arXiv subjects

Zhengwei Xie

Publications and source records attributed to Zhengwei Xie.

11 recordsLinked to original sources

arXiv2026

MemPrism: Task-Conditioned Relational Memory Views for Long-Horizon Agents

Long-horizon agents rely on memory to reuse experiences, yet existing memory systems often assume that evidence can be directly consumed through a fixed representation. This leads to representation mismatch, where relevant information is available but not organized for the current decision. To this end, we propose MemPrism, a task-conditioned relational memory framework that separates persistent experience storage from decision-time working memory. MemPrism records interactions as the event stream and dynamically constructs relational views according to the current task context. A lightweight view policy selects the relation structure, evidence range, outcome condition, and granularity, while a deterministic composer and render transform historical facts into a temporary optical working-memory view for a frozen task policy. Experiments on long-horizon embodied and web-agent benchmarks show that MemPrism consistently improves the task performance, especially as trajectories become longer, while reducing memory token consumption. Furthermore, the learned view policy transfers across different VLMs without additional adaptation, demonstrating the effectiveness of task-conditioned relational views as a general memory interface for agents.

cs.AI↗

arXiv2026

Modeling Cell-Cycle-Aware Single-Cell Drug Perturbation Responses

Single-cell drug perturbation models should capture transcriptional response magnitude and whether a treatment changes the proliferative state of the cell. This is difficult because cell-cycle variation is often treated as a nuisance factor, and benchmark processing rarely makes drug-induced phase changes a primary prediction target. We introduce scCycleMol, a cell-cycle-aware perturbation prediction framework built on a curated 24-hour SciPlex3 benchmark with standardized molecule identities, dose and cell-line metadata, modeled genes, and expression-derived cell-cycle supervision. scCycleMol derives cell-cycle supervision from the treated state and applies it to predicted treated expression without using phase as an input covariate. The model includes a learnable full-expression cell-cycle head with circular G1/S/G2M targets, and we evaluate readout-only supervision (with stop-gradient) versus closed-loop supervision (backpropagating through decoder, dose-response module, and drug representation). We also compare molecular representations and pretraining sources to isolate the effect of the cell-cycle objective. On a processed 24-hour SciPlex3 benchmark (635,541 cells, 186 perturbations, 188 compound embeddings, 3 cell lines, 4 doses plus DMSO, 5,080 genes), the best LINCS-pretrained circular variant reaches 0.9093 mean all-gene R-squared and 0.6843 mean DE-gene R-squared. Under matched preprocessing, closed-loop cell-cycle supervision improves phase accuracy by 0.54-0.62 points while keeping mean all-gene R-squared within 0.003 of matched chemCPA no-cell-cycle models; Tahoe-pretrained readout-only circular supervision achieves the strongest phase accuracy at 0.9609.

q-bio.QM↗

arXiv2026

PolarMem: A Training-Free Polarized Latent Graph Memory for Verifiable Vision-Language Models

Memory is not merely a storage mechanism for intelligent systems, but a structure for organizing evidence and constraining belief. This is especially important for multimodal reasoning, where retrieved evidence must be both query-relevant and visually consistent. However, current memory systems for vision-language models (VLMs) remain largely positive-associative: they retrieve what is similar or previously observed, but lack an explicit way to remember what has been verified as absent or logically excluded. To this end, we propose \textbf{PolarMem}, a training-free polarized latent graph memory framework for verifiable vision-language reasoning. PolarMem transforms frozen VLM perceptual signals into \textit{HAS}, \textit{NOT\_HAS}, and \textit{Uncertain} memory states through semantic consistency verification and adaptive distributional partitioning, and stores them in a polarized graph with distinct positive and negative memory relations. During inference, a lexicographical logic-aware retrieval protocol enforces logical consistency before semantic similarity, suppressing conflicting memories before they enter the model context. Across eight frozen VLM backbones and six multimodal benchmarks, PolarMem consistently improves retrieval-intensive tasks and reduces retrieval-level contradictions. These results highlight negative memory as a key mechanism for building more reliable multimodal memory systems. Our code is available at https://github.com/czs-ict/PolarMem.

cs.AI↗

arXiv2026

MolClaw: An Autonomous Agent with Hierarchical Skills for Drug Molecule Evaluation, Screening, and Optimization

Computational drug discovery, particularly the complex workflows of drug molecule screening and optimization, requires orchestrating dozens of specialized tools in multi-step workflows, yet current AI agents struggle to maintain robust performance and consistently underperform in these high-complexity scenarios. Here we present MolClaw, an autonomous agent that leads drug molecule evaluation, screening, and optimization. It unifies over 30 specialized domain resources through a three-tier hierarchical skill architecture (70 skills in total) that facilitates agent long-term interaction at runtime: tool-level skills standardize atomic operations, workflow-level skills compose them into validated pipelines with quality check and reflection, and a discipline-level skill supplies scientific principles governing planning and verification across all scenarios in the field. Additionally, we introduce MolBench, a benchmark comprising molecular screening, optimization, and end-to-end discovery challenges spanning 8 to 50+ sequential tool calls. MolClaw achieves state-of-the-art performance across all metrics, and ablation studies confirm that gains concentrate on tasks that demand structured workflows while vanishing on those solvable with ad hoc scripting, establishing workflow orchestration competence as the primary capability bottleneck for AI-driven drug discovery.

cs.AI↗

arXiv2026

MineEvolve: Self-Evolution with Accumulated Knowledge for Long-Horizon Embodied Minecraft Agents

Long-horizon embodied intelligence requires agents to improve through interaction, not merely to execute plans generated from static goals. A central challenge is therefore to transform past executions into knowledge that can shape future decisions. Minecraft provides a representative testbed for this problem, where tasks such as crafting tools, building redstone components, and obtaining diamond equipment involve long prerequisite chains and are frequently disrupted by missing tools, blocked paths, GUI failures, or stagnant execution. To this end, we propose \textbf{MineEvolve}, a knowledge-driven self-evolution framework that converts execution feedback into actionable behavioral knowledge. MineEvolve first uses \underline{\emph{\textbf{\ding{182}Monitor}}} to convert each subgoal execution into typed feedback, including state changes, inventory changes, failure types, progress signals, and stagnation indicators. \underline{\emph{\textbf{\ding{183}Inducer}}} then derives reusable skills from successful executions and remedies from failed or stagnant executions. \underline{\emph{\textbf{\ding{184}Curator}}} validates, merges, filters, and retrieves these knowledge entries, while \underline{\emph{\textbf{\ding{185}Adaptor}}} uses them to repair the unfinished part of the plan under repeated failures or stagnation. Experiments on the Minecraft MCU long-horizon task suite show that MineEvolve consistently improves performance across multiple language-model planners, with larger gains on high-dependency task groups. Ablation and knowledge-accumulation studies further demonstrate that converting execution signals into structured behavioral knowledge is an effective path toward self-evolving embodied agents in long-horizon environments. Our code is available at https://github.com/xzw-ustc/MC-MineEvolve.

cs.AI↗

arXiv2021

Testing Boolean Functions Properties

The goal in the area of functions property testing is to determine whether a given black-box Boolean function has a particular given property or is $\varepsilon$-far from having that property. We investigate here several types of properties testing for Boolean functions (identity, correlations and balancedness) using the Deutsch-Jozsa algorithm (for the Deutsch-Jozsa (D-J) problem) and also the amplitude amplification technique. At first, we study here a particular testing problem: namely whether a given Boolean function $f$, of $n$ variables, is identical with a given function $g$ or is $\varepsilon$-far from $g$, where $\varepsilon$ is the parameter. We present a one-sided error quantum algorithm to deal with this problem that has the query complexity $O(\frac{1}{\sqrt{\varepsilon}})$. Moreover, we show that our quantum algorithm is optimal. Afterwards we show that the classical randomized query complexity of this problem is $Θ(\frac{1}{\varepsilon})$. Secondly, we consider the D-J problem from the perspective of functional correlations and let $C(f,g)$ denote the correlation of $f$ and $g$. We propose an exact quantum algorithm for making distinction between $|C(f,g)|=\varepsilon$ and $|C(f,g)|=1$ using six queries, while the classical deterministic query complexity for this problem is $Θ(2^{n})$ queries. Finally, we propose a one-sided error quantum query algorithm for testing whether one Boolean function is balanced versus $\varepsilon$-far balanced using $O(\frac{1}{\varepsilon})$ queries. We also prove here that our quantum algorithm for balancedness testing is optimal. At the same time, for this balancedness testing problem we present a classical randomized algorithm with query complexity of $O(1/\varepsilon^{2})$. Also this randomized algorithm is optimal. Besides, we link the problems considered here together and generalize them to the general case.

quant-ph↗

arXiv2020

Prediction and optimization of NaV1.7 inhibitors based on machine learning methods

We used machine learning methods to predict NaV1.7 inhibitors and found the model RF-CDK that performed best on the imbalanced dataset. Using the RF-CDK model for screening drugs, we got effective compounds K1. We use the cell patch clamp method to verify K1. However, because the model evaluation method in this article is not comprehensive enough, there is still a lot of research work to be performed, such as comparison with other existing methods. The target protein has multiple active sites and requires our further research. We need more detailed models to consider this biological process and compare it with the current results, which is an error in this article. So we want to withdraw this article.

q-bio.QM↗

arXiv2019

Identification of key genes related to the mechanism and prognosis of lung squamous cell carcinoma using bioinformatics analysis

Objectives Lung squamous cell carcinoma (LUSC) often diagnosed as advanced with poor prognosis. The mechanisms of its pathogenesis and prognosis require urgent elucidation. This study was performed to screen potential biomarkers related to the occurrence, development and prognosis of LUSC to reveal unknown physiological and pathological processes. Materials and Methods Using bioinformatics analysis, the lung squamous cell carcinoma microarray datasets from the GEO and TCGA databases were analyzed to identify differentially expressed genes(DEGs). Furthermore, PPI and WGCNA network analysis were integrated to identify the key genes closely related to the process of LUSC development. In addition, survival analysis was performed to achieve a prognostic model that accomplished a high level of prediction accuracy. Results and Conclusion Eighty-five up-regulated and 39 down-regulated genes were identified, on which functional and pathway enrichment analysis was conducted. GO analysis demonstrated that up-regulated genes were principally enriched in epidermal development and DNA unwinding in DNA replication. Down-regulated genes were mainly involved in cell adhesion, signal transduction and positive regulation of inflammatory response. After PPI and WGCNA network analysis, eight genes, including AURKA, RAD51, TTK, AURKB, CCNA2, TPX2, KPNA2 and KIF23, have been found to play a vital role in LUSC development. The prognostic model contained 20 genes, 18 of which were detrimental to prognosis. The AUC of the established prognostic model for predicting the survival of patients at 1, 3, and 5 years was 0.828, 0.826 and 0.824, respectively. To conclude, this study identified a number of biomarkers of significant interest for additional investigation of the therapies and methods of prognosis of lung squamous cell carcinoma.

q-bio.GN↗

arXiv2017

Nesting-induced Local Electronic Patterns around a Single As (Te, Se) Vacancy in Iron-based Superconductors

The local electronic states around a single As (Te, Se) vacancy are investigated in order to shed light on the role of ligands in a series of iron-based superconductors. Such a vacancy can produce a local hopping correction ranging from $-0.22$ eV to 0.12 eV and always induce two in-gap resonance peaks in the local density of states (LDOS) at the fixed symmetrical bias voltages, which are rather robust and irrelevant to the phase of superconducting order parameter. The LDOS images near the defect predominantly possess $0^o$ and $45^o$ stripes. These energy-dependent charge modulations created by quasiparticle interference are originated in the nesting effect between the inner (outer) hole Fermi surface around $Γ$ point and the inner (outer) electron Fermi surface around $M$ point.

cond-mat.supr-con↗

arXiv2016

Predict genome-scale fluxes based solely on enzyme abundance by a novel Hyper-Cube Shrink Algorithm

One of the long-expected goals of genome-scale metabolic modeling is to evaluate the influence of the perturbed enzymes to the flux distribution. Both ordinary differential equation (ODE) models and the constraint-based models, like Flux balance analysis (FBA), lack of the room of performing metabolic control analysis (MCA) for large-scale networks. In this study, we developed a Hyper-Cube Shrink Algorithm (HCSA) to incorporate the enzymatic properties to the FBA model by introducing a pseudo reaction constrained by enzymatic parameters. Our algorithm was able to handle not only prediction of knockout strains but also strains with quantitative adjustment of expression level or activity. We first demonstrate the concept by applying HCSA to a simplest three-node network. Then we validate its prediction by comparing with ODE and with a synthetic network in Saccharomyces cerevisiae producing voilacein and analogues. Finally we show its capability of predicting the flux distribution in genome-scale networks by applying it to the sporulation in yeast.

q-bio.MN↗

arXiv2005

Monte Carlo simulation of melting transition on DNA nanocompartment

DNA nanocompartment is a typical DNA-based machine whose function is dependent of molecular collective effect. Fundamental properties of the device have been addressed via electrochemical analysis, fluorescent microscopy, and atomic force microscopy. Interesting and novel phenomena emerged during the switching of the device. We have found that DNAs in this system exhibit a much steep melting transition compared to ones in bulk solution or conventional DNA array. To achieve an understanding to this discrepancy, we introduced DNA-DNA interaction potential to the conventional Ising-like Zimm-Bragg theory and Peyrard-Bishop model of DNA melting. To avoid unrealistic numerical calculation caused by modification of the Peyrard-Bishop nonlinear Hamiltonian with the DNA-DNA interaction, we established coarse-gained Monte Carlo recursion relations by elucidation of five components of energy change during melting transition. The result suggests that DNA-DNA interaction potential accounts for the observed steep transition.

physics.bio-ph↗