Background: The systemic treatment of cancer typically requires the use of multiple anticancer agents in combination or sequentially. Clinical narrative texts often contain extensive descriptions of the temporal sequencing of systemic anticancer therapy (SACT), setting up an important task that may be amenable to automated extraction of SACT timelines.
Objective: We aimed to explore automatic methods for extracting patient-level SACT timelines from clinical narratives in the electronic medical records (EMRs).
Methods: We used two datasets from two institutions: (1) a colorectal cancer (CRC) dataset including the entire EMR of the 199 patients in the THYME (Temporal Histories of Your Medical Event) dataset and (2) the 2024 ChemoTimelines shared task dataset including 149 patients with ovarian cancer, breast cancer, and melanoma. We explored finetuning smaller language models trained to attend to events and time expressions, and few-shot prompting of large language models (LLMs). Evaluation used the 2024 ChemoTimelines shared task configuration-Subtask1 involving the construction of SACT timelines from manually annotated SACT event and time expression mentions provided as input in addition to the patient's notes and Subtask2 requiring extraction of SACT timelines directly from the patient's notes.
Results: Our task-specific finetuned EntityBERT model achieved 93% F1-score, outperforming the best results in Subtask1 of the 2024 ChemoTimelines shared task (90%). It ranked second in Subtask2. LLM (LLaMA2, LLaMA3.1, and Mixtral) performance lagged the task-specific finetuned model performance for both the THYME and shared task datasets. On the shared task datasets, the best LLM performance was 77% macro F1-score, 16% points lower than the task-specific finetuned system (Subtask1).
Conclusions: In this paper, we explored approaches for patient-level timeline extraction through the SACT timeline extraction task. Our results and analysis add to the knowledge of extracting treatment timelines from EMR clinical narratives using language modeling methods.
Background: Deep learning (DL) shows promise for automated lung cancer diagnosis, but limited clinical data can restrict performance. While data augmentation (DA) helps, existing methods struggle with chest computed tomography (CT) scans across diverse DL architectures.
Objective: This study proposes Random Pixel Swap (RPS), a novel DA technique, to enhance diagnostic performance in both convolutional neural networks and transformers for lung cancer diagnosis from CT scan images.
Methods: RPS generates augmented data by randomly swapping pixels within patient CT scan images. We evaluated it on ResNet, MobileNet, Vision Transformer, and Swin Transformer models, using 2 public CT datasets (Iraq-Oncology Teaching Hospital/National Center for Cancer Diseases [IQ-OTH/NCCD] dataset and chest CT scan images dataset), and measured accuracy and area under the receiver operating characteristic curve (AUROC). Statistical significance was assessed via paired t tests.
Results: The RPS outperformed state-of-the-art DA methods (Cutout, Random Erasing, MixUp, and CutMix), achieving 97.56% accuracy and 98.61% AUROC on the IQ-OTH/NCCD dataset and 97.78% accuracy and 99.46% AUROC on the chest CT scan images dataset. While traditional augmentation approaches (flipping and rotation) remained effective, RPS complemented them, surpassing the performance findings in prior studies and demonstrating the potential of artificial intelligence for early lung cancer detection.
Conclusions: The RPS technique enhances convolutional neural network and transformer models, enabling more accurate automated lung cancer detection from CT scan images.
Background: The protein A disintegrin and metalloprotease (ADAM) domain containing 17, also called tumor necrosis factor alpha-converting enzyme, is mainly responsible for cleaving a specific sequence Pro-Leu-Ala-Gln-Ala-/-Val-Arg-Ser-Ser-Ser in the membrane-bound precursor of tumor necrosis factor alpha. This cleavage process has significant implications for inflammatory and immune responses, and recent research indicates that genetic variants of ADAM17 may influence susceptibility to and severity of SARS-CoV-2 infection.
Objective: The aim of the study is to identify the most deleterious missense variants of ADAM17 that impact protein stability, structure, and function and to assess specific variants potentially involved in SARS-CoV-2 infection.
Methods: A bioinformatics approach was used on 12,042 single-nucleotide polymorphisms using tools including SIFT (Sorting Intolerant From Tolerant), PolyPhen2.0, PROVEAN (Protein Variation Effect Analyzer), PANTHER (Protein Analysis Through Evolutionary Relationships), SNP&GO (Single Nucleotide Polymorphisms and Gene Ontology), PhD-SNP (Predictor of Human Deleterious Single Nucleotide Polymorphisms), Mutation Assessor, SNAP2 (Screening for Non-Acceptable Polymorphisms 2), MUpro, I-Mutant, iStable, InterPro, Sparks-x, PROCHECK (Programs to Check the Stereochemical Quality of Protein Structures), PyMol, Project HOPE (Have (y)Our Protein Explained), ConSurf, and SWISS-MODEL. Missense variants of ADAM17 were collected from the Ensembl database for analysis.
Results: In total, 7 nonsynonymous single-nucleotide polymorphisms (P556L, G550D, V483A, G479E, G349E, T339P, and D232E) were identified as high-risk pathogenic by all prediction tools, and these variants were found to potentially have deleterious effects on the stability, structure, and function of the ADAM17 protein, potentially destroying the entire cleavage process. Additionally, 4 missense variants (Q658H, D657G, D654N, and F652L) in positions related to SARS-CoV-2 infection exhibited high conservation scores and were predicted to be deleterious, suggesting that they play an important role in SARS-CoV-2 infection.
Conclusions: Specific missense variants of ADAM17 are predicted to be highly pathogenic, potentially affecting protein stability and function and contributing to SARS-CoV-2 pathogenesis. These findings provide a basis for understanding their clinical relevance, aiding in early diagnosis, risk assessment, and therapeutic development.
Background: Cancer is one of the leading causes of disease burden globally, and early and accurate diagnosis is crucial for effective treatment. This study presents a deep learning-based model designed to classify 5 common types of cancer in Saudi Arabia: breast, colorectal, thyroid, non-Hodgkin lymphoma, and corpus uteri.
Objective: This study aimed to evaluate whether integrating RNA sequencing, somatic mutation, and DNA methylation profiles within a stacking deep learning ensemble improves cancer type classification accuracy relative to the current state-of-the-art multiomics models.
Methods: Using a stacking ensemble learning approach, our model integrates 5 well-established methods: support vector machine, k-nearest neighbors, artificial neural network, convolutional neural network, and random forest. The methodology involves 2 main stages: data preprocessing (including normalization and feature extraction) and ensemble stacking classification. We prepared the data before applying the stacking model.
Results: The stacking ensemble model achieved 98% accuracy with multiomics versus 96% using RNA sequencing and methylation individually, 81% using somatic mutation data, suggesting that multiomics data can be used for diagnosis in primary care settings. The models used in ensemble learning are among the most widely used in cancer classification research. Their prevalent use in previous studies underscores their effectiveness and flexibility, enhancing the performance of multiomics data integration.
Conclusions: This study highlights the importance of advanced machine learning techniques in improving cancer detection and prognosis, contributing valuable insights by applying ensemble learning to integrate multiomics data for more effective cancer classification.
Background: National and ethnic mutation frequency databases (NEMDBs) play a crucial role in documenting gene variations across populations, offering invaluable insights for gene mutation research and the advancement of precision medicine. These databases provide an essential resource for understanding genetic diversity and its implications for health and disease across different ethnic groups.
Objective: The aim of this study is to systematically evaluate 42 NEMDBs to (1) quantify gaps in standardization (70% nonstandard formats, 50% outdated data), (2) propose artificial intelligence/linked open data solutions for interoperability, and (3) highlight clinical implications for precision medicine across NEMDBs.
Methods: A systematic approach was used to assess the databases based on several criteria, including data collection methods, system design, and querying mechanisms. We analyzed the accessibility and user-centric features of each database, noting their ability to integrate with other systems and their role in advancing genetic disorder research. The review also addressed standardization and data quality challenges prevalent in current NEMDBs.
Results: The analysis of 42 NEMDBs revealed significant issues, with 70% (29/42) lacking standardized data formats and 60% (25/42) having notable gaps in the cross-comparison of genetic variations, and 50% (21/42) of the databases contained incomplete or outdated data, limiting their clinical utility. However, databases developed on open-source platforms, such as LOVD, showed a 40% increase in usability for researchers, highlighting the benefits of using flexible, open-access systems.
Conclusions: We propose cloud-based platforms and linked open data frameworks to address critical gaps in standardization (70% of databases) and outdated data (50%) alongside artificial intelligence-driven models for improved interoperability. These solutions prioritize user-centric design to effectively serve clinicians, researchers, and public stakeholders.
Background: Prediabetes is an intermediate stage between normal glucose metabolism and diabetes and is associated with increased risk of complications like cardiovascular disease and kidney failure.
Objective: It is crucial to recognize individuals with prediabetes early in order to apply timely intervention strategies to decelerate or prohibit diabetes development. This study aims to compare the effectiveness of machine learning (ML) algorithms in predicting prediabetes and identifying its key clinical predictors.
Methods: Multiple ML models are evaluated in this study, including random forest, extreme gradient boosting (XGBoost), support vector machine (SVM), and k-nearest neighbors (KNNs), on a dataset of 4743 individuals. For improved performance and interpretability, key clinical features were selected using LASSO (Least Absolute Shrinkage and Selection Operator) regression and principal component analysis (PCA). To optimize model accuracy and reduce overfitting, we used hyperparameter tuning with RandomizedSearchCV for XGBoost and random forest, and GridSearchCV for SVM and KNN. SHAP (Shapley Additive Explanations) was used to assess model-agnostic feature importance. To resolve data imbalance, SMOTE (Synthetic Minority Oversampling Technique) was applied to ensure reliable classifications.
Results: A cross-validated ROC-AUC (receiver operating characteristic area under the curve) score of 0.9117 highlighted the robustness of random forest in generalizing across datasets among the models tested. XGBoost followed closely, providing balanced accuracy in distinguishing between normal and prediabetic cases. While SVMs and KNNs performed adequately as baseline models, they exhibited limitations in sensitivity. The SHAP analysis indicated that BMI, age, high-density lipoprotein cholesterol, and low-density lipoprotein cholesterol emerged as the key predictors across models. The performance was significantly enhanced through hyperparameter tuning; for example, the ROC-AUC for SVM increased from 0.813 (default) to 0.863 (tuned). PCA kept 12 components while maintaining 95% of the variance in the dataset.
Conclusions: It is demonstrated in this research that optimized ML models, especially random forest and XGBoost, are effective tools for assessing early prediabetes risk. Combining SHAP analysis with LASSO and PCA enhances transparency, supporting their integration in real-time clinical decision support systems. Future directions include validating these models in diverse clinical settings and integrating additional biomarkers to improve prediction accuracy, offering a promising avenue for early intervention and personalized treatment strategies in preventive health care.
Background: Previous machine learning approaches for prostate cancer detection using gene expression data have shown remarkable classification accuracies. However, prior studies overlook the influence of racial diversity within the population and the importance of selecting outlier genes based on expression profiles.
Objective: We aim to develop a classification method for diagnosing prostate cancer using gene expression in specific populations.
Methods: This research uses differentially expressed gene analysis, receiver operating characteristic analysis, and MSigDB (Molecular Signature Database) verification as a feature selection framework to identify genes for constructing support vector machine models.
Results: Among the models evaluated, the highest observed accuracy was achieved using 139 gene features without oversampling, resulting in 98% accuracy for White patients and 97% for African American patients, based on 388 training samples and 92 testing samples. Notably, another model achieved a similarly strong performance, with 97% accuracy for White patients and 95% for African American patients, using only 9 gene features. It was trained on 374 samples and tested on 138 samples.
Conclusions: The findings identify a race-specific diagnosis method for prostate cancer detection using enhanced feature selection and machine learning. This approach emphasizes the potential for developing unbiased diagnostic tools in specific populations.
Unlabelled: Artificial intelligence (AI) is poised to become an integral component in health care research and delivery, promising to address complex challenges with unprecedented efficiency and precision. However, many clinicians lack training and experience with AI, and for those who wish to incorporate AI into research and practice, the path forward remains unclear. Technical barriers, institutional constraints, and lack of familiarity with computer and data science frequently stall progress. In this tutorial, we present a transparent account of our experiences as a newly established interdisciplinary team of clinical oncology researchers and data scientists working to develop a natural language processing model to identify symptomatic adverse events during pediatric cancer therapy. We outline the key steps for clinicians to consider as they explore the utility of AI in their inquiry and practice, including building a digital laboratory, curating a large clinical dataset, and developing early-stage AI models. We emphasize the invaluable role of institutional support, including financial and logistical resources, and dedicated and innovative computer and data scientists as equal partners in the research team. Our account highlights both facilitators and barriers encountered spanning financial support, learning curves inherent with interdisciplinary collaboration, and constraints of time and personnel. Through this narrative tutorial, we intend to demystify the process of AI research and equip clinicians with actionable steps to initiate new ventures in oncology research. As AI continues to reshape the research and practice landscapes, sharing insights from past successes and challenges will be essential to informing a clear path forward.
Unlabelled: Artificial intelligence (AI) and quantum computing will change the course of new drug discovery and approval. By generating computational data, predicting the efficacy of pharmaceuticals, and assessing their safety, AI and quantum computing can accelerate and optimize the process of identifying potential drug candidates. In this viewpoint, we demonstrate how computational models obtained from digital computers, AI, and quantum computing can reduce the number of laboratory and animal experiments; thus, computer-aided drug development can help to provide safe and effective combinations while minimizing the costs and time in drug development. To support this argument, 83 academic publications were reviewed, pharmaceutical manufacturers were interviewed, and AI was used to run computational data for determining the toxicity of collagen as a case example. The research evidence to date has mainly focused on the ability to create computational in silico data for comparison to actual laboratory data and the use of these data to discover or approve newly discovered drugs. In this context, "in silico" describes scientific studies performed using computer algorithms, simulations, or digital models to analyze biological, chemical, or physical processes without the need for laboratory (in vitro) or live (in vivo) experiments. Digital computers, AI, and quantum computing offer unique capabilities to tackle complex problems in drug discovery, which is a critical challenge in pharmaceutical research. Regulatory agents will need to adapt to these new technologies. Regulatory processes may become more streamlined, using adaptive clinical trials, accelerating pathways, and better integrating digital data to reduce the time and cost of bringing new drugs to market. Computational data methods could be used to reduce the cost and time involved in experimental drug discovery, allowing researchers to simulate biological interactions and screen large compound libraries more efficiently. Creating in silico data for drug discovery involves several stages, each using specific methods such as simulations, synthetic data generation, data augmentation, and tools to generate, collect, and affect human interaction to identify and develop new drugs.
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：