Maestría en Biociencias (tesis)

URI permanente para esta colección

https://hdl.handle.net/10784/30775

Examinar

Mostrando 1 - 3 de 3

Detection of carbapenem resistance in Klebsiella pneumoniae using convolutional vision transformers and MALDI-TOF proteomic profiles.
(Universidad EAFIT, 2025) Salazar Marín, Valentina; Fernández García, Geysson Javier; Bravo Ortíz, Mario Alejandro
Antimicrobial resistance is a growing global health problem, significantly increasing morbidity, mortality and healthcare costs. Traditionally, the identification of antibiotic resistance is based on phenotypic methods such as agar diffusion or automated systems such as VITEK, which require 24-72 hours to yield definitive results, delaying appropriate patient management. In this context, the need for faster and more accurate diagnostic aid strategies arises. This study explores the integration of artificial intelligence (AI) and mass spectrometry techniques for the classification of carbapenem-resistant Klebsiella pneumoniae strains. Specifically, matrix-assisted laser desorption/ionization-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) data were used to generate proteomic profiles potentially associated with resistance mechanisms. However, the complexity and high volume of these data make the use of AI tools capable of identifying robust patterns indispensable. A convolutional vision transformer (CVT) model was implemented to classify carbapenem-resistant Klebsiella pneumoniae strains from a set of 180 proteomic spectra collected by Synlab Colombia. The CVT model outperformed traditional convolutional neural networks and other automated learning approaches, achieving higher accuracy and stability. Grad-CAM visualization improved model interpretability by identifying key spectral regions associated with resistance. The results highlight the potential of Vision Transformers in microbiological diagnostics by significantly reducing resistance detection time and contributing to a timelier clinical response. Future studies should explore the applicability of this methodology on other resistant pathogens to improve global surveillance efforts against antimicrobial resistance.
Microbiología predictiva mediante aprendizaje automatizado para la optimización de procesos productivos : Metanálisis
(Universidad EAFIT, 2024) Yepes Medina, Verónica; Pinel Peláez, Nicolás; Ninguno
Unsafe food containing harmful bacteria, viruses, parasites or chemicals can cause more than 200 different illnesses, from diarrhea to cancer. Worldwide, an estimated 600 million (nearly 1 in 10 people) fall ill each year after eating contaminated food, resulting in 420.000 deaths and the loss of 33 million years of healthy life. Therefore, it is necessary to detect and respond to public health threats associated with unsafe food with enabling technologies or tools. Predictive microbiology is concerned with preventing, controlling or limiting the existence of microorganisms by mapping their potential responses to particular environmental conditions, such as temperature, pH, nutrients (protein and fat), water activity (aw) and others. And machine learning as a branch or artificial intelligence learns from these data, identifying patterns for decision making. Recent studies are based in the use of supervised machine learning models to predict the presence of a foodborne pathogenic microorganism at any stage of the production chain, the most commonly used models include Random Forest and support vector machine with rating metrics for accuracy and sensitivity >80%. The main evaluation metrics of the algorithms are: accuracy, F1 score, confusion matrix, sensitivity, specificity and area under the curve (ROC-AUC, Receiver-Operating-Characteristic). Studies have shown that Random Forest was the best model, exhibiting an accuracy of 95% and a F1 score of 98%. Here were evaluated twenty five (25) articles with library metafor of Rstudio version 4.2.1 and this information provides new opportunities to explore non-destructive models for rapid detection of microorganisms in the production chain.
The random forest machine learning model performs better in predicting drug repositioning using networks : systematic review and meta-analysis
(Universidad EAFIT, 2024) García Marín, Darlyn Juranny; García Zea, Jerson Alexander; García Zea, Jerson Alexander

Examinar

Examinando Maestría en Biociencias (tesis) por Materia "Aprendizaje automático"

Resultados por página

Opciones de ordenación