Projektportal

Startsida Om denna webbplats

Logga in

In silico protocols to predict stability of amorphous solid dispersions

Terrmin

HT24

Beskrivning

Aim
The objective of this project is to apply machine learning (ML) tools to the available molecular dynamics (MD) dataset to engineer features predicting the stability of the amorphous solid dispersions (ASDs), as observed experimentally.
Background
Amorphous solid dispersions is an innovative strategy to formulate drugs to increase solubility, and as a result, bioavailability. Drug molecules, in amorphous form would be entrapped in a matrix of excipient molecules, thus stabilized in high energy state. MD simulations of six different drug molecules were performed at various drug loadings and a significant body of data was accumulated from the trajectories of the molecules.1,2 Some of the criteria for formulation stability prediction have already been developed manually (for example, descriptors for hydrogen bond networks), but more accuracy in predictions can be achieved with help of various ML models.
Methods
Molecular dynamics simulations analysis, via Gromacs3 built-in commands, python MD libraries (MDAnalysis4,5 and MDTraj6) and in-house written scripts.
Machine learning libraries: scikit-learn7, Optuna8, TensorFlow9, RDKit.
Various models will be built, ranging from classical approaches (such as random forest) and to shallow deep learning networks.
My task
- Perform a thorough literature review on ML applied to MD data and regarding stabilization of formulations specifically.
- To engineer features using the data available from MD simulations and experiments done previously.
- Systematically build a broad range of diverse machine learning models to reach the optimal prediction of stability. Finetune the models with hyperparameters.

Expected results:
- Ranked models for prediction of ASD stability with MD data as input.
- Insights into the factors influencing the stability the most (including advanced engineered features, such as combinations of descriptors).
- High-quality literature review on applicability of ML tools to MD data.

Huvudområde

Läkemedelsutveckling

Ämne

Läkemedelsformulering och Molekylär galenisk farmaci

Typ

Beräkningsstudie

Företag

Uppsala University

Ort/Plats

Uppsala

Handledarens namn

Aleksei Kabedev

Handledarens e-post

aleksei.kabedev@farmaci.uu.se

Institution

Institutionen för farmaci

Program

Masterprogram i läkemedelsmodellering

Kurs

Degree Project in Pharmaceutical Modelling within Pharmaceutics and Biopharmaceutics 45 c - 3FG001

Omfattning/hp

45hp