Sequential Pattern Research Articles

GPSFun: geometry-aware protein sequence function predictions with language models.

Knowledge of protein function is essential for elucidating disease mechanisms and discovering new drug targets. However, there is a widening gap between the exponential growth of protein sequences and their limited function annotations. In our prior studies, we have developed a series of methods including GraphPPIS, GraphSite, LMetalSiteand SPROF-GO for protein function annotations at residue or protein level. To further enhance their applicability and performance, we now present GPSFun, a versatile web server for Geometry-aware Protein Sequence Function annotations, which equips our previous tools with language models and geometric deep learning. Specifically, GPSFun employs large language models to efficiently predict 3D conformations of the input protein sequences and extract informative sequence embeddings. Subsequently, geometric graph neural networks are utilized to capture the sequence and structure patterns in the protein graphs, facilitating various downstream predictions including protein-ligand binding sites, gene ontologies, subcellular locationsand protein solubility. Notably, GPSFun achieves superior performance to state-of-the-art methods across diverse tasks without requiring multiple sequence alignments or experimental protein structures. GPSFun is freely available to all users at https://bio-web1.nscc-gz.cn/app/GPSFun with user-friendly interfaces and rich visualizations.

A learning analytics‐based collaborative conversational agent to foster productive dialogue in inquiry learning

AbstractBackgroundSustaining productive student–student dialogue in online collaborative inquiry learning is challenging, and teacher support is limited when needed in multiple groups simultaneously. Collaborative conversational agents (CCAs) have been used in the past to support student dialogue. Yet, research is needed to reveal the characteristics and effectiveness of such agents.ObjectivesTo investigate the extent to which our analytics‐based Collaborative Learning Agent for Interactive Reasoning (Clair) can improve the productivity of student dialogue, we assessed both the levels at which students shared thoughts, listened to each other, deepened reasoning, and engaged with peer's reasoning, as well as their perceived productivity in terms of their learning community, accurate knowledge, and rigorous thinking.MethodIn two separate studies, 19 and 27 dyads of secondary school students from Brazil and the Netherlands, respectively, participated in digital inquiry‐based science lessons. The dyads were assigned to two conditions: with Clair present (treatment) or absent (control) in the chat. Sequential pattern mining of chat logs and the student's responses to a questionnaire were used to evaluate Clair's impact.ResultsAnalysis revealed that in both studies, Clair's presence resulted in dyads sharing their thoughts at a higher frequency compared to dyads that did not have Clair. Additionally, in the Netherlands' study, Clair's presence led to a higher frequency of students engaging with each other's reasoning. No differences were observed in students' perceived productivity.ConclusionThis work deepens our understanding of how CCAs impact student dialogue and illustrates the importance of a multidimensional perspective in analysing the role of CCAs in guiding student dialogue.

MvStHgL: Multi-view Hypergraph Learning with Spatial-temporal Periodic Interests for Next POI Recommendation

Providing potential next point-of-interest (POI) suggestions for users has become a prominent task in location-based social networks, which receives more and more attention from the industry and academia, and it remains challenging due to highly dynamic and personalized interactions in user movements. Currently, state-of-the-art works develop various graph- and sequential-based learning methods to model user-POI interactions and transition regularities. However, there are still two significant shortcomings in these works: (1) Ignoring personalized spatial- and temporal-aspect interactive characteristics capable of exhibiting periodic interests of users; (2) Insufficiently leveraging the sequential patterns of interactions for beyond-pairwise high-order collaborative signals among users’ sequences. To jointly address these challenges, we propose a novel multi-view hypergraph learning with spatial-temporal periodic interests for next POI recommendation (MvStHgL). In the local view, we attempt to learn the POI representation of each interaction via jointing periodic characteristics of spatial and temporal aspects. In the global view, we design a hypergraph by regarding interactive sequences as hyperedges to capture high-order collaborative signals across users, for further POI representations. More specifically, the output of POI representations in the local view is used for the initialized embedding, and the aggregation and propagation in the hypergraph are performed by a novel Node-to-Hypergraph-to-Node scheme. Furthermore, the captured POI embeddings are applied to achieve sequential dependency modeling for next POI prediction. Extensive experiments on three real-world datasets demonstrate that our proposed model outperforms the state-of-the-art models.

Fundamentals for predicting transcriptional regulations from DNA sequence patterns.

Cell-type-specific regulatory elements, cataloged through extensive experiments and bioinformatics in large-scale consortiums, have enabled enrichment analyses of genetic associations that primarily utilize positional information of the regulatory elements. These analyses have identified cell types and pathways genetically associated with human complex traits. However, our understanding of detailed allelic effects on these elements' activities and on-off states remains incomplete, hampering the interpretation of human genetic study results. This review introduces machine learning methods to learn sequence-dependent transcriptional regulation mechanisms from DNA sequences for predicting such allelic effects (not associations). We provide a concise history of machine-learning-based approaches, the requirements, and the key computational processes, focusing on primers in machine learning. Convolution and self-attention, pivotal in modern deep-learning models, are explained through geometrical interpretations using dot products. This facilitates understanding of the concept and why these have been used for machine learning for DNA sequences. These will inspire further research in this genetics and genomics field.

Using Permutation Tests to Identify Statistically Sound and Nonredundant Sequential Patterns in Educational Event Sequences

Frequent sequential pattern mining is a valuable technique for capturing the relative arrangement of learning events, but current algorithms often return excessive learning event patterns, many of which may be noise or redundant. These issues exacerbate researchers’ burden when interpreting the patterns to derive actionable insights into learning processes. This study proposed permutation tests for identifying sequential patterns whose occurrences are statistically significantly greater than the chance value and different from their superpatterns. Simulations demonstrated that the test for detecting sound patterns had a low false discovery rate and high power, while the test for detecting nonredundant patterns also showed a high accuracy. Empirical data analyses found that the patterns detected in training data were generalizable to test data.

Network-based visualisation of frequent sequences.

Frequent sequence pattern mining is an excellent tool to discover patterns in event chains. In complex systems, events from parallel processes are present, often without proper labelling. To identify the groups of events related to the subprocess, frequent sequential pattern mining can be applied. Since most algorithms provide too many frequent sequences that make it difficult to interpret the results, it is necessary to post-process the resulting frequent patterns. The available visualisation techniques do not allow easy access to multiple properties that support a faster and better understanding of the event scenarios. To answer this issue, our work proposes an intuitive and interactive solution to support this task, introducing three novel network-based sequence visualisation methods that can reduce the time of information processing from a cognitive perspective. The proposed visualisation methods offer a more information rich and easily understandable interpretation of sequential pattern mining results compared to the usual text-like outcome of pattern mining algorithms. The first uses the confidence values of the transitions to create a weighted network, while the second enriches the adjacency matrix based on the confidence values with similarities of the transitive nodes. The enriched matrix enables a similarity-based Multidimensional Scaling (MDS) projection of the sequences. The third method uses similarity measurement based on the overlap of the occurrences of the supporting events of the sequences. The applicability of the method is presented in an industrial alarm management problem and in the analysis of clickstreams of a website. The method was fully implemented in Python environment. The results show that the proposed methods are highly applicable for the interactive processing of frequent sequences, supporting the exploration of the inner mechanisms of complex systems.

On Semi-Vector Spaces and Semi-Algebras with Applications in Fuzzy Automata

In this paper, we expand the theory of semi-vector spaces and semi-algebras, both over the semi-field of nonnegative real numbers R0+. More precisely, we prove several new results concerning these theories. We introduce to the literature the concept of eigenvalues and eigenvectors of a semi-linear operator, describing how to compute them. The topological properties of semi-vector spaces, such as completeness and separability, are also investigated here. New families of semi-vector spaces derived from the semi-metric, semi-norm and semi-inner product, among others, are exhibited. Furthermore, we show several new results concerning semi-algebras. After this theoretical approach, we apply such a theory in fuzzy automata. More precisely, we describe the semi-algebra of A-fuzzy regular languages and we apply the theory of fuzzy automata for counting patterns in DNA sequences.

Dissecting Mismatch Negativity: Early and Late Subcomponents for Detecting Deviants in Local and Global Sequence Regularities.

Mismatch negativity (MMN) is commonly recognized as a neural signal of prediction error evoked by deviants from the expected patterns of sensory input. Studies show that MMN diminishes when sequence patterns become more predictable over a longer timescale. This implies that MMN is comprised of multiple subcomponents, each responding to different levels of temporal regularities. To probe the hypothesized subcomponents in MMN, we record human electroencephalography during an auditory local-global oddball paradigm where the tone-to-tone transition probability (local regularity) and the overall sequence probability (global regularity) are manipulated to control temporal predictabilities at two hierarchical levels. We find that the size of MMN is correlated with both probabilities and the spatiotemporal structure of MMN can be decomposed into two distinct subcomponents. Both subcomponents appear as negative waveforms, with one peaking early in the central-frontal area and the other late in a more frontal area. With a quantitative predictive coding model, we map the early and late subcomponents to the prediction errors that are tied to local and global regularities, respectively. Our study highlights the hierarchical complexity of MMN and offers an experimental and analytical platform for developing a multi-tiered neural marker applicable in clinical settings.Significance statement Our study provides new insights into the intricate architecture of mismatch negativity (MMN), a key neural indicator for deviant detection. Using a refined oddball paradigm with dual-level temporal controls, we identified two unique MMN subcomponents, each linked to prediction errors at different brain hierarchies. This work establishes a practical platform for a multi-tiered neural marker, offering clinical applications for assessing brain function across various hierarchies.

Integrated convolution and self-attention for improving peptide toxicity prediction.

Peptides are promising agents for the treatment of a variety of diseases due to their specificity and efficacy. However, the development of peptide-based drugs is often hindered by the potential toxicity of peptides, which poses a significant barrier to their clinical application. Traditional experimental methods for evaluating peptide toxicity are time-consuming and costly, making the development process inefficient. Therefore, there is an urgent need for computational tools specifically designed to predict peptide toxicity accurately and rapidly, facilitating the identification of safe peptide candidates for drug development. We provide here a novel computational approach, CAPTP, which leverages the power of convolutional and self-attention to enhance the prediction of peptide toxicity from amino acid sequences. CAPTP demonstrates outstanding performance, achieving a Matthews correlation coefficient of approximately 0.82 in both cross-validation settings and on independent test dataset. This performance surpasses that of existing state-of-the-art peptide toxicity predictors. Importantly, CAPTP maintains its robustness and generalizability even when dealing with data imbalances. Further analysis by CAPTP reveals that certain sequential patterns, particularly in the head and central regions of peptides, are crucial in determining their toxicity. This insight can significantly inform and guide the design of safer peptide drugs. The source code for CAPTP is freely available at https://github.com/jiaoshihu/CAPTP. Supplementary data are available at Bioinformatics online.

Abstract PO4-17-04: CDK 4/6 inhibitor switching patterns in Swedish patients with metastatic breast cancer: 5-year update from the SIRI study

Abstract Background: CDK 4/6 inhibitors (CDK 4/6i) combined with endocrine therapy is a well-established treatment option in patients with hormone receptor-positive (HR+), human epidermal growth factor receptor 2 (HER2)-negative metastatic breast cancer (MBC). However, there is limited evidence on switching patterns within this class of drugs. The Swedish Ibrance Registries Insights (SIRI) study investigated CDK 4/6i switching patterns in real-world setting using a nationwide cohort of MBC patients. Methods: This was a retrospective study utilizing population-based Swedish Health Data Registers. The overall cohort included all breast cancer patients ≥ 18 years with ≥ 1 dispensation of palbociclib from January 2017 – June 2022. This subgroup analysis focused on patients with ≥ 1 dispensation of either ribociclib or abemaciclib in addition to palbociclib. Minimum follow-up was 3 months. CDK 4/6-I sequencing patterns in total and over time as well as time from the first CDK 4/6i to the subsequent, was investigated. No information on the reason for switching was available. Results: Out of a total 2314 patients with ≥ 1 dispensation of palbociclib, 256 patients (11%) had ≥ 1 dispensation of either ribociclib or abemaciclib, either prior to (60% of cases) or following (40 % of cases) palbociclib treatment. The share of patients with dispensation of > 1 CDK 4/6i increased over the study period from 7% of the patients initiating treatment in 2017 to 15% in 2021. Of the total 2161 patients initiating CDK 4/6i treatment on palbociclib, about 5% were subsequently prescribed another CDK 4/6i. The median age at treatment initiation in patients with > 1 dispensed CDK 4/6-i was similar to the overall study cohort (67.1 vs 68.4 years). Half of the patients identified with > 1 CDK 4/6i were initiated on ribociclib and prescribed palbociclib subsequently. The second and third most common switches were palbociclib to abemaciclib (25%) and palbociclib to ribociclib (10%), respectively. Seven patients received all three CDK 4/6i. In terms of time to subsequent CDK 4/6i, 108 patients (42%) were prescribed a subsequent CDK 4/6i within three months of treatment initiation with ribociclib-palbociclib being the most common sequence in 74 patients whereas 17%, 8%, and 4% of switches happened 4-6, 7-9, and 10-12 months after treatment initiation, respectively. Seventy-two patients (28%) had a dispensation of a subsequent CDK 4/6i more than 12 months after initiation of the first CDK 4/6i, with palbociclib-abemaciclib being the most common switch. Conclusions: A relatively high proportion of Swedish patients treated with CDK4/6i is switched from one to another CDK 4/6i soon after treatment initiation, implying that this strategy might be applied mainly, but not solely, due to adverse events. The increased trend over time might be associated with the regulatory approval of new CDK4/6i over time. Future studies should focus on the potential impact of switching CDK 4/6i on treatment effectiveness and toxicity as these issues have yet not been clarified. Declaration of Interest: This study is sponsored by Pfizer. AV has reported receiving research funding from Roche. HL has reported receiving consultant/advisory fees from Lilly, Novartis, Daichii, Pfizer, MSD, Pierre Fabre, Astra-Zeneca and research funding from Roche. MJ and DN are employees of Pfizer. RL and ML are employees of Quantify Research and were paid consultants to Pfizer for this research. Table. Table. Citation Format: Henrik Lindman, Antonis Valachis, Rosa Lauppe, Mathias Lilja, Daniel Nyqvist, Maria Jakobsson. CDK 4/6 inhibitor switching patterns in Swedish patients with metastatic breast cancer: 5-year update from the SIRI study [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO4-17-04.

Insomnia severity among caregivers of children with type 1 diabetes: A mixed method study

AimThis research was conducted to assess the severity of insomnia experienced by caregivers of children with type 1 diabetes. MethodIn this study, a mixed-method descriptive sequential pattern design was employed, incorporating both qualitative and quantitative research methods. In the study, 152 caregivers were surveyed for quantitative data on survey form and insomnia severity. Additionally, 9 mothers underwent semi-structured interviews for qualitative insights. ResultsOf caregivers, 13.8% were identified to experience clinical insomnia. There were statistically significant differences in insomnia severity index scores in terms of experiencing sleep problems after the child's type 1 diabetes diagnosis, caregivers having adequate sleep duration at night, how they felt on waking in the morning, difficulty managing their child's diabetes the next day due to lack of sleep, difficulty getting to sleep at night, sleeping more than one hour during the day, lack of sleep due to caring and treatment, lack of sleep due to not meeting the desired targets for blood glucose values and receiving support for type 1 diabetes management due to lack of sleep (p < 0.05). Three themes were identified as a result of qualitative analysis: Sleep status of caregivers, effect of sleep problems on daily life and solving sleep problems. ConclusionsCaregivers experience insomnia during the process of type 1 diabetes management and this situation affects their daytime care duties. Practical implicationsResearch shows caregivers of children with type 1 diabetes often face insomnia, highlighting the need for solutions.

A federated learning-based zero trust intrusion detection system for Internet of Things

The exponential growth of Internet of Things (IoT) devices poses distinctive challenges to safeguarding the security and privacy of interconnected systems. As the frequency of cyberattacks continues to escalate, the development of an effective and scalable Intrusion Detection System (IDS) based on Federated Learning (FL) for IoT becomes increasingly complex. Existing methodologies struggle to balance spatial and temporal feature extraction, particularly when confronted with dynamic and evolving cyber threats. Additionally, the lack of diversity in datasets employed for FL-based IDS evaluations further hinders progress. Furthermore, a notable tradeoff between performance and scalability emerges, particularly as the number of edge devices in communication grows. Motivated by the aforementioned challenges, this article presents a horizontal FL model that combines Convolutional Neural Networks (CNN) and Bidirectional Long-Term Short Memory (BiLSTM) for effective intrusion detection. This hybrid approach aims to address the limitations of existing methods and enhance the effectiveness of intrusion detection in the context of FL for IoT. Specifically, CNN plays a pivotal role in spatial feature extraction, allowing the model to identify and comprehend local patterns indicative of potential intrusions, and the BiLSTM component contributes by capturing temporal dependencies and learning sequential patterns within the data. The proposed IDS adheres to a zero-trust model by keeping the data on local edge devices, sharing only the learned weights with the centralized FL server. In turn, the FL server aggregates updates from diverse sources to optimize the accuracy of the global learning model. The experimental results using CICIDS2017 and Edge-IIoTset prove the effectiveness of the proposed approach over centralized and federated deep learning-based IDS.

Chronology of core sediments from the south coastal Bohai Sea of eastern Asia and its implications for sedimentary history

The Shandong Peninsula along the coastal Bohai Sea in China is a vast alluvial plain formed by the Yellow River. It is an ideal place for investigating the evolution of landform in a river–sea interaction area where chronology is key. However, detailed optically stimulated luminescence (OSL) chronology studies of cores are still limited and have mainly concentrated on the Yellow River Delta, while they are scarce on the south coast of the Bohai Sea. In this study 16 OSL samples were collected from the upper 32 m of core DZK01 (192 m in depth) to establish a chronological framework using quartz OSL. Results indicated that the quartz OSL ages ranged from 217 ± 19 ka to 1.63 ± 0.14 ka, and we believe they were the minimum ages when De>ca. 200 Gy owing to OSL signal saturation. Our dating results reveal two sedimentary hiatuses with age gaps of 71–37 ka and 37–9.5 ka. Holocene sediments are located above a depth of 11.8 m with an onset age of approximately 9.5 ± 0.6 ka. Based on the sedimentary facies and dating results, no transgression sediment has been observed during marine isotopic stage (MIS) 3. This sedimentary sequence pattern is consistent with global sea-level change. We also summarized the extents of three transgressions around the Bohai Sea, and found very different patterns for different part of the coasts, e.g., in the north, the west, and the south. This requires further investigation.

Decoding Nifty50’s Price Enigma: Machine Learning Powered Approach

Abstract: This study uses machine learning and Long Short-Term Memory (LSTM) networks to decipher the price riddle of Nifty50, a well-known stock market index in India. Using Long Short-Term Memory (LSTM) recurrent neural networks, which are renowned for their capacity to recognise sequential patterns, we aim to forecast and comprehend the intricate dynamics influencing Nifty50's price fluctuations.

SecureNet: Network Intrusion Detection using Machine Learning and Deep Learning Techniques

Abstract: In the ever-evolving landscape of cybersecurity, the need for robust intrusion detection systems has become paramount. This paper introduces a cutting-edge intrusion detection algorithm designed to enhance network security through the integration of advanced machine learning and deep learning methodologies. The proposed algorithm capitalizes on the strengths of both paradigms to achieve a comprehensive and adaptive approach to identifying malicious activities within a network. This research focuses on enhancing network security through the development and evaluation of a novel intrusion detection system leveraging both deep learning and traditional machine learning approaches. Utilizing the NSL-KDD dataset, we employ the Long Short-Term Memory (LSTM) model, a superior version of Recurrent Neural Networks (RNNs), and the K-Nearest Neighbors (KNN) algorithm for binary and multi-class classification of network intrusion anomalies. The LSTM model excels in capturing temporal dependencies, enabling the detection of nuanced sequential patterns, while the KNN algorithm contributes to a comprehensive classification framework. Experimental results demonstrate the effectiveness of the hybrid methodology, showcasing improved accuracy, precision, and recall compared to traditional methods. This research underscores the potential of integrating deep learning and classical machine learning techniques to bolster the capabilities of intrusion detection systems in safeguarding against evolving cyber threats

Ranking on user–item heterogeneous graph for Ecommerce next basket recommendations

Recommender systems have become an indispensable engine for online applications, which can help users locate the items they need among numerous other candidate items. Despite the success of general models, sequential recommender systems based on sequential patterns have garnered attention, especially in e-commerce, because customers do not necessarily purchase the same items over time. Markov chains and neural networks are the most adopted models in previous studies, but few integrated user sequential order information and user numeric rating feedback. To simultaneously capture the time-aware demands and time-invariant interests of e-commerce customers, this study proposes a unique heterogeneous graph structure to model sequential buying order and rating feedback and develops a query-based recommendation approach to predict future purchase behavior. Experiments on four different datasets show that the approach demonstrates improved performance for different metrics.

Predictability‐Aware Subsequence Modeling for Sequential Recommendation

Sequential recommendation frames the recommendation task as a next‐item prediction problem, where the model is trained to predict the next item given a user behavior sequence. While recent research has made significant progress in developing advanced models for this task, there exists a notable gap in the exploration of subsequences and the predictability inherent in user behavior sequences. This oversight can lead models to recall inconsequential sequential patterns, adversely affecting recommendation quality. In this paper, we introduce a novel approach to augmenting sequential recommendation by integrating predictability awareness into subsequence modeling. Our method begins by discerning the predictability of target items; those easily predicted often align with the preceding subsequence, while those that are hard to predict typically indicate transitions to other subsequences. Leveraging this predictability information, we enhance the discovery of meaningful subsequences within individual user behavior sequences. Evaluation of four benchmark data sets using various state‐of‐the‐art sequential models illustrates the efficacy of our approach in enhancing recommendation performance. © 2024 Institute of Electrical Engineer of Japan and Wiley Periodicals LLC.

Sequence basis of transcription initiation in the human genome.

Transcription initiation is a process that is essential to ensuring the proper function of any gene, yet we still lack a unified understanding of sequence patterns and rules that explain most transcription start sites in the human genome. By predicting transcription initiation at base-pair resolution from sequences with a deep learning-inspired explainable model called Puffin, we show that a small set of simple rules can explain transcription initiation at most human promoters. We identify key sequence patterns that contribute to human promoter activity, each activating transcription with distinct position-specific effects. Furthermore, we explain the sequence basis of bidirectional transcription at promoters, identify the links between promoter sequence and gene expression variation across cell types, and explore the conservation of sequence determinants of transcription initiation across mammalian species.

Group-aware graph neural networks for sequential recommendation

Sequential recommendations play a crucial role in recommender systems. Existing methods commonly focus on extracting sequential patterns within individual item sequences to make personalized recommendations. However, this paper argues that such a closure strategy has limitations. Firstly, the inherent uncertainty and sparsity within these sequences pose considerable challenges in accurately capturing sequential patterns. Secondly, ignoring group behaviors beyond the sequence hinders the modeling of user similarities, which are essential for recommendations. To address these limitations, this paper proposes a novel collaborative sequential recommendation approach, called Group-aware Graph Neural Networks (GaGNN). GaGNN seeks to incorporate group behaviors within sequential recommendations, thereby enhancing the modeling of user preferences through the assimilation of collective wisdom. GaGNN constructs two transition graphs, each offering a distinct behavioral perspective. The group transition graph is designed to capture the collaborative effects and learn collaborative information, whereas the individual transition graph aims to articulate individual behaviors, thereby extracting sequential information. Additionally, a multi-source information fusion module is designed to bridge semantic gaps between the two types of information and yield a holistic understanding of user preferences. Through rigorous experimental comparisons and ablation studies, the superiority of GaGNN and the beneficial integration of collaborative information are emphatically validated.

Unitig-centered pan-genome machine learning approach for predicting antibiotic resistance and discovering novel resistance genes in bacterial strains

In current genomic research, the widely used methods for predicting antimicrobial resistance (AMR) often rely on prior knowledge of known AMR genes or reference genomes. However, these methods have limitations, potentially resulting in imprecise predictions owing to incomplete coverage of AMR mechanisms and genetic variations. To overcome these limitations, we propose a pan-genome-based machine learning approach to advance our understanding of AMR gene repertoires and uncover possible feature sets for precise AMR classification. By building compacted de Brujin graphs (cDBGs) from thousands of genomes and collecting the presence/absence patterns of unique sequences (unitigs) for Pseudomonas aeruginosa, we determined that using machine learning models on unitig-centered pan-genomes showed significant promise for accurately predicting the antibiotic resistance or susceptibility of microbial strains. Applying a feature-selection-based machine learning algorithm led to satisfactory predictive performance for the training dataset (with an area under the receiver operating characteristic curve (AUC) of > 0.929) and an independent validation dataset (AUC, approximately 0.77). Furthermore, the selected unitigs revealed previously unidentified resistance genes, allowing for the expansion of the resistance gene repertoire to those that have not previously been described in the literature on antibiotic resistance. These results demonstrate that our proposed unitig-based pan-genome feature set was effective in constructing machine learning predictors that could accurately identify AMR pathogens. Gene sets extracted using this approach may offer valuable insights into expanding known AMR genes and forming new hypotheses to uncover the underlying mechanisms of bacterial AMR.