Enhancing Diagnosis, Treatment, and Patient Results Using Artificial Intelligence in Healthcare

Advances in healthcare information systems and administration book series, Journal Year: 2025, Volume and Issue: unknown, P. 461 - 476

Published: Jan. 17, 2025

Artificial intelligence (AI) is revolutionizing healthcare by improving diagnostic accuracy, treatment personalization, and operational efficiency. This chapter explores AI applications in medical imaging, predictive analytics, drug discovery, patient care. We examine current AI-driven systems, such as deep learning models for early diagnosis natural language processing electronic health record (EHR) analysis. The also discusses challenges integrating into healthcare, including ethical considerations, data privacy, regulatory frameworks. Potential improvements outcomes workflows are highlighted, supported case studies research findings.

Language: Английский

---

A multi-task domain-adapted model to predict chemotherapy response from mutations in recurrently altered cancer genes

iScience, Journal Year: 2025, Volume and Issue: 28(3), P. 111992 - 111992

Published: Feb. 11, 2025

Next-generation sequencing (NGS) is increasingly utilized in oncological practice; however, only a minority of patients benefit from targeted therapy. Developing drug response prediction (DRP) models important for the "untargetable" majority. Prior DRP typically use whole-transcriptome and whole-exome data, which are clinically unavailable. We aim to develop model toward repurposing chemotherapy, requiring information clinical-grade NGS (cNGS) panels restricted gene sets. Data sparsity limited patient make this challenging. firstly show that existing DRPs perform equally with versus cNGS (∼300 genes) data. Drug IDentifier (DruID) then described, sets using transfer learning, variant annotations, domain-invariant representation multi-task learning. DruID outperformed state-of-the-art methods on pan-cancer data showed robust classification two real-world clinical datasets, representing step applicable tool.

Language: Английский

---

Effects of capecitabine chemotherapy on the expression of CASC18, CASC19, and CASC20 lncRNAs in patients with colorectal cancer and the HCT-116 colorectal cancer cell line

Pharmacia, Journal Year: 2025, Volume and Issue: 72, P. 1 - 13

Published: Feb. 27, 2025

Background : Colorectal cancer (CRC) is the third leading cause of cancer-related deaths. Capecitabine a key chemotherapy drug for colorectal (CRC). Long non-coding RNAs (lncRNAs) play role in pathogenic pathways associated with cancer. This investigation compared expression patterns CASC18, CASC19, and CASC20 CRC patients before after chemotherapy, as well healthy individuals. The effects capecitabine on cell viability, apoptosis, cycle, expressions were examined using HCT-116 cells. Methods assessed qRT-PCR patients’ tissues. Furthermore, receiver operating characteristic (ROC) curve analysis was used to evaluate prognostic diagnostic value CRC. MTT test assess capecitabine’s cytotoxicity line. Annexin-V/PI staining examine apoptosis cycle progression by flow cytometry. Gene differences between treatments qRT-PCR. Results showed that had lower levels CASC18 but higher CASC19 than group. Moreover, treated an increased expression, while non-significant changes observed CASC20. clinical factors not lncRNAsCASC18, results indicated are poor biomarkers diagnosis revealed dose- time-dependent cytotoxic effect treatment led accumulation cells Sub-G1 phase, suggesting apoptotic solid impact arrested G2/M phases capecitabine. also upregulated Conclusion These support anticancer highlight its ability increase expression. Additionally, findings indicate diagnosis.

Language: Английский

---

SensitiveCancerGPT: Leveraging Generative Large Language Model on Structured Omics Data to Optimize Drug Sensitivity Prediction

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 28, 2025

The fast accumulation of vast pharmacogenomics data cancer cell lines provide unprecedented opportunities for drug sensitivity prediction (DSP), a crucial prerequisite the advancement precision oncology. Recently, Generative Large Language Models (LLM) have demonstrated performance and generalization prowess across diverse tasks in field natural language processing (NLP). However, structured format poses challenge utility LLM DSP. Therefore, objective this study is multi-fold: to adapt prompt engineering toward optimizing LLM's DSP performance, evaluate real-world scenarios, compare against that state-of-the-science baselines. We systematically investigated capability Pre-trained Transformer (GPT) as model on four publicly available benchmark datasets, which are stratified by five tissue types encompass both oncology non-oncology drugs. Essentially, predictive landscape GPT assessed effectiveness task via learning paradigms: zero-shot learning, few-shot fine-tuning clustering pretrained embeddings. To facilitate seamlessly data, domain-specific novel employed implementing three templates (i.e., Instruction, Instruction-Prefix, Cloze) integrating pharmacogenomics-related features into prompt. validated GPT's scenarios: cross-tissue generalization, blind tests, analyses drug-pathway associations top sensitive/resistant lines. Furthermore, we conducted comparative evaluation multiple Transformer-based models existing Extensive experiments datasets cohorts demonstrate yields best (28% F1 increase, p-value= 0.0003) followed embeddings (26% 0.0005), outperforming in-context few-shot). setting had big gap, resulting worst performance. Within scope engineering, enhancement was achieved directly instructing about resorting concise context instruction-prefix), leading gain 22% (p-value=0.02); while incorporation drug-cell line derived from genomics and/or molecular further boosted score 2%. Compared baselines, significantly asserted superior mean (16% gain, p-value<0.05) GDSC dataset. In crosstissue analysis, showcased comparable generalizability within-tissue performances PRISM statistically significant improvements CCLE (8%, p-value=0.001) DrugComb (19%, p-value=0.009) datasets. Evaluation challenging tests suggests competitiveness compared random splitting. log probabilities provided valuable insights align with previous findings. experiment setups in-depth analysis underscore importance generative LLM, such GPT, viable silico approach guide https://github.com/bioIKEA/SensitiveCancerGPT.

Language: Английский

---

The specification game: rethinking the evaluation of drug response prediction for precision oncology

Journal of Cheminformatics, Journal Year: 2025, Volume and Issue: 17(1)

Published: March 14, 2025

Precision oncology plays a pivotal role in contemporary healthcare, aiming to optimize treatments for each patient based on their unique characteristics. This objective has spurred the emergence of various cancer cell line drug response datasets, driven by need facilitate pre-clinical studies exploring impact multi-omics data response. Despite proliferation machine learning models Drug Response Prediction (DRP), validation remains critical reliably assess usefulness discovery, precision and actual ability generalize over immense space cells chemical compounds. Scientific contribution In this paper we show that commonly used evaluation strategies DRP methods can be easily fooled occurring dataset biases, they are therefore not able truly measure drugs lines ("specification gaming"). problem hinders development reliable application experimental pipelines. Here propose new protocol composed three Aggregation Strategies (Global, Fixed-Drug, Fixed-Cell Line) integrating them with most train-test settings, ensure realistic assessment prediction performance. We also scrutinize challenges associated using IC50 as label, showing how its close correlation concentration ranges worsens risk misleading performance assessment, indicate an additional reason replace it Area Under Dose-Response Curve instead.

Language: Английский

---

Anticancer drug response prediction integrating multi-omics pathway-based difference features and multiple deep learning techniques

PLoS Computational Biology, Journal Year: 2025, Volume and Issue: 21(3), P. e1012905 - e1012905

Published: March 31, 2025

Individualized prediction of cancer drug sensitivity is vital importance in precision medicine. While numerous predictive methodologies for response have been proposed, the precise an individual patient’s to and a thorough understanding differences responses among individuals continue pose significant challenges. This study introduced deep learning model PASO, which integrated transformer encoder, multi-scale convolutional networks attention mechanisms predict cell lines anticancer drugs, based on omics data SMILES representations molecules. First, we use statistical methods compute gene expression, mutation, copy number variations between within outside biological pathways, utilized these pathway difference values as line features, combined with drugs’ chemical structure information inputs model. Then integrates various technologies encoder extract properties molecules from different perspectives, while network devoted complex interactions features aforementioned Finally, multilayer perceptron (MLP) outputs final predictions response. Our exhibits higher accuracy predicting drugs comparing other proposed recently. It found that PARP inhibitors, Topoisomerase I inhibitors were particularly sensitive SCLC when analyzing lung lines. Additionally, capable highlighting pathways related accurately capturing critical parts drug’s structure. We also validated model’s clinical utility using The Cancer Genome Atlas. In summary, PASO suggests potential robust support individualized treatment. are implemented Python freely available GitHub ( https://github.com/queryang/PASO ).

Language: Английский

---

Data imbalance in drug response prediction: multi-objective optimization approach in deep learning setting

Briefings in Bioinformatics, Journal Year: 2025, Volume and Issue: 26(2)

Published: March 1, 2025

Abstract Drug response prediction (DRP) methods tackle the complex task of associating effectiveness small molecules with specific genetic makeup patient. Anti-cancer DRP is a particularly challenging requiring costly experiments as underlying pathogenic mechanisms are broad and associated multiple genomic pathways. The scientific community has exerted significant efforts to generate public drug screening datasets, giving path various machine learning models that attempt reason over data space compounds biological characteristics tumors. However, depth still lacking compared application domains like computer vision or natural language processing domains, limiting current capabilities. To combat this issue improves generalizability models, we exploring strategies explicitly address imbalance in datasets. We reframe problem multi-objective optimization across drugs maximize deep model performance. implement approach by constructing Multi-Objective Optimization Regularized Loss Entropy loss function plugging it into Deep Learning model. demonstrate utility proposed discovery make suggestions for further potential work achieve desirable outcomes healthcare field.

Language: Английский

---

DTLCDR: A target-based multimodal fusion deep learning framework for cancer drug response prediction

Journal of Pharmaceutical Analysis, Journal Year: 2025, Volume and Issue: unknown, P. 101315 - 101315

Published: April 1, 2025

Language: Английский

---

AI in Cancer Research: Challenges, Applications, and Future Directions

Lecture notes in networks and systems, Journal Year: 2025, Volume and Issue: unknown, P. 201 - 216

Published: Jan. 1, 2025

Language: Английский

---

A Comprehensive Investigation of Active Learning Strategies for Conducting Anti-Cancer Drug Screening

Cancers, Journal Year: 2024, Volume and Issue: 16(3), P. 530 - 530

Published: Jan. 26, 2024

It is well-known that cancers of the same histology type can respond differently to a treatment. Thus, computational drug response prediction paramount importance for both preclinical screening studies and clinical treatment design. To build models, data need be generated through experiments used as input train models. In this study, we investigate various active learning strategies selecting generate purposes (1) improving performance models built on (2) identifying effective treatments. Here, focus constructing drug-specific cancer cell lines. Various approaches have been designed applied select lines screening, including random, greedy, uncertainty, diversity, combination greedy sampling-based hybrid, iteration-based hybrid approach. All these are evaluated compared using two criteria: number identified hits selected validated responsive, model trained experiments. The analysis was conducted 57 drugs results show significant improvement with random sampling method. Active also an some runs

Language: Английский

---