context and are often confined to specific target pairs, resulting in limited generalizability. Here, we introduce LaMGen, a general-purpose multi-target drug design framework powered by large language models (LLMs). Built on MTD2025, a dataset comprising over 600,000 quantum-accurate molecular conformations and 700,000 multi-target associations, LaMGen directly yields energy-favorable conformations with quantum-level accuracy. The framework integrates ESM-C protein embeddings, rotation-aware ligand tokens, and a TriCoupleAttention module to capture multi-level target–ligand interactions. Across independent benchmarks, LaMGen outperforms diffusion-based model across multiple properties, generating molecules in an average of 0.44 s, while preserving high conformational plausibility. Retrospective analyses demonstrate that LaMGen not only can reproduce molecules identical to known actives, but also consistently produces structurally novel candidates with conserved core scaffolds and superior binding affinities. Designing effective multi-target therapeutics remains a major challenge, as existing ligand- or protein-centric methods struggle to generate biologically contextualized, spatially valid 3D molecules, particularly for triple-target systems. This study introduces LaMGen, an LLM-powered framework that leverages large-scale protein-ligand data and rotation-aware molecular encoding to rapidly produce chemically plausible multi-target candidates, achieving strong zero-shot generalization, superior molecular quality, and robust performance across dual- and triple-target design tasks.
Qun Su, Qiaolin Gou, Hui Zhang et al.· Nature Communications· 1 citation
Accurate prediction of blood-brain barrier permeability (BBBP) is essential for central nervous system drug discovery, yet existing models are often limited by their reliance on predefined physicochemical descriptors, small-molecule-centered training sets, or conformation-dependent representations, which restricts their transferability across chemically diverse modalities especially peptides. In addition, publicly available BBBP datasets remain fragmented, inconsistently standardized, and weakly controlled for molecular redundancy, increasing the risk of data leakage and overestimated model performance. In this study, we propose BBBP-Atlas, a structure-aware BBB permeability prediction model designed for unified modeling of small molecules and peptides with the first cross-modal dataset OmniBBBP. Designed to bypass descriptor and conformation dependencies, our model represents standardized molecular structures as atom-level graphs to capture local atom-bond environments and long-range topological dependencies associated with BBB transport. This design enables direct learning of structure-permeability relationships from molecular topology. For model training and evaluation, we curated a cross-modal, redundancy-filtered database OmniBBBP that seamlessly unifies small molecules and complex peptides, containing 10,218 unique compounds with 9,316 small molecules and 902 peptides. BBBP-Atlas achieved an accuracy of 0.8914 and an MCC of 0.7678 on the independent test set. On a balanced external benchmark of 200 compounds, our model reached an AUC of 0.9108, an accuracy of 0.8500, and an MCC of 0.7000, outperforming LightBBB by an absolute MCC gain of 6%. Case studies further showed that BBBP-Atlas captured clinically meaningful BBB permeability patterns, correctly identifying lorlatinib as BBB-permeable and vancomycin as BBB-impermeable with high confidence. The OmniBBBP-backed BBBP-Atlas offers a versatile and cross-modal approach for single-compound prediction, batch screening, and dataset exploration for CNS drug discovery. BBBP-Atlas is available at https://cadd.drugflow.com/bbbp/.
Xin Shen, Qun Su, Hao Luo et al.· bioRxiv· 0 citations
Targeting the intrinsically disordered N-terminal domain of the androgen receptor (AR-NTD) represents a promising strategy to overcome resistance in prostate cancer. However, its inherent lack of a stable tertiary structure and highly dynamic conformational ensemble pose formidable challenges for rational drug design. This study introduces an integrated computational workflow that combines enhanced sampling techniques and machine learning collective variables to identify druggable conformations of the AR-NTD and elucidate the binding mechanism of its modulator, EPI-002. We characterize nine metastable states of the Tau-5 region and reveal that ligand recognition is driven by π–π stacking and structured water-mediated hydrogen bonds. Leveraging these insights, we perform structure-based virtual screening based on the identified druggable conformations and identify K53, a rationally designed AR-NTD antagonist, which exhibits potent anti-proliferative activity in enzalutamide-resistant prostate cancer cells. K53 directly binds the AR-NTD, suppresses AR transcriptional activity, and demonstrates high selectivity for cancer cells. This work provides a rational design paradigm for targeting intrinsically disordered proteins and offers a therapeutic candidate for resistant prostate cancer. In this work, the authors develop a machine learning–based enhanced sampling workflow to target the intrinsically disordered AR-NTD, identifying druggable conformations and enabling transferable modeling of ligand binding for rational drug discovery.
Kai Zhu, Huating Wang, Jintu Zhang et al.· Nature Communications· 0 citations
This project is a prospective systematic review and meta-analysis examining the effects of stroboscopic visual training (SVT) on cognitive and perceptual–motor performance in elite athletes. SVT delivers intermittent visual occlusion through liquid-crystal shutter eyewear, forcing athletes to complete perceptual–decision–action sequences under incomplete visual information. Although prior meta-analyses report moderate benefits of SVT in general athlete samples, it remains unresolved whether these gains transfer to elite athletes, whose visual–motor systems are already highly specialized and may approach a functional ceiling. The mechanistic evidence for SVT likewise derives largely from non-elite populations. The review restricts its synthesis to elite athletes, defined by the McKay et al. (2022) framework as Tier 3 (highly trained/national) through Tier 5 (world-class).Its primary aim is to quantify the effect of SVT on two outcome domains — decision-making accuracy (the quality of perceptual–motor output) and reaction time (information-processing speed) — relative to normal-vision training, sham (clear-lens) intervention, active control, or no intervention. A secondary aim is to synthesize the candidate mechanisms by which SVT operates, including acute neural perturbation with neuroplastic adaptation, sensory reweighting with cognitive enhancement, and optimization of motor-control strategies. PubMed, Web of Science, Scopus, and CNKI will be searched from inception to 15 September 2026 with no language restriction. Effect sizes will be expressed as Hedges' g and pooled with random-effects models. The expected outcomes are that SVT effects in elite athletes will be smaller or more domain-specific than in mixed-skill samples, and that SVT will improve decision accuracy more reliably than raw reaction time — consistent with the view that occlusion training strengthens anticipatory, feedforward control rather than simple processing speed.
An atmosphere has one length in it, H=kT/(mg). This paper asks what that length decides──the answer is both the thickness and whether the atmosphere stays. No new mathematical theorem and no new law is claimed. Scope of this paper (scope note): No new mathematical theorem and no new law is claimed──scale height, the isothermal barometric law, and the Jeans parameter as an escape criterion are all standard. We do not build atmospheric science──all we use is one length and one ratio. We assume isothermality──a real atmosphere changes temperature with height, and H with it. We compute at one representative temperature. We do not treat escape mechanisms──hydrodynamic escape, non-thermal escape and solar-wind stripping are not entered. lambda is a guide, not a verdict. We do not follow compositional evolution──which gas leaves first, and when, is not treated. We claim no accuracy for the representative values──T and mean molecular weight are numbers for seeing orders of magnitude. Relation to earlier papers: Paper 309 showed that l_c=sqrt(gamma/rho g) is the length dividing gravity from surface tension──this is a third of the same kind, dividing gravity from heat this time. Paper 291 showed three lengths inside one ocean──this paper is the reverse: one length answering two questions. Paper 268 showed that the triple point’s being a “point” is settled by the phase rule before measurement──the same territory, pressure, and this paper looks at the distribution inside a single phase. Paper 276 showed the decibel is a ratio and not a quantity──R/H is a ratio too, and carries no dimension. Paper 316 showed the same integer returning from three counts──this paper is its mirror: one ratio answering two questions. What is added is computing the scale height for five bodies, tabulating H/R as a percentage, giving the pressure profile and the cumulative mass in numbers, and showing that the Jeans parameter equals R/H. First, on Earth it is 7.319 km.0.1149% of the 6371 km radius (Section 2). Second, pressure falls by 1/e every H. At Everest’s height, 0.2985 (Section 3). Third, 63.2121% of the mass sits below one H (Section 3). Fourth, this is the core of the paper. The number deciding escape is R/H itself (Section 4). Fifth, the Moon is at 25.20 and leaks. Earth 871.65, Titan 124.90 (Section 4). Sixth, the separator is which of gravity and heat wins, and it can only be written as a ratio (Section 5). An isothermal atmosphere has one length in it, H=kT/(mg), a ratio of heat to gravity that has come out as a length. On Earth it is 7.319 km──only 0.1149% of the 6371 km radius, 0.17 mm on a 30 cm globe, thinner than a sheet of paper. Pressure falls by 1/e every H, reaching 0.2985 at the summit of Everest──and 63.2121% of the mass sits below a single H. That 1-1/e depends on neither T nor g nor mu, and is the same on Venus, Jupiter and Titan. And the number deciding whether the atmosphere escapes comes out of that same H──putting g=GM/R^2 into the Jeans parameter lambda=GMm/(kTR) makes it exactly lambda=R/H. Earth stays at 871.65, Mars at 313.60, Titan at 124.90, and the Moon leaks at 25.20──Titan is smaller than the Moon and keeps its atmosphere because 94 K of cold compensates for the smallness. One thing separates them──which is larger, kT or mgR. And that can only be written as a ratio. H alone settles nothing──Jupiter’s H exceeds Earth’s, and at R/H=2804.51 it is far safer, so the ordering by length runs opposite to the ordering by safety. This is a third of the kind Paper 309 began with “the length dividing gravity from surface tension is not the size of the material”──H is not the size of the body but what the body is compared with. One last thing──one ratio is answering two questions. “How thin is it” and “does it stay” look different, and the answers are the two faces of one number. A thin atmosphere is an atmosphere in the act of leaving. On the making of this work: The ideas and content of this work stem from the author's own considerations. Assistance from an AI (a large language model) was used for structuring, English translation, and checking the algebra. Any remaining errors or misinterpretations are solely the author's. Feedback and corrections are sincerely appreciated. ----- 大気には H=kT/(mg) という長さが一つある。本稿が問うのは、この長さは何を決めているのかである──答は、厚さと、残るかどうかの両方である。新しい数学定理も新しい法則も主張しない。 本稿の射程(射程注記):新しい数学定理も新しい法則も主張しない──スケールハイト、等温大気の気圧分布、ジーンズ係数による散逸の目安は、いずれも標準的である。大気科学を作らない──使うのは一つの長さと、一つの比だけである。等温を仮定する──実際の大気は高さで温度が変わり、H も高さで変わる。本稿は一つの代表温度で計算する。散逸の機構を扱わない──ジーンズ散逸以外(流体力学的散逸・非熱的散逸・太陽風の剥ぎ取り)には立ち入らない。 lambda は目安であって判定ではない。組成の進化を追わない──どの気体が先に逃げるかの時間発展は扱わない。代表値に精度を主張しない──T と平均分子量は桁を見るための値である。既刊との関係:論文309 は l_c=sqrt(gamma/rho g) が重力と表面張力を分ける長さだと示した──本稿は同じ型の三本目であり、今度は重力と熱を分ける。論文291 は同じ海に三つの長さがあると示した──本稿は一つの長さが二つの問いに答えるという、逆向きの話である。論文268 は三重点が「点」であるのは相律が測る前に決めていると示した──気圧の話をする場所が同じであり、本稿は一相の中の分布を見る。論文276 はデシベルが比であって量ではないと示した──R/H も比であり、次元を持たない。論文316 は同じ整数を三つの数え方が返すと示した──本稿は一つの比が二つの問いに答えるという、鏡像である。加えたのは五つの天体でスケールハイトを計算したこと、H/R を百分率で並べたこと、気圧の高度依存と質量の累積を数で出したこと、ジーンズ係数が R/H に一致することを示したことである。 第一に、地球では 7.319 km である。半径 6371 km の0.1149%(第2節)。 第二に、高度 H ごとに 1/e になる。エベレストの高さで 0.2985 倍(第3節)。 第三に、H 一つぶんの下に 63.2121% が入っている(第3節)。 第四に、これが本稿の芯である。逃げるか残るかを決める数は R/H そのものである(第4節)。 第五に、月は 25.20 で漏れる。地球は 871.65、タイタンは 124.90(第4節)。 第六に、分離子は「重力と熱のどちらが勝つか」であり、それは比でしか書けない(第5節)。 等温の大気には H=kT/(mg) という長さが一つあり、熱と重力の比が長さになったものである。地球では 7.319 km──半径 6371 km の0.1149%にすぎず、直径 30 cm の地球儀なら 0.17 mm、紙一枚より薄い。高度 H ごとに気圧は 1/e になり、エベレストの頂上で 0.2985 倍──そしてH 一つぶんの下に 63.2121% の質量が入っている。この 1-1/e は T にも g にも mu にも依らず、金星でも木星でもタイタンでも同じである。そして大気が逃げるか残るかを決める数は、この H そのものから出る──ジーンズ係数 lambda=GMm/(kTR) に g=GM/R^2 を入れるとlambda=R/H に厳密になる。地球は 871.65、火星は 313.60、タイタンは 124.90 で残り、月は 25.20 で漏れる──タイタンが月より小さいのに大気を持つのは、94 K という冷たさが小ささを補っているからである。分けるものは一つ──kT と mgR のどちらが大きいか。そしてそれは比でしか書けない。 H 単独では何も決まらない──木星の H は地球より大きいのに R/H=2804.51 で地球よりずっと安泰であり、長さの大小と安泰さの大小が逆を向いている。論文309 が「重力と表面張力を分ける長さは材料の寸法ではない」と書いたのと同じ型の三本目である──H は天体の寸法ではなく、比べる相手である。最後に一つ──同じ一つの比が、二つの問いに答えている。「どれだけ薄いか」と「残るか」は別の問いに見えて、答は同じ数の表と裏であった。薄い大気とは、逃げかけている大気のことである。 作成にあたって:本稿の着想と内容は、著者自身の考察に基づくものです。文章の構成整理や英訳、数式の確認には AI(大規模言語モデル)の助力を得ました。最終的な内容の解釈や誤りがあれば、それらはすべて著者の責に帰します。お気づきの点があれば、ご教示いただければ幸いです。
Yuuki Yamagishi· Zenodo (CERN European Organi...· 0 citations
WHITE PAPER Dynamic Prompt Specialization via Semantic Micro-Model Routing STRATEGIC TECHNICAL MEMORANDUM: DPS Author: Valerii Khalif (VALEO), AVA LIVE Status: Draft for DOI / Zenodo submission Date: August 31, 2026 License: Open / free to use (attribution requested; voluntary donations welcome, not required) Abstract Modern development of large language models (LLMs) based on the Transformer architecture relies heavily on scaling: increasing parameter counts, training data volumes, and computational resources, alongside regular retraining and model updates. This document proposes an alternative, additive adaptation layer that does not require modifying the weights of the core LLM. The proposed approach is designated as Dynamic Prompt Specialization (DPS). DPS employs a small, high-speed auxiliary model—a semantic micro-model—that continuously analyzes the semantic and communicative structure of the current user input, extracts concise semantic fragments, and maps them to a predefined code table. Each code corresponds to a pre-engineered and validated prompt fragment, optionally accompanied by a confidence level or signal intensity score. Based on the active set of signals, DPS constructs a dynamic specialized context transmitted to the unmodified primary LLM. The primary goal of DPS is user interaction personalization: the system can adapt to stable individual communication patterns, ongoing conversational context, professional expertise level, task domain, feedback to previous answers, and other observable signals. Furthermore, DPS is not restricted to user communication. The same architecture can be applied to domain-specific pre-specialization of queries prior to reaching the primary LLM. This pre-structures multi-faceted queries and potentially reduces the necessity for dedicated specialist agents and complex orchestration pipelines. DPS operates without altering main model weights or requiring retraining. In the absence of a sufficiently confident signal, the system defaults to the baseline behavior of the main LLM. Consequently, DPS represents an orthogonal adaptation layer that can be integrated into existing LLM infrastructure without altering the underlying foundation model. 1. Problem Statement The proposed approach addresses two interconnected constraints in current LLM-based systems. 1.1. Scaling Does Not Solve Individual Specialization The main vector of foundation model advancement focuses on increasing: Parameter counts; Training data volume; Compute resources; Training duration and complexity; Model update frequency. However, expanding overall model capabilities does not yield a proportional enhancement in interaction quality for a specific individual user. In many practical scenarios, a user operates within a relatively bounded domain of knowledge and exhibits consistent individual patterns of task formulation, communication, verification, clarification, and correction. Thus, incremental general model knowledge often carries less practical utility than the system's ability to precisely align with a specific user's traits and the current interaction state. DPS shifts part of the specialization process from model parameter modification to dynamic context adaptation. 1.2. Static System Prompt Most LLM systems rely on a single primary system prompt or a narrow set of predefined instructions. As a result, interactions are typically treated as static: $$\text{System Instructions} \longrightarrow \text{User Request} \longrightarrow \text{Response}$$ DPS frames interaction as a dynamic process where conversational state evolves continuously: $$\text{Current User Signal} \longrightarrow \text{Analysis} \longrightarrow \text{Active Context Update} \longrightarrow \text{Response} \longrightarrow \text{New User Signal} \longrightarrow \text{State Update}$$ Thus, specialization occurs continuously rather than as a single static setup. 2. Core DPS Architecture DPS comprises three functionally separated layers, each holding distinct responsibilities. 2.1. Semantic Micro-Model – Fast Semantic Layer The first layer consists of a lightweight, ultra-fast auxiliary model. It is not designed for full reasoning or final output generation. Its scope is narrowly constrained to: Analyzing input utterances; Extracting short semantic fragments; Identifying corresponding semantic signals; Mapping signals to designated codes; Computing confidence or intensity levels for each signal. A typical semantic fragment may consist of a few words, though the architecture does not impose fixed-length constraints. Fragment size depends on input signal characteristics and implementation design. A single user turn can simultaneously trigger multiple independent codes. For instance, a query might concurrently express: Professional technical context; Financial domain context; Request for conciseness; Requirement for high precision; Emotional tone/signal; Affirmation of a previously established rule. Consequently, DPS is not forced to select a single category or agent, but instead builds a multidimensional set of co-active signals. 2.2. Code-Prompt Correspondence Table The second layer operates as a mapping repository: Code (e.g., PFA-0281) $\rightarrow$ Pre-engineered Prompt Fragment. A key principle of DPS is that prompt fragments are not automatically generated by the system on the fly. They are authored by engineers and undergo a controlled lifecycle: $$\text{Design} \longrightarrow \text{Testing} \longrightarrow \text{Verification} \longrightarrow \text{Approval} \longrightarrow \text{Deployment}$$ This establishes prompt engineering as a structured software engineering discipline. Codes map to behavioral, domain, or contextual specializations, such as: Technical engineering; Software programming; Financial analysis; Scientific discussion; Conversational interaction; Concise messaging; In-depth explanation; Critical evaluation; Re-affirmation of established context. 2.3. Dynamic Composite Prompt At the third layer, active codes are compiled into a unified composite context. Each signal maintains an intensity score, denoted internally as a temperature signal—a metric reflecting the activation strength of the given signal. Note: This term is distinct from the generation temperature parameter used in LLM sampling. DPS temperature measures signal activation strength rather than generative randomness. Based on these active signals, the system constructs a Composite Prompt passed to the primary LLM. Crucially, the primary LLM: Undergoes no retraining; Retains its original weights; Requires no specialized model variant; Executes a standard single generation pass. 3. Dynamic Temperature Profile – Breadcrumb State A core feature of DPS is that user profiling is not a one-off classification. It exists as a dynamic interaction state termed the Breadcrumb State. Profiling begins during the first exchanges of a conversation, eliminating the need to aggregate multiple historical sessions before initial activation. As dialogue progresses, individual signals can strengthen ($\uparrow$), weaken ($\downarrow$), disappear, be superseded, or interact with other active signals. Signal Dimension Trend Professional context $\uparrow$ Conciseness $\uparrow$ Need for detailed explanation $\downarrow$ Resistance to suggestion $\uparrow$ Confirmation of understanding $\uparrow$ This enables tracking not only topic content, but the evolving communicative dynamics of the session. 4. User Personalization The main purpose of DPS is adapting system behavior to an individual user. Across recurring sessions, a profile can persist (stored on the user side as a table of code mappings and signal temperatures) and evolve. Over time, the system continuously adapts to: Consistent professional contexts; Preferred task formulation styles; Desired response depth; Typical modes of confirmation or disagreement; Receptivity to suggestions; Preferred level of directness; Recurring communication patterns. This does not alter core LLM weights; instead, the interaction context shifts, allowing a single foundation model to serve diverse users through distinct dynamic adaptation profiles. 5. Professional Specialization Personalization does not degrade technical rigor. DPS can serve as a domain-specialization pre-layer between user requests and the primary LLM. When a user simultaneously addresses a technical problem, financial constraints, demands rigorous analysis, cites prior rules, and signals skepticism, traditional agentic architectures route the query through multiple stages: $$\text{Router} \longrightarrow \text{Technical Agent} \longrightarrow \text{Financial Agent} \longrightarrow \text{Context Aggregator} \longrightarrow \text{Main LLM}$$ DPS streamlines this pathway: $$\text{User Input} \longrightarrow \text{Semantic Micro-Model} \longrightarrow \text{Active Codes Selection} \longrightarrow \text{Composite Prompt} \longrightarrow \text{Main LLM}$$ Multiple context facets are integrated into a single pre-generation context, avoiding multi-agent pipeline overhead. 6. Reducing Agentic Complexity Modern AI implementations frequently rely on multi-agent orchestration: $$\text{Router} \longrightarrow \text{Specialist Agents (A, B, C)} \longrightarrow \text{Aggregator} \longrightarrow \text{Final LLM}$$ While effective, this paradigm increases inference latency, model call counts, compute costs, context size, and orchestration complexity. DPS offloads specialization that can be expressed via engineered prompt fragments to a lightweight micro-model prior to
Valeo Khalif· Zenodo (CERN European Organi...· 0 citations
Bacteriophage therapy is re-emerging as a potential strategy to address antimicrobial resistance, but standardized patient education materials are limited. Large language models (LLMs) are increasingly used for patient-facing medical information. The quality of LLM-generated responses to 20 patient-relevant questions was evaluated by 12 clinicians and research experts in bacteriophage therapy independently rated each response for accuracy, completeness, clarity, and tone/empathy using 5-point Likert scales. Expert suggestions for improvement were recorded. A total of 960 ratings were analyzed. Adjusted mean scores ranged from 3.36 to 3.96 across domains, indicating generally favorable evaluations for all models. Significant differences among LLMs were observed for completeness and tone/empathy (Holm-adjusted p = 0.042 for both), but not for accuracy or clarity. Differences were small in magnitude (Cohen’s d = 0.12–0.29). Claude scored significantly lower than the other models for completeness and tone/empathy, while Perplexity achieved the highest completeness scores. Experts recommended improvements for 34–40% of responses; wrong information was given in 20%. The best responses were revised into an expert-informed patient guide provided as Supplementary Material, presenting a hybrid model in which LLMs generate draft patient information that is subsequently refined by clinical experts, particularly in rapidly evolving therapeutic domains lacking standardized educational resources.
N. Walter, D. Amanatullah, Laurent Debarbieux et al.· npj Viruses· 0 citations
Background Bilingual education is increasingly understood as a socio-educational system shaped by learner, instructional, and institutional conditions. In multilingual contexts such as Vietnam, the effectiveness of Khmer-Vietnamese bilingual education (KhVBE) depends not only on linguistic competence but also on broader instructional and institutional processes. Purpose This study develops and empirically examines an ecosystem-informed framework to assess how learner-level, instructional, and contextual factors are associated with implementation-level perceptions of Khmer-Vietnamese bilingual education effectiveness. Methods Using a cross-sectional survey design, data were collected from 269 administrators and teachers across three Mekong Delta provinces. Partial Least Squares Structural Equation Modeling (PLS-SEM) was employed to examine a multi-layered model incorporating learner-related, instructional, and contextual determinants of KhVBE effectiveness. Findings Results indicate that information channels emerged as the strongest predictor of implementation-level KhVBE effectiveness, while Khmer language ability and professional development showed smaller but significant positive associations. By contrast, internal learner factors, Vietnamese language ability, contextual-institutional conditions, community factors, and family factors did not show significant direct effects within the tested model. Implications The study contributes to bilingual education scholarship by proposing an ecosystem-informed perspective on KhVBE implementation and provides policy-relevant insights for strengthening instructional coordination, heritage-language integration, and institutional communication systems in minority education contexts.
Minh-Quang Duong, Vu-Thanh-Tam Nguyen, Van-Tuong Nguyen et al.· 0 citations
What if pathology foundation models could do more with less? GigaPath-Flash and GigaTIME-Flash cut computational demands while maintaining strong performance, opening the door to larger studies and broader exploration. The post GigaPath-Flash and GigaTIME-Flash: Toward population-scale discovery with efficient pathology foundation models appeared first on Microsoft Research.
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
