https://bioinform.jmir.org/issue/feed 2023-01-10T09:30:04-05:00 JMIR Publications editor@jmir.org Open Journal Systems Unless stated otherwise, all articles are open-access distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work ("first published in the Journal of Medical Internet Research...") is properly cited with original URL and bibliographic citation information. The complete bibliographic information, a link to the original publication on http://www.jmir.org/, as well as this copyright and license information must be included. Methods, web-based platforms, open data and open software tools for big data analytics, machine learning-based predictive models using genomic and imaging data, and information retrieval in biology and medicine. JMIR Bioinformatics and Biotechnology is the official journal of the MidSouth Computational Biology and Bioinformatics Society https://bioinform.jmir.org/2026/1/e70553 2026-01-13T16:00:11-05:00 Rayan Mubarak Fahim Islam Anik Jean T Rodriguez Nazmus Sakib Mohammad A Rahman

Background: Non-small cell lung cancer (NSCLC) is one of the leading causes of cancer-related mortality worldwide. PD-1 immunotherapy has shown promising results in the treatment of NSCLC; however, not all patients respond effectively to this treatment. Identifying predictive biomarkers for PD-1 therapy response is critical to improving patient outcomes and optimizing treatment strategies. Traditional methods of biomarker discovery often fall short in terms of accuracy and comprehensiveness. Recent advancements in deep learning provide a powerful approach to analyze complex genomic data and identify novel biomarkers that may predict therapeutic responses. Objective: This study aims to leverage machine learning techniques, particularly deep neural networks (DNN), to identify genomic biomarkers for predicting responses to PD-1 immunotherapy in NSCLC patients. By applying the DeepImmunoGene model to RNA-seq data, the study compares the performance of DNN, SVM, and XGBoost in predicting patient responses. It focuses on identifying key biomarkers through feature selection and deep learning that can enhance patient stratification and improve the accuracy of PD-1 immunotherapy predictions, contributing to more personalized treatment strategies. Methods: Differentially expressed genes (DEGs) were identified in RNA-seq data from 355 NSCLC patients using the LIMMA package in R, followed by preprocessing with log2 transformation. Machine learning models, including Support Vector Machines (SVM), XGBoost, and Deep Neural Networks (DNN), were employed to analyze gene expression data, with hyperparameters optimized using GridSearchCV. The DNN model's predictive performance was evaluated with permutation importance to identify genes critical for immunotherapy response. The models were trained on 284 patients, with 71 used for testing. Evaluation metrics like accuracy, AUC, precision, recall, specificity, and F1 score were used to assess performance. Statistical significance was tested using the Kruskal-Wallis test. Results: Initially, we identified 1,093 differentially expressed genes from RNA-seq data of 355 patients. We then trained models using SVM, XGBoost, and DNN to predict immunotherapy response. The DNN model outperformed both SVM and XGBoost with an accuracy of 82%, AUC of 90%, and recall of 0.85, significantly improving predictive performance by capturing non-linear relationships in gene expression data. To identify key biomarkers, we performed a permutation importance analysis, narrowing down the gene set to 98 genes. DeepImmunoGene, trained on these 98 genes, showed superior results, with an accuracy of 85% and an AUC of 90%. The top 36 upregulated genes in responders and 62 upregulated genes in non-responders were identified, which could serve as potential biomarkers for predicting response to PD-1 inhibitors. These findings suggest that the DeepImmunoGene model, with its ability to capture complex gene interactions, can reliably predict immunotherapy outcomes and provide insights into the molecular mechanisms of response, paving the way for more personalized treatment strategies. Conclusions: The DeepImmunoGene predictive model has successfully identified 36 upregulated genes that may serve as potential genomic biomarkers for predicting NSCLC patient responses to PD-1 immunotherapy. Notably, the ten most significant genes—GSTT2B, HMGA2, AC135050.2, ANKRD33B, MMP13, PLA2G2D, RASGEF1A, BIRC7, DCAF4L2, and CHMP7—offer valuable insights into the underlying mechanisms of treatment responses. These biomarkers not only help predict which patients are most likely to respond to PD-1 immunotherapy but also shed light on the molecular factors that explain non-response.

2026-01-13T16:00:11-05:00 https://bioinform.jmir.org/2026/1/e80539 2026-01-08T16:00:12-05:00 Qiang Fu Wei Ji Yu-Ping Fan Jian Yao Ming-Xia Song Qiao-Jing Yan

Background: Plant-derived exosome-like nanovesicles (P-ELNs) effectively deliver bioactive compounds due to their high biocompatibility and low immunogenicity. While LC-MS profiles compounds in complex samples, its analysis of large datasets remains limited by traditional methods. Recent advances in large language models (LLMs) and domain-specific systems now enhance Chinese biomedical data processing and cross-modal pharmaceutical research. Objective: To create a multimodal framework of liquid chromatography-mass spectrometry (LC-MS) combined with DeepSeek models for data mining of compounds with wound-healing properties from exosome-like nanovesicles derived from Cayratia japonica (CJ-ELNs). Methods: LC-MS identified compounds enriched in CJ (N=3) and CJ-ELNs (N=3), then compounds specifically enriched in CJ-ELNs were filtered via a four-step filtering workflow. The CJ-ELNs-specific compounds were processed by DeepSeek models for screening naturally active compounds with targeted functions of antioxidation, anti-inflammation, anti-cellular damage, anti-apoptosis, wound healing and tissue regeneration, and cell proliferation. Results: A multimodal framework of LC-MS combined with the DeepSeek-DF model was created. With the assistance of artificial intelligence (AI), a total of 46 naturally active compounds derived from CJ-ELNs with targeted functions were identified. Conclusions: A self-designed multimodal framework of LC-MS combined with DeepSeek models rapidly and accurately identifies naturally active compounds from CJ-ELNs. This AI-powered system innovatively integrates the traditional analytical technique with modern large language models, thus greatly favoring data mining of active ingredients in traditional Chinese medicine (TCM) herbs.

2026-01-08T16:00:12-05:00 https://bioinform.jmir.org/2026/1/e70708 2026-01-06T16:30:03-05:00 Marvin N Carlisle William A Pace Andrew W Liu Robert Krumm Janet E Cowan Peter R Carroll Matthew R Cooperberg Anobel Y Odisho

Background: Manual abstraction of unstructured clinical data is often necessary for granular clinical outcomes research but is time consuming and can be of variable quality. Large language models (LLMs) show promise in medical data extraction yet integrating them into research workflows remains challenging and poorly described. Objective: To develop and integrate an LLM-based system for automated data extraction from unstructured electronic health record (EHR) text reports within an established clinical outcomes database. Methods: We implemented a generative artificial intelligence (genAI) pipeline (UODBLLM) utilizing a flexible language model interface that supports various LLM implementations, including Health Insurance Portability and Accountability Act (HIPAA)-compliant cloud services and local open-source models. We used Extensible Markup Language (XML)-structured prompts and integrated using an open database connectivity interface to generate structured data from clinical documentation in the EHR. We evaluated UODBLLM's performance on completion rate, processing time, and extraction capabilities across multiple clinical data elements, including quantitative measurements, categorical assessments, and anatomical descriptions, using sample MRI reports as test cases. System reliability was tested across multiple batches to assess scalability and consistency. Results: Piloted against MRI reports, UODBLLM processed 1,800 clinical documents with a 100% completion rate and an average processing time of 8.90 seconds per report. Token utilization averaged 2,692 tokens per report, with an input-to-output ratio of approximately 13:2, resulting in a processing cost of $0.009 per report. UODBLLM had consistent performance across 18 batches of 100 reports each and completed all processing in 4.45 hours. From each report, UODBLLM extracted 16 structured clinical elements, including prostate volume, PSA values, PI-RADS scores, clinical staging, and anatomical assessments. All extracted data was automatically validated against predefined schemas and stored in standardized JSON format. Conclusions: We demonstrated successful integration of an LLM-based extraction system within an existing clinical outcomes database, achieving rapid, comprehensive data extraction at minimal cost. UODBLLM provides a scalable, efficient solution for automating clinical data extraction while maintaining protected health information security. This approach could significantly accelerate research timelines and expand feasible clinical studies, particularly for large-scale database projects.

2026-01-06T16:30:03-05:00 https://bioinform.jmir.org/2025/1/e89673 2025-12-29T17:00:11-05:00 Edem K Netsey Samuel M Naandam Joseph Asante Jnr Kuukua E Abraham Aayire C Yadem Gabriel Owusu Jeffrey G Shaffer Sudesh K Srivastav Seydou Doumbia Ellis Owusu-Dabo Chris E Morkle Desmond Yemeh Stephen Manortey Ernest Yankson Mamadou Sangare Samuel Kakraba

 

2025-12-29T17:00:11-05:00 https://bioinform.jmir.org/2025/1/e83872 2025-12-08T16:15:04-05:00 Diana Isabel Pachón-Suárez Germán Mejía-Salgado Oscar Correa Andrés Sánchez Marlon Munera Alejandra de-la-Torre

Background: Adalimumab, a monoclonal antibody targeting TNFα, treats autoimmune diseases but induces anti-drug antibodies in 30–60% of patients, reducing its efficacy. Objective: This study investigates molecular mimicry as a mechanism behind this immunogenicity, where bacterial immunoglobulin domains structurally resemble adalimumab’s light chain, triggering immune responses. Methods: Using PSI-BLASTp and IBIVU Praline, there are 40 bacterial antigens homologous to adalimumab, with eight clinically relevant strains. Results: Structural analysis revealed 94% amino acid identity between the immunoglobulin domain of Escherichia coli strain B1 and adalimumab’s light chain, and 89.67% similarity with Corynebacterium pyruviciproducens. Root Mean Square Deviation values confirmed strong structural homology. Additionally, five cross-reactive B-cell epitopes were predicted, suggesting overlapping surfaces that may promote immune cross-reactivity and anti-drug antibody development. Conclusions: These findings emphasize the role of bacterial immunoglobulin domains in adalimumab immunogenicity and highlight further need for experimental validation and improved strategies to reduce immune responses in biological therapies.

2025-12-08T16:15:04-05:00 https://bioinform.jmir.org/2025/1/e73637 2025-12-08T16:15:04-05:00 Edem K Netsey Samuel M Naandam Joseph Asante Jnr Kuukua E Abraham Aayire C Yadem Gabriel Owusu Jeffrey G Shaffer Sudesh K Srivastav Seydou Doumbia Ellis Owusu-Dabo Chris E Morkle Desmond Yemeh Stephen Manortey Ernest Yankson Mamadou Sangare Samuel Kakraba

Background: The COVID-19 pandemic requires a deep understanding of SARS-CoV-2, particularly how mutations in the Spike Receptor Binding Domain (RBD) Chain E affect its structure and function. Current methods lack comprehensive analysis of these mutations at different structural levels. Objective: To analyze the impact of specific COVID-19 associated point mutations (N501Y, L452R, N440K, K417N, E484A) on the SARS-CoV-2 Spike RBD structure and function using predictive modeling, including a graph-theoretic model, protein modeling techniques, and molecular dynamics simulations. Methods: The study employed a multi-tiered graph-theoretic framework to represent protein structure across three interconnected levels. This model incorporated 19 top-level vertices, connected to intermediate graphs based on 6-angstrom proximity within the protein's 3D structure. Graph-theoretic molecular descriptors/invariants were applied to weigh vertices and edges at all levels. The study also used Iterative Threading Assembly Refinement (I-TASSER) to model mutated sequences and molecular dynamic simulation (MD) tools to evaluate changes in protein folding and stability compared to the wildtype. Results: Three distinct predictive modeling and analytical approaches successfully identified structural and functional changes in the SARS-CoV-2 Spike RBD (Chain E) resulting from point mutations. The novel graph-theoretic model detected notable structural changes, with N501Y and L452R showing the most pronounced effects on conformation and stability compared to the wildtype. K147N and E484A mutations demonstrated less significant impacts compared to the severe mutations, N501Y and L452R. Ab initio modeling and molecular simulation dynamics findings corroborated the results from graph-theoretic analysis. The multi-level analytical approach provided a comprehensive visualization of mutation effects, deepening our understanding of their functional consequences. Conclusions: This study advanced our understanding of SARS-CoV-2 Spike RBD mutations and their implications. The multi-faceted approach characterized the effects of various mutations, identifying N501Y and L452R as having the most substantial impact on RBD conformation and stability. The findings have important implications for vaccine development, therapeutic design, and variant monitoring. Our research underscores the power of combining multiple predictive analytical approaches in virology, contributing valuable knowledge to ongoing efforts against the COVID-19 pandemic and providing a framework for future studies on viral mutations and their impacts on protein structure and function.

2025-12-08T16:15:04-05:00 https://bioinform.jmir.org/2025/1/e76736 2025-11-27T16:00:06-05:00 Charissa Chou Yiğit Baykara Sean Hacking Ali Amin Liang Cheng Alper Uzun Ece Dilber Gamsiz Uzun

Background: Bladder cancer is a disease with complex perturbations in gene networks and heterogeneous in terms of histology, mutations, and prognosis. Advances in high-throughput sequencing technologies, genome-wide association studies, and bioinformatics methods have revealed greater insights into the pathogenesis of complex diseases. Network biology-based approaches have been used to demonstrate the complex physical or functional interactions between molecules which can lead to potential drug targets. Objective: There is a need to better understand gene networks and protein-protein interactions (PPI) specific to urothelial carcinoma. Methods: We performed a multi-sample PPI study comparing two urothelial carcinoma architectures: papillary and non-papillary. We used a novel PPI analysis tool, Proteinarium to identify clusters of patients with shared PPI networks in each architecture. The feature of this tool is to analyze the PPI networks of patients and visualize them in clusters based on their network similarities from any genomic data including Next Generation Sequencing (NGS). Results: We observed distinct networks for the papillary and non-papillary groups. Proteins unique to the papillary urothelial carcinoma detected in two separate datasets included UBA52, RPS27A, UBR4, CUL1, UBE2K, and CDC5L. Proteins found in the non-papillary urothelial carcinoma specific PPI network were GNB1, UBC, RHOA, FPR2, GNGT1, PIK3CA, PIK3CG, HSP90AA1, SLC11A1, CCT7, ARHGEF1, PAK1, PAK2, PSMA7, and TRIO. Conclusions: We identified distinct PPI networks specific to papillary and non-papillary urothelial carcinomas presenting unique molecular entities. Clinical Trial: N/A

2025-11-27T16:00:06-05:00 https://bioinform.jmir.org/2025/1/e68476 2025-10-20T14:15:04-04:00 Miguel Bosch Adriana Moreno Raul Colmenares Jose Arocha Sina Hoche Auris Garcia Daniela Hall Dawlyn Garcia Lindsey Rudtner Nol Salcedo Irene Bosch

Background: Sensitivity is a critical measure of lateral-flow antigen test (AT) performance, typically compared to qRT-PCR as the gold standard. For COVID-19 diagnostics, sensitivity reflects the AT’s ability to detect SARS-CoV-2 nucleoprotein. However, estimates of sensitivity can be skewed by differences in target concentration distributions within clinical sample sets, complicating performance comparisons across ATs from different suppliers. Regulatory guidelines generally recommend a balanced representation of low, mid, and high viral loads, yet real-world sample distributions are often variable. Previous studies have largely focused on raw sensitivity without adjusting for variability in viral load distribution (Ct values). While logistic regression has been used to model positive agreement as a function of viral load, no prior method adjusts sensitivity estimates based on a standardized reference distribution. Objective: To develop a method for estimating antigen test sensitivity aligned with a standard target concentration distribution using clinical test results from an uncontrolled concentration distribution. Methods: Sensitivity is calculated by modeling the probability of positive agreement (PPA) as a function of qRT-PCR cycle thresholds (Cts) through logistic regression on AT results. Raw sensitivity is computed as the ratio of AT positives to total PCR positives. Adjusted sensitivity is then derived by applying the PPA function to a reference concentration distribution, enabling uniform sensitivity comparisons across tests. This approach reduces the impact of sampling variability, as demonstrated using data from a study in Chelsea, Massachusetts, USA. Results: Over two years, paired antigen and PCR-positive tests from four AT suppliers were analyzed: A (211 tests), B (156), C (85), and D (43). Significant differences were found in Ct distributions, with suppliers A and D showing more high viral load samples, and supplier C showing more low viral load samples, leading to discrepancies in raw sensitivity. Using the PPA function estimated from each dataset, we calculated adjusted sensitivities for common reference Ct distributions, showing how sample variability affects raw sensitivity. Our method mitigated these discrepancies, enabling more accurate sensitivity comparisons across suppliers. Conclusions: This study demonstrates that real-world sensitivity estimates are vulnerable to deviations due to variability in qRT-PCR Ct distributions across studies. We introduce a novel methodology that compensates for this variability by calculating the PPA function from raw data and adjusting sensitivity based on a standardized reference distribution of Cts, ensuring more consistent and accurate sensitivity assessments. Our approach provides a robust mathematical solution for aligning sensitivity estimates with a standardized viral load distribution, enhancing the precision of this key performance metric. By adjusting for sample variability, this method improves quality control and supports regulatory oversight, offering a reliable framework for AT performance evaluation. Clinical Trial: https://clinicaltrials.gov/study/NCT05884515

2025-10-20T14:15:04-04:00 https://bioinform.jmir.org/2025/1/e76553 2025-10-10T16:00:06-04:00 Ei-Wen Yang Brigette Waldrup Enrique Velazquez-Villarreal

Background: Achieving equity in translational precision medicine requires the integration of genomic, clinical, and social determinants of health (SDoH) data to uncover disease mechanisms, personalize treatment, and reduce health disparities. Yet, existing bioinformatics tools are often hindered by fragmented data structures, steep technical barriers, and limited capacity to incorporate SDoH variables-challenges that disproportionately affect underserved populations. Objective: To address this, we developed AI-HOPE-PM (Artificial Intelligence agent for High-Optimization and Precision mEdicine in Population Metrics), a conversational AI platform that allows users to conduct multi-dimensional cancer analyses through natural language interaction. By unifying large-scale clinical, genomic, and SDoH data within a dynamic and accessible interface, AI-HOPE-PM lowers the barrier to integrative research and supports inclusive, hypothesis-driven investigation. Methods: AI-HOPE-PM leverages large language models (LLMs), structured natural language processing, retrieval-augmented generation (RAG), and an internal Python-based workflow engine to automate data ingestion, filtering, cohort stratification, and statistical analysis. The platform operates on harmonized datasets from TCGA, cBioPortal, and AACR GENIE, enriched with simulated SDoH variables such as financial strain, food insecurity, and healthcare access. Free-text queries (e.g., Compare survival outcomes in CRC patients with TP53 mutations and limited access to care) are parsed into executable scripts aligned with biomedical ontologies. The system performs survival modeling, odds ratio testing, and case-control comparisons, generating interpretable visualizations and narrative reports in real time. Benchmarking against platforms like cBioPortal and UCSC Xena demonstrated 92.5% query interpretation accuracy and efficient performance across both CPU and GPU cloud environments. Results: AI-HOPE-PM successfully translated diverse user queries into real-time, executable analyses across colorectal cancer (CRC) datasets, enabling integration of clinical, genomic, and SDoH data. In one case study, the platform identified significantly worse survival in FOLFOX-treated CRC patients with TP53 mutations experiencing financial strain (p = 0.0481). Another analysis revealed poorer progression-free survival in APC wild-type patients with good healthcare access (p = 0.0233). Additional findings highlighted the influence of social support (p = 0.0220), food insecurity (p = 0.0162), and health literacy on outcomes and treatment access. Odds ratio analyses revealed disparities in chemotherapy exposure (OR = 0.356 for food-insecure patients) and KRAS mutation prevalence by sex and literacy status. AI-HOPE-PM also surfaced racial and ethnic differences in progression-free survival, emphasizing the importance of SDoH integration in population-level cancer research. All analyses were completed in under one minute, significantly reducing manual workload and improving scalability. Conclusions: AI-HOPE-PM marks a significant leap forward in the field of precision oncology by uniting clinical, genomic, and SDoH data within a single, conversational AI framework. Instead of relying on traditional, code-heavy approaches, the platform enables users to perform complex, multi-layered analyses through simple natural language interactions. This functionality not only democratizes access to integrative cancer research but also enhances the ability to uncover disparities in outcomes linked to genetic, clinical, and social variables. By contextualizing molecular insights within real-world social environments, AI-HOPE-PM delivers a more comprehensive understanding of cancer biology and care inequities. Its high performance, interpretability, and scalability position it as a powerful tool for accelerating hypothesis generation, guiding biomarker discovery, and informing equity-driven treatment strategies. As a flexible and user-centered platform, AI-HOPE-PM lays the groundwork for a new paradigm in AI-assisted, health equity-focused translational research.

2025-10-10T16:00:06-04:00 https://bioinform.jmir.org/2025/1/e80735 2025-10-08T15:00:05-04:00 Mahdieh Shabanian Nima Pouladi Liam Wilson Mattia Prosperi Yves A Lussier

Background: Ninety percent of the 65,000 human diseases are infrequent, collectively affecting ~ 400 million peo-ple, substantially limiting cohort accrual. This low prevalence constrains the development of robust transcriptome-based machine learning (ML) classifiers. Standard data-driven classifiers typically require cohorts of over 100 subjects per group to achieve clinical accuracy while managing high-dimensional input (~25,000 transcripts). These requirements are infeasible for micro-cohorts of ~20 individuals, where overfitting becomes pervasive. Objective: To overcome these constraints, we developed a classification method that integrates three enabling strategies: (i) paired-sample transcriptome dynamics, (ii) N-of-1 pathway-based analytics, and (iii) reproducible machine learning operations (MLOps) for continuous model refinement. Methods: Unlike ML approaches relying on a single transcriptome per subject, within-subject paired-sample designs — such as pre- versus post-treatment or diseased versus adjacent-normal tissue — effectively control intra-individual variability under isogenic conditions and within-subject environmental expo-sures (e.g. smoking history, other medications, etc.), improve signal-to-noise ratios, and, when pre-processed as single-subject studies (N-of-1), can achieve statistical power comparable to that ob-tained in animal models. Pathway-level N-of-1 analytics further reduces each sample’s high-dimensional profile into ~4,000 biologically interpretable features, annotated with effect sizes, dis-persion, and significance. Complementary MLOps practices—automated versioning, continuous monitoring, and adaptive hyperparameter tuning—improve model reproducibility and generalization. Results: In two case studies of distinct diseases, human rhinovirus infection (HRV) versus matched healthy controls (n=16 training; 3 test) and breast cancer tissues harboring TP53 or PIK3CA mutations versus adjacent normal tissue (n=27 training; 9 test)—this approach achieved 90% precision and recall on an unseen breast cancer test set and 92% precision with 90% recall in rhinovirus fivefold cross-validation. Incorporating paired-sample dynamics boosted precision by up to 12% and recall by 13% in BC, and by 5% each in HRV. MLOps workflows yielded an additional ~14.5% accuracy im-provement compared to traditional pipelines. Moreover, our method identified 42 critical gene sets (pathways) for rhinovirus response and 21 for breast cancer mutation status, selected as the most im-portant features (Mean Decrease Impurity) of the best performing model, with retroactive ablation of Top-20 features reducing accuracy by ~25%. Conclusions: These proof-of-concept results support the utility of integrating intra-subject dynamics, “biological knowledge”-based feature reduction (pathway-level feature reduction grounded in prior biological knowledge; e.g., N-of-1-pathways analytics), and reproducible MLOps workflows can overcome cohort-size limitations in infrequent disease, offering a scalable, interpretable solution for high-dimensional transcriptomic classification. Future work will extend these advances across various therapeutic and small-cohort designs. https://github.com/shabanian2018/MLOps-Micro-Cohort Clinical Trial: Not applicable

2025-10-08T15:00:05-04:00