Dr. Vivian Lipari, executive director of the Ibero-American University Foundation (FUNIBER) in Chile, collaborated with an international team on a study published in The American Journal of Pathology that integrates machine learning and genomic analysis to improve the detection and classification of gastric cancer.
A Complex Disease
Gastric cancer is a major global health issue. The article ranks it among the most commonly diagnosed cancers and notes that it affects more than one million people each year. Early detection is essential for improving prognosis, although many cases are identified only when the disease is already advanced.
One of the main challenges stems from the molecular heterogeneity of gastric cancer. Genetic characteristics can vary among patients and tumor subtypes, making it difficult to apply uniform diagnostic and therapeutic strategies. Furthermore, conventional histopathological analyses, while still essential, do not always capture the full complexity of the genomic profiles associated with the disease.
Beyond Conventional Diagnosis
Among the solutions previously studied are machine learning models applied to genomic databases, gene expression signatures in blood samples, the identification of microRNAs, and prediction systems based on clinical or lifestyle factors.
These studies have shown promising results but also have limitations related to the use of retrospective data, small sample sizes, or insufficient external validation. Furthermore, some models focus on a single stage of the analysis and do not combine the discovery of molecular patterns with the predictive classification of cancer subtypes.
An Integrated Architecture
The study proposes a methodology that combines different artificial intelligence and data analysis tools to identify patterns in genetic information. The researchers grouped the profiles according to their characteristics and used these groups to develop models capable of recognizing similar patterns in new data.
In addition, they applied various methods to verify the consistency of the results, identify potential outliers, and facilitate the interpretation of the information. Finally, they analyzed which genetic characteristics had the greatest influence on the models’ predictions, with the aim of obtaining more understandable and reliable results.
High-Performance Results
The results demonstrated a high level of accuracy in the evaluated models. The support vector machine performed best, with 99.15% accuracy, followed by the stacking model at 98.87%, and Random Forest at 97.29%. Taken together, the results confirm the ability of these tools to classify genomic profiles accurately and consistently.
Further analysis of the results supported the robustness of the models and showed that their performance was significantly superior to what would be expected by chance. These findings underscore the potential of artificial intelligence to support the analysis and classification of genetic information.
Future Clinical Implications
The study concludes that the integration of bioinformatics, genomic analysis, and machine learning can help improve the classification of molecular subtypes and facilitate the search for potential biomarkers. The combination of unsupervised and supervised techniques also offers a pathway to developing more interpretable tools geared toward personalized medicine.
The results provide initial evidence of these tools’ potential to identify and differentiate genetic patterns, but they do not yet constitute a clinical tool for direct application. As next steps, the authors propose expanding the database, incorporating new genetic profiles, and exploring more advanced techniques to improve the accuracy and utility of these systems.
If you’d like to learn more about this study, click here.
To read more research, check out the UNEATLANTICO repository.
Through its Scholarship Program and University Network, FUNIBER facilitates access to undergraduate and graduate studies in various academic fields, including nutrition, public health, health management and research, and nursing. Fill out the form to request information about available scholarships and training options in these disciplines.