Learning Framework Research Articles

Abstract IA021: Plasma-only ultrasensitive liquid biopsy to monitor tumor burden and clonal evolution

Abstract Individual bulk tumor biopsies are invasive and may fail to capture clonal heterogeneity within primary tumors and metastatic sites of disease. Further, analyses of subclonal variants are often hindered by background error rates related to tissue preservation methods (e.g. FFPE) and sequencing artifact. Use of plasma-only ultrasensitive liquid biopsy approaches may allow for noninvasive inference of clonality and dynamically track changes in tumor burden. We previously developed MRD-EDGE, a tumor-informed machine learning classifier for ultrasensitive liquid biopsy monitoring of single nucleotide variants (SNVs) and copy number variants (CNVs) using standard plasma whole genome sequencing. For SNVs, MRD- EDGE uses a novel machine learning framework to classify individual cell free DNA (cfDNA) fragments as circulating tumor DNA or cfDNA sequencing artifact. We demonstrated the potential of plasma-only MRD-EDGE to identify SNVs in advanced melanoma and monitor tumor burden throughout treatment with immune checkpoint inhibiton. In tumor-informed settings, MRD-EDGE was used to identify minimal residual disease at tumor fractions (TFs) of 1:100,000 and below. Accurate SNV classification is a function of classifier performance and background error rate. To reduce error rates, we combined MRD-EDGE SNV with paired plus- minus sequencing (ppmSeq) from Ultima Genomics. ppmSeq is a PCR-free approach that leverages the hydrogen bonds of complementary DNA to carry matched native strands directly into sequencing, enabling scalable duplex sequencing. Use of MRD-EDGE SNV together with ppmSeq enables ultrasensitive tracking of early-stage non-small cell lung cancer throughout neoadjuvant immunotherapy. Plasma-only MRD-EDGE is therefore poised to enable dynamic monitoring of tumor burden and clonal composition outside the scope of tumor-informed approaches, including the comparison of SNV composition in cancer recurrence with preoperative disease. Citation Format: Adam Widman, Alexandre P. Cheng, John Zinno, Srinivas Rajagopalan, Ashish Saxena, Nasser Altorki, Dan A. Landau. Plasma-only ultrasensitive liquid biopsy to monitor tumor burden and clonal evolution [abstract]. In: Proceedings of the AACR Special Conference: Liquid Biopsy: From Discovery to Clinical Implementation; 2024 Nov 13-16; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2024;30(21_Suppl):Abstract nr IA021.

Abstract PR011: Advance prostate cancer detection through epigenomic profiling of cell-free DNA

Abstract Introduction: Metastatic castration-resistant prostate cancer (mCRPC) is a heterogeneous disease which can be classified into clinically relevant subtypes based on the expression of genes, such as the androgen receptor (AR) and neuroendocrine markers. Neuroendocrine prostate cancer (NEPC), characterized by gain of stem-like and neuroendocrine features and lack of AR expression is a clinically aggressive variant. Due to the lack of adequate biomarkers, NEPC is usually detected at a very advanced stage. There is mounting evidence that molecular subtype changes seen in NEPC are enforced by widespread epigenetic alterations, in particular DNA methylation changes. In this study, we aim to devise a novel DNA methylation-based assay for molecular subtyping and disease monitoring from cell-free DNA (cfDNA). Methods: We analyzed genome wide methylation patterns in 56 prostate cancer patient-derived xenograft (PDX) and 128 mCRPC tumors using array- and sequencing-based assays. We integrated DNA methylation at promoters, gene bodies and transcription factor binding site (TFBS) to determine the landscape of methylation alterations at key lineage specific genes. Using whole genome methylation derived from tissue with matched expression data we developed a deep learning framework to predict gene expression directly from tissue or cfDNA. Using key marker genes, the model was used to discern tumor molecular phenotypes from tissue and cfDNA in three independent cohorts of mCRPC patients using whole genome bisulfite sequencing and low-pass Enzymatic Methyl-Seq (EM-seq). Results: We observed a tight association between promoter, gene body and TFBS methylation with gene expression. Inferring gene expression from methylation for lineage specific markers such as AR, KLK3, ASCL1, INSM1, SRRM4 and DLL3 we classified molecular subtypes from both tissue and cfDNA. Additionally, for AR and ASCL1, we identified core sets of TFBSs whose differential methylation allowed for accurate assay-independent molecular subtype quantification. Applying the optimized quantitative model to mCRPC patients who underwent comprehensive tissue sampling by rapid autopsy we observed accurate subtype classification from both tissue samples and cfDNA for all cases. A similar analytical performance was observed in additional clinical mCRPC cohorts with cfDNA. Conclusion: Whole-genome methylation analysis of cfDNA allows for the prediction of gene expression patterns in tumor tissues, enabling non-invasive tumor subclassification and assessment of therapeutic targets. Citation Format: Mohamed Adil, Brian Hanratty, Pallabi Mustafi, Chitvan Mittal, Helen Richards, Ilsa Coleman, Radhika Patel, Anna-Lisa Doebley, Robert Patton, Eden Cruikshank, Patricia Galipeau, Ruth Dumpit, Martine Roudier, Jin-Yih Low, Navonil Sarkar, Robert Montgomery, Eva Corey, Colm Morrissey, Peter Nelson, Gavin Ha, Michael Haffner. Advance prostate cancer detection through epigenomic profiling of cell-free DNA [abstract]. In: Proceedings of the AACR Special Conference: Liquid Biopsy: From Discovery to Clinical Implementation; 2024 Nov 13-16; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2024;30(21_Suppl):Abstract nr PR011.

Target Structure Learning Framework for Unsupervised Multi-Class Domain Adaptation

Unsupervised multi-class domain adaptation (multi-class UDA) has recently been proposed to fill the gap between empirically practical methods for multi-class classification and well-founded theory with the setting of binary classification. Nevertheless, the multi-class UDA methods use model predictions to characterize the disagreement of multi-class scoring hypotheses, which is used to optimize the divergence between domain distributions. Such self-training manner may bring inaccurate model predictions, which would damage the target structure due to the absence of labels of the target domain, leading to sub-optimal performance. On the other hand, this disagreement between multi-class scoring hypotheses does not involve the relationships among all of the multiple classes. It causes that multi-class UDA cannot properly connect the advanced practical UDA methods that consider class-conditional distribution alignment. Thus, we propose to exploit the target structure information and then incorporate it into multi-class UDA to achieve class-conditional distribution alignment. We theoretically and experimentally explain the importance of accurate target structure information to reduce the expected error on the target domain. Notably, our method achieves state-of-the-art results on three commonly-used benchmarks with different scales. In addition, using the target structure information, we propose a variant to cope with noisy open-world source domains such as noisy labels and out-of-distribution samples, enhancing the robustness of our method. The source code is available at https://github.com/jingzhengli/Multi_Class_UDA .

Societal Impact of Machine Learning on Human Emotions

Modern networking conversations generate annotated metadata, necessitating a method for synthesizing insights from statistics. Emotion detection is crucial for practical conversations, distinguishing joy, grief, and wrath. Corpora are becoming the standard for human–machine interaction, aiming to make interactions feel natural and real. A paradigm that identifies debates and customer views can provide a human touch to these interactions. Researchers developed a machine learning framework for assessing emotions in English phrases, utilizing Long Short Term Memory perspective and real-time emotion recognition in idiomatic speech. Emotion recognition rule is created using ontologies like Word Net and Concept Net, Naive Bayes, and Random Forest. Real-time analysis of written words and facial expressions significantly outperforms current algorithms and commandment classifiers in identifying emotional states.

Research on Key Technologies of Image Steganography Based on Simultaneous Deception of Vision and Deep Learning Models

As machine learning continues to evolve, traditional image steganography techniques find themselves increasingly unable to meet the dual challenge of deceiving both the human visual system and machine learning models. In response, this paper introduces the Visually Robust Image Steganography (VRIS) model, specifically tailored for this dual deception task. The VRIS model elevates visual security through the use of specialized feature extraction and processing methodologies. It meticulously conducts feature-level fusion between secret images and cover images, ensuring a high level of visual similarity between them. To effectively mislead machine learning models, the VRIS model incorporates a sophisticated strategy utilizing random noise factors and discriminators. This involves adding controlled amounts of random Gaussian noise to the encrypted image, thereby enhancing the difficulty for machine learning frameworks to recognize it. Furthermore, the discriminator is trained to discern between the noise-infused encrypted image and the original cover image. Through adversarial training, the discriminator and VRIS model refine each other, successfully deceiving the machine learning systems. Additionally, the VRIS model presents an innovative method for extracting and reconstructing secret images. This approach safeguards secret information from unauthorized access while enabling legitimate users to non-destructively extract and reconstruct images by leveraging multi-scale features from the encrypted image, combined with advanced feature fusion and reconstruction techniques. Experimental results validate the effectiveness of the VRIS model, achieving high PSNR and SSIM scores on the LFW dataset and demonstrating significant deception capabilities against the ResNet50 model on the Mini-ImageNet dataset, with an impressive misclassification rate of 99.24%.

Alternative Learning Paradigms for Image Quality Transfer

Image Quality Transfer (IQT) aims to enhance the contrast and resolution of low-quality medical images, e.g. obtained from low-power devices, with rich information learned from higher quality images. In contrast to existing IQT methods in the literature which adopt supervised learning frameworks, in this work, we propose two novel formulations of the IQT problem. The first approach uses an unsupervised learning framework, whereas the second is a combination of both supervised and unsupervised learning. The unsupervised learning approach considers a sparse representation (SRep) and dictionary learning model, which we call IQT-SRep, whereas the combination of supervised and unsupervised learning ap- proach is based on deep dictionary learning (DDL), which we call IQT-DDL. The IQT-SRep approach trains two dictionaries using a sparse representation model using pairs of low- and high-quality volumes. Subsequently, the sparse representation of a low-quality block, in terms of the low-quality dictionary, can be directly used to recover the corresponding high-quality block using the high-quality dictionary. On the other hand, the IQT-DDL ap- proach explicitly learns a high-resolution dictionary to upscale the input volume, while the entire network, including high dictionary generator, is simultaneously optimised to take full advantage of deep learning methods. The two models are evaluated using a low-field mag- netic resonance imaging (MRI) application aiming to recover high-quality images akin to those obtained from high-field scanners. Experiments comparing the proposed approaches against state-of-the-art supervised deep learning IQT method (IQT-DL) identify that the two novel formulations of the IQT problem can avoid bias associated with supervised meth- ods when tested using out-of-distribution data that differs from the distribution of the data the model was trained on. This highlights the potential benefit of these novel paradigms for IQT.

Weather-aware energy management for unmannedaerial vehicles: a machine learning application with global data integration

This study introduces a machine learning (ML) framework to predict unmanned aerial vehicle (UAV) energy requirements under diverse environmental conditions. The framework correlates UAV flight patterns with publicly accessible weather data, to yield an energy management tool applicable to a wide range of UAV configurations. The model employs the Cross-industry standard process for data mining and advanced feature engineering, offering an in-depth analysis of meteorological factors and UAV energy demands. The study assesses several multi-regression linear and ML models, whereby ensemble models gradient boosting (GB) and eXtreme gradient boosting demonstrate superior performance and accuracy. Specifically, the GB model achieved a test mean absolute error (MAE) of 0.0395 V (V) for voltage, 0.808 A (A) for current, and 9.758 mA-hours (mAh) for discharge, with prediction accuracy of over 99.9% for voltage and discharge, and 97% for current, derived from the coefficient of determination (R2). A novel integration of real-world UAV logs and weather data underpins the development of a weather-aware ML prediction model for UAV energy consumption. Our framework is capable of concurrently predicting three components of energy and power with almost uniform accuracy, a feature not found in contemporary models. Empirical test flights show a discrepancy of only 0.005 W-hour (Wh) between total predicted and actual energy consumption. This work enhances both efficiency and safety in UAV operations. The resulting energy-predictive flight planning tool sets a new benchmark for artificial intelligence (AI) applications in intelligent automation for UAVs.

PSATF-6mA: an integrated learning fusion feature-encoded DNA-6 mA methylcytosine modification site recognition model based on attentional mechanisms

DNA methylation is of crucial importance for biological genetic expression, such as biological cell differentiation and cellular tumours. The identification of DNA-6mA sites using traditional biological experimental methods requires more cumbersome steps and a large amount of time. The advent of neural network technology has facilitated the identification of 6 mA sites on cross-species DNA with enhanced efficacy. Nevertheless, the majority of contemporary neural network models for identifying 6 mA sites prioritize the design of the identification model, with comparatively limited research conducted on the statistically significant DNA sequence itself. Consequently, this paper will focus on the statistical strategy of DNA double-stranded features, utilising the multi-head self-attention mechanism in neural networks applied to DNA position probabilistic relationships. Furthermore, a new recognition model, PSATF-6 mA, will be constructed by continually adjusting the attentional tendency of feature fusion through an integrated learning framework. The experimental results, obtained through cross-validation with cross-species data, demonstrate that the PSATF-6 mA model outperforms the baseline model. The in-Matthews correlation coefficient (MCC) for the cross-species dataset of rice and m. musus genomes can reach a score of 0.982. The present model is expected to assist biologists in more accurately identifying 6 mA locus and in formulating new testable biological hypotheses.

Adversarial imitation learning with deep attention network for swarm systems

AbstractSwarm systems consist of a large number of interacting individuals, which exhibit complex behavior despite having simple interaction rules. However, crafting individual motion policies that can manifest desired collective behaviors poses a significant challenge due to the intricate relationship between individual policies and swarm dynamics. This paper addresses this issue by proposing an imitation learning method, which derives individual policies from collective behavior data. The approach leverages an adversarial imitation learning framework, with a deep attention network serving as the individual policy network. Our method successfully imitates three distinct collective behaviors. Utilizing the ease of analysis provided by the deep attention network, we have verified that the individual policies underlying a certain collective behavior are not unique. Additionally, we have analyzed the different individual policies discovered. Lastly, we validate the applicability of the proposed method in designing policies for swarm robots through practical implementation on swarm robots.

Abstract 4118735: Interpretable deep learning translation of GWAS findings for drug repurposing in Atrial Fibrillation

Introduction: Translating human genetic findings, such as genome-wide association studies (GWAS) to pathobiology and the discovery of therapeutic target remains a major challenge for Atrial Fibrillation (AF). We previously published a network topology-based deep learning framework to identify disease-associated genes (NETTAG). Hypothesis: By using a deep learning framework, we can efficiently identify AF risk genes, druggable targets, and candidates of repurposable drugs. Aims: To identify potential AF related genes and repurposable drug candidates using our deep learning framework. Methods: First, we collected the reported quantitative trait loci (QTLs) for AF from human heart tissues. Then, we identified the overlaps between the QTLs and the previously reported 150 AF GWAS loci in the latest meta-analysis. We previously built a comprehensive human protein-protein interactome using 18 publicly available databases, containing 351,444 unique PPIs and 17,706 proteins. Using the human protein-protein interactome and the overlaps between AF GWAS hits and QTLs, we prioritized genes and defined the genes with the top 1 % predicted score as AF risk genes (afRGs) using the NETTAG. Then, we assembled drugs from the Drugbank database relating 2,938 FDA-approved drugs or clinically investigated molecules. Using network proximity approaches to evaluate the closest distance between afRGs and a drug’s targets within the human protein-protein interactome, we computationally predicted drugs for AF using Z scores <-2.0. Results: We first collected the overlaps between AF GWAS hits and QTLs, which constituted 27 expression and 12 splicing QTLs. Via NETTAG, we identified 176 afRGs. Among the 176 predicted afRGs, 12 proteins (gene products of afRGs) have been identified as known drug targets with FDA-approved medicines. In total, 1,275 targets have been widely investigated as therapeutic targets for treating AF. Using the closest-based network proximity approach, we computationally identified 49 candidate drugs. These included drugs both reportedly potentially treating AF, such as Pioglitazone (Z=2.29), Telmisartan (Z=-2.52), Sildenafil (Z=-2.86), and those have never been reported before, such as Balsalazide (Z=-2.32). Conclusion: Using a deep learning methodology that utilized GWAS and QTL findings, we identified risk genes and repurposing drug candidates for AF.

Abstract 4116410: A Novel Deep Learning Approach for Prediction of Right Heart Failure After Left Ventricular Assist Device Implantation using Pulmonary Artery Pressure Tracings

Introduction: Left ventricular assist device (LVAD) offers a lifeline for advanced heart failure patients, but ~30% experience post-operative right heart failure (RHF) with significant morbidity and mortality. Current methods for predicting RHF risk have limited accuracy. This study is the first of its kind to explore the feasibility of using a convolutional neural network (CNN) deep learning model with pulmonary artery pressure (PAP) tracings acquired during pre-operative hemodynamic assessment to identify patients at risk for early RHF. Methods: A CNN model was pre-trained on PAP tracings from the MIMIC-III Waveform Database to leverage transferable knowledge and improve model performance. The model learned general features and temporal relationships in hemodynamic tracing morphology for RHF prediction. It was then adapted for classification by modifying the output layers. Post-LVAD RHF was definitively adjudicated for each instance included from our 246 patient single center LVAD database. RHF was adjudicated using the 2020 Academic Research Consortium defintion. The database included 205 non-RHF instances and 41 RHF instances. We developed a novel signal processing method utilizing computer vision to extract PAP waveform data from pre-operative right heart catheterization reports. Data was reviewed to ensure accurate pressure tracing representation. SMOTE-ENN (synthetic oversampling) and class weighting addressed dataset imbalance during training to enhance model generalizability and RHF prediction accuracy. Results: The model achieved strong performance in classifying RHF from non-RHF outcomes, as evidenced by consistent AUC scores of 0.87 and 0.85 on validation and unseen test data, respectively. These findings indicate the potential for hemodynamic tracings to predict RHF after LVAD implantation. Notably, the model maintained its performance on unseen data, highlighting its generalizability. Conclusions: This study explored the feasibility of using a deep learning model for RHF risk stratification in patients with LVADs. The model, pre-trained on PAP tracings, achieved promising classification performance assessed by AUC. Further validation with larger, more diverse datasets will refine model performance and enhance generalizability for clinical application. Leveraging hemodynamic tracings within a deep learning framework holds promise for improved clinical outcome prediction.

Enhancing image-based diagnosis of gastrointestinal tract diseases through deep learning with EfficientNet and advanced data augmentation techniques.

The early detection and diagnosis of gastrointestinal tract diseases, such as ulcerative colitis, polyps, and esophagitis, are crucial for timely treatment. Traditional imaging techniques often rely on manual interpretation, which is subject to variability and may lack precision. Current methodologies leverage conventional deep learning models that, while effective to an extent, often suffer from overfitting and generalization issues on medical image datasets due to the intricate and subtle variations in disease manifestations. These models typically do not fully utilize the potential of transfer learning or advanced data augmentation, leading to less-than-optimal performance, especially in diverse real-world scenarios where data variability is high. This study introduces a robust model using the EfficientNetB5 architecture combined with a sophisticated data augmentation strategy. The model is tailored for the high variability and intricate details present in gastrointestinal tract disease images. By integrating transfer learning with maximal pooling and extensive regularization, the model aims to enhance diagnostic accuracy and reduce overfitting. The proposed model achieved a test accuracy of 98.89%, surpassing traditional methods by incorporating advanced regularization and augmentation techniques. The application of horizontal flipping and dynamic scaling during training significantly improved the model's ability to generalize, evidenced by a low-test loss of 0.230 and high precision metrics across all classes. The proposed deep learning framework demonstrates superior performance in the automated classification of gastrointestinal diseases from image data. By addressing key limitations of existing models through innovative techniques, this study contributes to the enhancement of diagnostic processes in medical imaging, potentially leading to more accurate and timely disease interventions.

Abstract 4136501: Machine Learning Uncovers the Contribution of Rare Genetic Variants and Enhances Risk Prediction for Coronary Artery Disease in the Japanese Population

Background: Genome-wide association studies (GWASs) have enhanced our understanding of the genetic basis of coronary artery disease (CAD), and polygenic risk scores (PRSs) have facilitated the assessment of genetic risk. However, these methods predominantly focus on common variants due to statistical power, potentially leaving rare variants insufficiently analyzed and thus limiting the predictive performance of PRS. Methods: We conducted whole genome sequencing (WGS) of 1,752 Japanese early-onset myocardial infarction (MI) patients and 3,019 controls from Biobank Japan (BBJ). We performed case-control association studies including GWAS and gene-based tests, as well as a novel machine learning-based framework. In this framework, we developed a penalized regression model to predict the CAD status from genome-wide rare nonsynonymous variants. The model identified the minimal set of most distinguishing features (genes) and generated a rare variant-based risk score (RVS). The RVS was evaluated on an independent validation WGS cohort of 200 cases and 824 controls. We also derived a PRS based on CAD-GWAS (25,668 CAD cases vs 141,667 controls from BBJ) to compare the properties and performance between the RVS and PRS. Results: In the case-control studies, only two common variants in chromosome 12 were identified in GWAS, with no genes in gene-based analysis (SKAT-O), suggesting the challenges in rare variant analysis. On the other hand, our machine-learning framework identified 59 CAD-related genes, including LDLR , a causal gene of familial hypercholesteremia. Functional analyses revealed that various biological pathways, including lipid metabolism, immune system, and vessel development, are involved in CAD. For the genetic risk prediction, RVS significantly predicted CAD (area under the curve [AUC], 0.58; p=0.001, pseudo-R 2 , 0.051; p=7.92*10 -9 ). RVS was significantly associated with LDL cholesterol levels and coagulation function (Pearson’s r, 0.21; p=4.5*10 -5 and 0.10; p=0.03, respectively) and MI patients with high RVS (top 5%) showed higher cardiovascular mortality rate (p=0.03, log-rank test), highlighting the clinical importance of RVS. Finally, the combined risk score (CRS) of RVS and PRS significantly improved CAD prediction compared to PRS (AUC, 0.66 (CRS) vs 0.61 (PRS); p=0.007, pseudo-R 2 , 0.093 (CRS) vs 0.040 (PRS); p=0.0018, Figure ). Conclusions: Our machine learning framework successfully characterized rare variants and enhanced genetic risk prediction in CAD.

Towards harmonization of SO(3)-equivariance and expressiveness: a hybrid deep learning framework for electronic-structure Hamiltonian prediction

Abstract Deep learning for predicting the electronic-structure Hamiltonian of quantum systems necessitates satisfying the covariance laws, among which achieving SO(3)-equivariance without sacrificing the non-linear expressive capability of networks remains unsolved. To navigate the harmonization between SO(3)-equivariance and expressiveness, we propose HarmoSE, a deep learning method synergizing two distinct categories of neural mechanisms as a two-stage encoding and regression framework. The first stage corresponds to group theory-based neural mechanisms with inherent SO(3)-equivariant properties prior to the parameter learning process, while the second stage is characterized by a non-linear 3D graph Transformer network we propose, featuring high capability on non-linear expressiveness. Their combination lies in the point that, the first stage predicts baseline Hamiltonians with abundant SO(3)-equivariant features extracted, assisting the second stage in empirical learning of equivariance; and in turn, the second stage refines the first stage’s output as a fine-grained prediction of Hamiltonians using powerful non-linear neural mappings, compensating for the intrinsic weakness on non-linear expressiveness capability of mechanisms in the first stage. Our method enables precise, generalizable predictions while capturing SO(3)-equivariance under rotational transformations, and achieves state-of-the-art performance in Hamiltonian prediction tasks under multiple mean absolute error (MAE) metrics, such as the average MAE across all samples and matrix elements, the MAE for challenging samples, the MAE for different Hamiltonian blocks, and the MAE for the challenging blocks. It also demonstrates significant improvements in accuracy for downstream quantities, such as occupied orbital energy and the electronic wavefunction, as measured by MAE and cosine similarity, respectively.

Domain knowledge-guided machine learning framework for state of health estimation in Lithium-ion batteries.

Accurate estimation of battery state of health is crucial for effective electric vehicle battery management. Here, we propose five health indicators that can be extracted online from real-world electric vehicle operation and develop a machine learning-based method to estimate the battery state of health. The proposed indicators provide physical insights into the energy and power fade of the battery and enable accurate capacity estimation even with partially missing data. Moreover, they can be computed for portions of the charging profile and real-world driving discharging conditions, facilitating real-time battery degradation estimation. The indicators are computed using experimental data from five cells aged under electric vehicle conditions, and a linear regression model is used to estimate the state of health. The results show that models trained with power autocorrelation and energy-based features achieve capacity estimation with maximum absolute percentage error within 1.5% to 2.5%.

Machine-Learning-Assisted Screening of Nanocluster Electrocatalysts: Mapping and Reshaping the Activity Volcano for the Oxygen Reduction Reaction.

In computational heterogeneous catalysis, Sabatier's principle-based activity volcano plots provide an intuitive guide to catalyst design but impose a fundamental constraint on the maximum achievable catalytic performance. Recently, subnano clusters have emerged as an exciting platform, offering high noble metal utilization and superior performance for various reactions compared to extended surfaces, reflecting a complex structure-activity relationship in the non-scalable regime. However, understanding their non-monotonic catalytic activity, attributed to the large configurational space and their fluxional identity, poses a formidable challenge. Here, we present a machine learning (ML) framework that captures the non-monotonic trends in oxygen reduction reaction (ORR) activity at the subnanometer scale, attributed to their dynamic fluxional characteristics. We demonstrate a size-dependent shifting and reshaping of the ORR activity volcano, with Au replacing Pt at the peak. Leveraging only upon the non-ab initio geometric and electronic properties, our trained ML model accurately captures the site-specific adsorption energies of intermediates at the subnanometer regime. To account for the inconsistent trend in activity, we analyzed the correlation between electronic and geometric properties. Our findings reveal that the d-filling and coupling matrix of the neighboring metal atom significantly influences the intermediate adsorption on the local chemical environment compared to the d-band center. Following this analysis, we utilized ML to map the catalyst distribution in the activity volcano and identified the five best sub-nano electrocatalysts, demonstrating overpotential values lower than or comparable to the Pt(111) surface for the ORR. This study provides intuitive guidelines for the rational designing of highly efficient electrocatalysts for fuel cell applications while modifying the activity volcano plots for electrocatalysts at the subnanometer regime.

Data free knowledge distillation with feature synthesis and spatial consistency for image analysis.

Privacy and security concerns restrict access to original training datasets, posing significant challenges for model compression. Data-Free Knowledge Distillation (DFKD) emerges as a solution, aiming to transfer knowledge from teacher to student networks without accessing original data. Existing DFKD methods struggle to generate high-quality synthetic samples that capture the complexities of real-world data, leading to suboptimal knowledge transfer. Moreover, these approaches often fail to preserve the spatial attributes of the teacher network, resulting in shortcut learning and limited generalization.To address these issues, a novel DFKD strategy is proposed with three innovations: (1) an enhanced DCGAN generator with an attention module for synthesizing samples with improved micro-discriminative features; (2) a Multi-Scale Spatial Activation Region Consistency (MSARC) mechanism to accurately replicate the teacher's spatial attributes; and (3) an adversarial learning framework that creates a dynamic competitive environment between the generative and distillation phases. Rigorous evaluation of the method on several benchmark datasets, including CIFAR-10, CIFAR-100, Tiny-ImageNet, and medical imaging datasets such as PathMNIST, BloodMNIST, and PneumoniaMNIST, demonstrates superior performance compared to existing DFKD methods. Specifically, on CIFAR-100, the student network attains an accuracy of 77.85%, surpassing previous methods like CMI and SpaceshipNet. On BloodMNIST, the method achieves an accuracy of 80.50%, outperforming the next best method by over 5%.

Application of Nexus Learning framework for Personalized Medicine and Drug Development

Biological systems provide an enormous amount of knowledge on development and disease. The pharmaceutical industry is facing new possibilities and challenges with the rise of high-throughput methods to study disease and biology. The aim is to develop drugs based on acceptable therapeutic assumptions. Nexus Learning provides a variety of tools that help improve discovery and decision making using extensive, high-quality data and well-specified queries. Potential applications of NLF can be found across the whole drug development process. Clinical study target validation, prognostic biomarker discovery, and digital pathology data processing are a few examples. There has been a wide variety of applications in terms of technique and environment, with some methods producing reliable forecasts and insights. The primary problem with using NLF is that the findings provided by NL tend to be easy to understand or reproduce, which can limit their value. The collection of complete and methodical high-dimensional data remains an essential need across all domains. By addressing these challenges and increasing understanding about what is required to verify NLF methodologies, we can employ NLF to make data-driven decisions, which can speed up drug discovery and development and minimize failure rates.

Improving the Classification of Defect Levels in Main Electrical Equipment Through Advanced Prompting on a Large Language Model

The stable operation of the power system is closely related to the national economy and people’s livelihoods. Therefore, the timely detection, qualitative assessment, and handling of major equipment defects are crucial. The classification of defect levels in main electrical equipment is a fundamental task in this process, which is often manually completed, supplemented by knowledge bases or expert systems. However, this approach is time-consuming, labor-intensive, involves challenging human–machine interaction, and relies on expert experience. Conversational large language models, such as ChatGPT, ERNIE Bot, ChatGLM, have garnered widespread recognition in various domains. However, these models may have errors in the reasoning process, resulting in biased or even erroneous outputs, which is referred to as the “hallucinations”. The hallucinations’ problem of large language models poses challenges in specific fields. To mitigate the hallucinations in large language models, researchers often seek to incorporate domain-specific knowledge into these models through methods like fine-tuning or prompt learning. In order to enhance model performance while minimizing computational costs, this study adopts the prompt learning approach. Specifically, we propose a large language model prompt learning framework based on knowledge graphs, aiming to provide the large language model with reasoning support by leveraging specific information stored within the knowledge graph and receive explainable reasoning result. Experimental results demonstrate that our module achieved superior results on the power defect dataset compared to the non-prompt method.

Arts Integration in an Undergraduate General Education Course to Improve Engagement in Bioengineering

AbstractRapid advancements in bioengineering call for broad public literacy to help individuals better understand the changes these technologies bring to our lives. However, making bioengineering concepts accessible and relevant, especially to non-science majors, is often challenging. To address this challenge, we designed a general education course aimed at demystifying bioengineering through an inclusive, interdisciplinary approach. Grounded in STEAM (Science, Technology, Engineering, Arts, and Mathematics), sociocultural learning frameworks, and inquiry-based learning, the course integrates artistic perspectives with scientific instruction to encourage creativity and critical thinking. Students explore core bioengineering concepts, ethical considerations, and societal impacts through a combination of art, hands-on making, and collaborative learning. We observed high engagement and interdisciplinary connections in the classroom and the practical experiences shared here highlight the potential for such an approach to enhance scientific literacy and appreciation of bioengineering’s role in society.