Briefings in bioinformatics最新文献

Structure-based virulence factor classification using a dual-driven graph transformer with a pretrained language model. 基于结构的毒力因子分类，使用双驱动图转换器和预训练语言模型。

IF 7.3 2区生物学

Briefings in bioinformatics Pub Date : 2026-09-01 DOI: 10.1093/bib/bbag508

Peihao Bai, Guanghui Li, Nan Jiang, Jiao Chen, Cheng Liang, Yijie Ding

{"title":"Structure-based virulence factor classification using a dual-driven graph transformer with a pretrained language model.","authors":"Peihao Bai, Guanghui Li, Nan Jiang, Jiao Chen, Cheng Liang, Yijie Ding","doi":"10.1093/bib/bbag508","DOIUrl":"10.1093/bib/bbag508","url":null,"abstract":"Bacterial pathogenicity relies on a multitude of virulence factors (VFs). Identifying these factors is crucial for comprehending the molecular mechanisms driving bacterial pathogenesis and pinpointing potential targets for antivirulence strategies. While the identification of VFs has received extensive attention, there is still a lack of prediction from the perspective of VF classes. Moreover, most computational methods focus solely on sequence information for virulence proteins, neglecting spatial structure information. Here, a novel Structure-based Dual-driven Graph Transformer framework named SDGT is proposed to identify multiple distinct VFs by leveraging graph data structures of virulence proteins and sequence representations generated from a pretrained protein language model. By representing virulence proteins at both the atom and residue levels, SDGT adaptively and comprehensively learns a graph representation of protein structure through the utilization of self-attention pooling. The results of extensive experiments demonstrate that SDGT significantly enhances the classification performance, achieving an accuracy of 0.7623 on an independent test dataset, outperforming other sequence-based and structure-based methods. Moreover, cluster analysis of pathogen VF features learned by SDGT proves the excellent specificity and similarity feature learning power of the proposed model. The current analysis suggests that SDGT is an effective tool for identifying of VF classes.","PeriodicalId":9209,"journal":{"name":"Briefings in bioinformatics","volume":"27 5","pages":""},"PeriodicalIF":7.3,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13577672/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148975779","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

From prior knowledge to data-informed models: a review of Boolean network inference. 从先验知识到数据知情模型：布尔网络推理综述。

IF 7.3 2区生物学

Briefings in bioinformatics Pub Date : 2026-09-01 DOI: 10.1093/bib/bbag499

Pierre Klemmer, Ahmed Abdelmonem Hemedan, Reinhard Schneider, Marek Ostaszewski

{"title":"From prior knowledge to data-informed models: a review of Boolean network inference.","authors":"Pierre Klemmer, Ahmed Abdelmonem Hemedan, Reinhard Schneider, Marek Ostaszewski","doi":"10.1093/bib/bbag499","DOIUrl":"10.1093/bib/bbag499","url":null,"abstract":"Molecular mechanisms are highly complex and involve numerous components in non-linear interactions, imposing challenges in analysis and understanding. Networks provide intuitive representations of these systems, enabling visualisation of experimental data and the aggregation of prior knowledge. However, static networks cannot account for key temporal dynamics. Boolean networks address this limitation as discrete dynamical models representing temporal properties of biological regulation through binary variables connected by logical functions, offering scalable frameworks for hypothesis generation while avoiding parameterisation challenges of quantitative modelling approaches. Construction of (Boolean) networks is a mostly manual, labour-intensive and bias-prone process, relying on the review of a large corpora of prior knowledge. This review surveys methods for Boolean network inference that integrate experimental data with or without prior knowledge to automatically construct context-specific, data-informed models. We begin by examining network curation methods from text mining to manual curation given their use as backbones of Boolean networks, then analyse several inference algorithms categorised as heuristic (MIBNI, ATEN, LogicGep, CANTATA, and CellNetOptimizer) and exact (BoNesis, caspo, Sketchbook, RE:IN, and Griffin) methods. For each method, we discuss computational representation, inputs, inference algorithms, and benchmarking. We identify key considerations for method selection, including handling of uncertainty, choice of updating scheme, scalability, and data discretisation challenges. We advise that method selection be guided by the type and extent of available prior knowledge and experimental data, and modelling objectives.","PeriodicalId":9209,"journal":{"name":"Briefings in bioinformatics","volume":"27 5","pages":""},"PeriodicalIF":7.3,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13577669/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148975784","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

circMAC: microRNA-conditioned binding-site localization on circular RNA isoforms. circMAC：环状RNA异构体上的微RNA条件结合位点定位。

IF 7.3 2区生物学

Briefings in bioinformatics Pub Date : 2026-09-01 DOI: 10.1093/bib/bbag492

Juseong Kim, Sanghun Sel, Giltae Song

{"title":"circMAC: microRNA-conditioned binding-site localization on circular RNA isoforms.","authors":"Juseong Kim, Sanghun Sel, Giltae Song","doi":"10.1093/bib/bbag492","DOIUrl":"10.1093/bib/bbag492","url":null,"abstract":"Circular RNAs (circRNAs) regulate gene expression in part through interactions with microRNAs (miRNAs), but identifying miRNA binding sites on full-length circRNA isoforms remains challenging. Binding context can differ across circRNA isoforms, and sequence continuity across the back-splice junction may be overlooked when circRNAs are represented as linear transcripts. Existing circRNA-miRNA resources and computational approaches mainly support association-level prediction or rule-based candidate-site screening. Although rule-based tools can be applied to full-length circRNA sequences, they do not directly learn miRNA-conditioned nucleotide-level binding-site localization while preserving circular sequence continuity. Here, we formulate circRNA-miRNA binding-site prediction as a miRNA-conditioned sequence-labeling task and present circMAC, a purpose-built framework for nucleotide-level localization on full-length circRNA isoforms. Given a full-length circRNA isoform and a mature miRNA sequence, circMAC predicts a binding probability for each circRNA nucleotide. circMAC combines established sequence-modeling components in a task-specific architecture, including attention-based global context modeling, Mamba-based sequential modeling, and convolutional local motif extraction. Paired miRNA information is incorporated through cross-attention, allowing each circRNA nucleotide to be evaluated in a miRNA-specific context. We evaluated circMAC against pretrained RNA language models, conventional sequence encoders, alternative pretraining strategies, architectural ablations, and stricter isoform-disjoint and back-splice-junction-disjoint splits. circMAC showed improved nucleotide-level localization performance under the evaluated benchmark settings, while qualitative and aggregate analyses indicated concentration of prediction signals around annotated binding-site regions. These results support task-specific full-length circular isoform modeling for prioritizing candidate circRNA-miRNA binding-site regions.","PeriodicalId":9209,"journal":{"name":"Briefings in bioinformatics","volume":"27 5","pages":""},"PeriodicalIF":7.3,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13557716/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148918668","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A computational framework for proteome-wide target profiling of natural products: mechanistic and therapeutic insights into ginsenosides. 天然产物蛋白质组范围靶标分析的计算框架：人参皂苷的机制和治疗见解。

IF 7.3 2区生物学

Briefings in bioinformatics Pub Date : 2026-09-01 DOI: 10.1093/bib/bbag483

Dohyeon Kim, Charuvaka Muvva, Jung-Seok Yang, Hak Cheol Kwon, Cheol-Ho Pan, Kyungsu Kang, Taejung Kim, Ki Cheong Park, Keunwan Park

{"title":"A computational framework for proteome-wide target profiling of natural products: mechanistic and therapeutic insights into ginsenosides.","authors":"Dohyeon Kim, Charuvaka Muvva, Jung-Seok Yang, Hak Cheol Kwon, Cheol-Ho Pan, Kyungsu Kang, Taejung Kim, Ki Cheong Park, Keunwan Park","doi":"10.1093/bib/bbag483","DOIUrl":"10.1093/bib/bbag483","url":null,"abstract":"Natural products (NPs) serve as valuable sources of pharmacologically active compounds, yet their therapeutic potential remains constrained by incomplete knowledge of their molecular targets. To address this, a computational framework based on evolutionary chemical binding similarity (ECBS) is developed to generate proteome-wide NP-target binding profiles. By integrating evolutionarily related ligands and optimizing chemical pairing strategies, the framework constructs target-specific ECBS models covering 6203 protein targets with enhanced predictive performance. As a proof of concept, this approach is applied to ginsenosides, a structurally diverse NP class with broad pharmacological relevance. The analysis reveals comprehensive ginsenoside-target associations, enabling functional clustering, inference of mechanisms of action, and identification of chemotype-specific targets. Experimental validation confirms a novel direct interaction with protein kinase C isoforms by a direct binding assay, and provides phenotypic (indirect) evidence consistent with activity at the sarcoplasmic/endoplasmic reticulum calcium adenosine triphosphatase (SERCA), demonstrating the practical utility of the approach. To enhance data accessibility and exploration, GinsenBank (http://ginseng.aslla.net) is developed as an open web resource providing predicted binding profiles, molecular clustering, multi-target analysis, and drug similarity assessment. These findings highlight the potential of proteome-scale NP-target profiling to accelerate the discovery of novel therapeutic applications for NPs.","PeriodicalId":9209,"journal":{"name":"Briefings in bioinformatics","volume":"27 5","pages":""},"PeriodicalIF":7.3,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13557714/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148918737","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

DLRNA-BERTa: a transformer approach for predicting molecule-RNA binding affinities. DLRNA-BERTa：一种预测分子- rna结合亲和力的转换方法。

IF 7.3 2区生物学

Briefings in bioinformatics Pub Date : 2026-09-01 DOI: 10.1093/bib/bbag488

Pasquale Lobascio, Khalid Saeed, Asifullah Khan, Ziaurrehman Tanoli

{"title":"DLRNA-BERTa: a transformer approach for predicting molecule-RNA binding affinities.","authors":"Pasquale Lobascio, Khalid Saeed, Asifullah Khan, Ziaurrehman Tanoli","doi":"10.1093/bib/bbag488","DOIUrl":"10.1093/bib/bbag488","url":null,"abstract":"Therapies targeting RNA are rapidly expanding, with 24 FDA-approved RNA therapeutics and over 130 currently in clinical trials, highlighting RNA's growing role in drug discovery. In this context, transformer-based language models provide a scalable and cost-effective approach to accelerate RNA-targeted drug discovery by enabling the prediction of molecule-RNA binding affinities directly from sequence information. This study introduces DLRNA-BERTa, a RoBERTa-based framework that integrates RNA-BERTa and ChemBERTa-v2 to model molecule-RNA binding affinities. The framework includes six RNA class-specific models, aptamers, repeats, ribosomal RNAs, riboswitches, microRNAs, and viral RNAs, along with a general model for other RNA classes. DLRNA-BERTa outperformed existing approaches across different RNA classes on a randomly split validation dataset. On an independent and relatively diverse test set, the model achieved AUROC values of 0.57-0.59, demonstrating performance comparable to the other evaluated methods. Application of DLRNA-BERTa to a library of 3492 approved drugs identified 2859 compounds with predicted binding affinities (pKd ≥ 6) across 294 RNA targets, suggesting its potential utility for RNA-based drug repurposing and prioritization of candidate molecules for further investigation. To facilitate broader use and reproducibility, we provide a publicly accessible web application and API are available at https://huggingface.co/spaces/IlPakoZ/DLRNA-BERTa, enabling users to predict binding affinities between custom compounds and RNA sequences.","PeriodicalId":9209,"journal":{"name":"Briefings in bioinformatics","volume":"27 5","pages":""},"PeriodicalIF":7.3,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13563671/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148944339","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

DNAmBERT: a transformer-based model for non-invasive cancer diagnosis using DNA sequence and methylation data. DNAmBERT：基于变压器的非侵入性癌症诊断模型，使用DNA序列和甲基化数据。

IF 7.3 2区生物学

Briefings in bioinformatics Pub Date : 2026-09-01 DOI: 10.1093/bib/bbag455

Maryam Yassi, Mark Ezegbogu, Euan J Rodger, Peter Stockwell, Aniruddha Chatterjee, Matthew Parry

{"title":"DNAmBERT: a transformer-based model for non-invasive cancer diagnosis using DNA sequence and methylation data.","authors":"Maryam Yassi, Mark Ezegbogu, Euan J Rodger, Peter Stockwell, Aniruddha Chatterjee, Matthew Parry","doi":"10.1093/bib/bbag455","DOIUrl":"10.1093/bib/bbag455","url":null,"abstract":"DNA methylation alterations are early and stable hallmarks of cancer and represent promising biomarkers for non-invasive detection using circulating cell-free DNA (cfDNA). However, current computational approaches often model DNA sequence and methylation features separately and struggle to capture complex read-level methylation architecture in heterogeneous, low-signal liquid biopsy data. Here, we present DNAmBERT, a Transformer-based deep learning framework designed to jointly model DNA sequence context and read-level methylation haplotype structure from cfDNA methylation sequencing data. DNAmBERT integrates k-mer-encoded DNA sequences with methylation haplotype tokens using a unified representation and masked language modelling objective, enabling context-aware learning of sequence-epigenetic dependencies through self-attention. We evaluated DNAmBERT across multiple cfDNA methylation platforms (RRBS, cfRRBS, and cfMethyl-seq) and cancer types, including colorectal cancer, lung adenocarcinoma and hepatocellular carcinoma. In binary classification tasks, the model achieved high performance across platforms (AUC up to 0.99-1.00) and outperformed conventional machine learning and existing deep learning approaches. Aggregation of read-level predictions enabled quantitative tumour probability estimation at the sample level. Beyond binary detection, DNAmBERT supported multi-cancer and stage-aware classification, including early-stage disease, with multiclass AUC values up to 0.99. The framework further demonstrated effective cross-cancer transfer learning, maintaining robust performance under limited data availability. These results indicate that integrated sequence-haplotype representation learning provides an accurate and scalable approach for cfDNA-based multi-cancer detection.","PeriodicalId":9209,"journal":{"name":"Briefings in bioinformatics","volume":"27 5","pages":""},"PeriodicalIF":7.3,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13537528/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148878782","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Machine learning approaches for cancer prognosis and diagnosis via non-coding RNA: a comprehensive review. 通过非编码RNA进行癌症预后和诊断的机器学习方法：综述。

IF 7.3 2区生物学

Briefings in bioinformatics Pub Date : 2026-09-01 DOI: 10.1093/bib/bbag346

Md Shafayet Hossain, Yi-Ping Phoebe Chen

{"title":"Machine learning approaches for cancer prognosis and diagnosis via non-coding RNA: a comprehensive review.","authors":"Md Shafayet Hossain, Yi-Ping Phoebe Chen","doi":"10.1093/bib/bbag346","DOIUrl":"10.1093/bib/bbag346","url":null,"abstract":"Non-coding RNAs (ncRNAs), once considered genomic dark matter, are now established as key regulators of gene expression with widespread roles in cellular homeostasis and disease. In cancer, ncRNA expression is frequently and systematically dysregulated, and many of these molecules circulate in stable, protected form within biofluids, offering a compelling basis for non-invasive or minimally invasive diagnostic strategies. However, their clinical translation remains substantially hindered to date due to biological complexity, technical noise, and high dimensionality inherent to ncRNA expression datasets. In this context, machine learning (ML) has emerged as a powerful analytical tool to address these challenges, enabling the identification of subtle, reproducible ncRNA signatures predictive of diverse malignancies. This review critically evaluates ML-driven frameworks for cancer diagnosis and prognosis across four ncRNA subclasses, namely miRNAs, lncRNAs, circRNAs, and piRNAs, while also acknowledging the biophysical and thermodynamic models that reinforce ncRNA bioinformatics. Despite substantial methodological progress in ML-based cancer diagnosis and prognosis, key challenges persist, including tumor biological heterogeneity, limited multicenter validation, and the lack of widely adopted standardized protocols for preprocessing, normalization, and reporting workflows. Furthermore, many current ML models lack interpretability in biological or clinical context, constraining their translational utility. By synthesizing recent advances and identifying unresolved barriers, this review charts a roadmap for developing a robust, clinically actionable ncRNA biomarker platform for cancer detection. With global cancer incidence projected to exceed 35 million annual cases by 2050, validated ncRNA-ML-driven frameworks hold potential to revolutionize early-stage detection and personalized therapeutic strategies, thereby reducing the escalating socio-economic burden of cancer worldwide.","PeriodicalId":9209,"journal":{"name":"Briefings in bioinformatics","volume":"27 5","pages":""},"PeriodicalIF":7.3,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13546767/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148900040","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Somatic likelihood tiering: an interpretable post-calling triage protocol for tumor-only whole-exome variant review. 体细胞可能性分级：一种可解释的呼叫后分诊方案，用于肿瘤全外显子组变异回顾。

IF 7.3 2区生物学

Briefings in bioinformatics Pub Date : 2026-09-01 DOI: 10.1093/bib/bbag487

Konrad Stawiski, Sophia C Kamran, Júlia Perera-Bel, Jihyun Lee, Joaquim Bellmunt, Kent W Mouw, Filipe L F De Carvalho

{"title":"Somatic likelihood tiering: an interpretable post-calling triage protocol for tumor-only whole-exome variant review.","authors":"Konrad Stawiski, Sophia C Kamran, Júlia Perera-Bel, Jihyun Lee, Joaquim Bellmunt, Kent W Mouw, Filipe L F De Carvalho","doi":"10.1093/bib/bbag487","DOIUrl":"10.1093/bib/bbag487","url":null,"abstract":"Tumor-only whole-exome sequencing (WES) is used when matched normal tissue is unavailable, but one sample can produce thousands of variants. Somatic likelihood tiering (SLT) is an interpretable post-calling protocol that ranks Mutect2 calls into four review-priority tiers using population-frequency, germline-quality, cancer-knowledge, PureCN posterior, and clonal-hematopoiesis evidence. Layer 2 distinguishes common, rare-callable, and unevaluable gnomAD states; missing or unmatchable gnomAD evidence is not positive rarity evidence. On the SEQC2 HCC1395 benchmark, the callability-aware SLT-A row contained 101 calls, 78 truth variants, 77.2% PPV (95% Wilson confidence interval 68.1%-84.3%), and a Number Needed to Review (NNR) of 1.29 (1.19-1.47). The conservative SLT-C catchment retained 352 of 455 truth variants (77.4%, 73.3%-81.0%) and all tiers together retained 430 of 455 truth variants. SNV performance is the primary calibration frame: SLT-C retained 341 of 439 SNV truth variants, whereas indel results were exploratory because only 16 truth indels were available. Clinical cohorts are reported as recall and concordance versus partially dependent matched-normal Mutect2 references, not independent clinical sensitivity. Patient-level bootstrap intervals were principal: HdM-BLCA-1 SLT-A recall was 18.2% (14.0%-23.5%), and LUAD-TW SLT-A recall was 49.1% (26.6%-63.3%) among 32 evaluable patients. The HdM-BLCA-1 median SLT-A queue remained 1277 variants per patient, so SLT reduces first-pass candidate counts but does not measure review time or eliminate FFPE candidate-count burden. SLT provides an auditable tumor-only WES review queue, not a substitute for matched-normal sequencing, independent orthogonal validation, or definitive somatic classification.","PeriodicalId":9209,"journal":{"name":"Briefings in bioinformatics","volume":"27 5","pages":""},"PeriodicalIF":7.3,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13561304/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148928973","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

DFRL-Mol: a dual-stage framework of reinforcement learning for multi-scenario molecule optimization. DFRL-Mol：用于多场景分子优化的双阶段强化学习框架。

IF 7.3 2区生物学

Briefings in bioinformatics Pub Date : 2026-09-01 DOI: 10.1093/bib/bbag493

Zhenyi Wu, Pengcheng Zhao, Jianing Li, Qiong Wang, Qin Zhang, Hui Yu, Zhe Yu, Jian-Yu Shi

{"title":"DFRL-Mol: a dual-stage framework of reinforcement learning for multi-scenario molecule optimization.","authors":"Zhenyi Wu, Pengcheng Zhao, Jianing Li, Qiong Wang, Qin Zhang, Hui Yu, Zhe Yu, Jian-Yu Shi","doi":"10.1093/bib/bbag493","DOIUrl":"10.1093/bib/bbag493","url":null,"abstract":"As a critical step in drug discovery, molecule optimization aims to improve specific properties of lead compounds. Inspired by isosteres (i.e. molecular substructures with similar reactive electron shells), existing AI-based methods have been proposed for molecule optimization. However, they often rely on available matched molecular pairs (MMPs), thus facing two essential challenges. First, the limited diversity of available MMPs leads to biased training and a lack of novelty in the optimized molecules, in both single-objective (SO) and multi-objective (MO) optimization. Second, the low number of available MMPs leads to limited generalization in MO optimization. To overcome the limitations of MMPs, we propose DFRL-Mol, a novel dual-stage reinforcement learning framework. It consists of three main modules: the Structure-Property Analyst (SPA), the Decorator (DCR), and Curriculum-guided Dual-Stage Policy Optimization (CDSPO). SPA and DCR are two LLM-based models, which are responsible for locating the substructures to be optimized and inferring their appropriate isosteric replacement, respectively. CDSPO leverages curriculum-guided reinforcement learning to facilitate collaboration between the two models. DFRL-Mol ensures the generation of diverse molecular representations while effectively overcoming the issues of bias and data sparsity associated with MMPs. The comparisons with state-of-the-art methods demonstrate the enhanced performance of DFRL-Mol in multiple scenarios of molecule optimization, including SO optimization, substructure-constrained (SC) SO optimization, MO optimization, and MO molecule generation. More importantly, detailed analysis of experimental results demonstrates that the models' behavior is consistent with chemical intuition.","PeriodicalId":9209,"journal":{"name":"Briefings in bioinformatics","volume":"27 5","pages":""},"PeriodicalIF":7.3,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13561309/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148928870","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Response to "Caution: Fundamental data quality issues underlying intelligent prediction of enzyme optimal pH". 对“警告：智能预测酶最优pH值的基础数据质量问题”的回应。

IF 7.3 2区生物学

Briefings in bioinformatics Pub Date : 2026-09-01 DOI: 10.1093/bib/bbag537

Sizhe Qiu, Aidong Yang

引用次数: 0