Neural ordinary differential equations in pharmacokinetics for long-term Dalbavancin dose predictions

Giacometti, Tommaso (2024) Neural ordinary differential equations in pharmacokinetics for long-term Dalbavancin dose predictions. [Laurea magistrale], Università di Bologna, Corso di Studio in Physics [LM-DM270], Documento ad accesso riservato.
Documenti full-text disponibili:
[thumbnail of Thesis] Documento PDF (Thesis)
Full-text non accessibile fino al 1 Maggio 2025.
Disponibile con Licenza: Creative Commons: Attribuzione - Non commerciale - Condividi allo stesso modo 4.0 (CC BY-NC-SA 4.0)

Download (10MB) | Contatta l'autore

Abstract

This thesis explores the use of Machine Learning, specifically Neural Ordinary Differential Equations (NODEs), in Pharmacokinetics to address the limitations of traditional methods. The focus is on predicting drug concentration for Dalbavancin, a long-acting antibiotic, in the human body. Traditional pharmacokinetic models, such as compartment models, struggle to capture complex drug dynamics and often lack personalization due to simplifying assumptions. In contrast, NODEs provide a more flexible and adaptive approach, as they relax many of these assumptions, allowing for more accurate and tailored predictions. The study used a small and sparse clinical dataset of Dalbavancin, involving multiple administrations, which presents significant challenges for machine learning models. To overcome these, data augmentation techniques were applied to enhance the stability and training of NODE models. Results showed that NODEs effectively model drug concentration dynamics, aligning well with real clinical data. They particularly excelled in long-term prediction accuracy, where traditional models typically under predict drug levels over extended periods. Additionally, the integration of SHAP (Shapley Additive Explanations) analysis provided valuable insights into how patient-specific factors influence drug concentrations, opening the door to more personalized dosing strategies. However, the overall prediction accuracy of NODEs did not show significant improvement, likely due to the limited dataset size. Furthermore, NODE models require substantial computational resources and fine-tuning. Despite these challenges, the study demonstrates the potential of NODEs as a powerful tool for pharmacokinetic modelling and personalized medicine, especially when dealing with small, sparse clinical datasets. These findings suggest that NODEs could enhance drug therapies by improving treatment personalization and long-term prediction capabilities.

Abstract
Tipologia del documento
Tesi di laurea (Laurea magistrale)
Autore della tesi
Giacometti, Tommaso
Relatore della tesi
Correlatore della tesi
Scuola
Corso di studio
Indirizzo
Applied Physics
Ordinamento Cds
DM270
Parole chiave
Deep-learning,NODE,Pharmacokinetic,Neural-networks,SHAP,AI-interpretability,Differential equations,Precision medicine
Data di discussione della Tesi
30 Ottobre 2024
URI

Altri metadati

Gestione del documento: Visualizza il documento

^