CBL - Campus del Baix Llobregat

Projecte llegit

Títol: Predicció del model lipoproteic en models animals

Estudiant que ha llegit aquest projecte:

Tutor/Cotutor: PRATS SOLER, CLARA

Departament: FIS

Títol: Predicció del model lipoproteic en models animals

Data inici oferta: 21-02-2022      Data finalització oferta: 21-10-2022

Estudis d'assignació del projecte:
    GR ENG SIS BIOLÒGICS

Lloc de realització:


En empresa (cal signar un conveni de cooperació)

        Tutor Extern: ENRIQUE OZCARIZ GARCÍA
        Institució/Empresa: BIOSFER TESLAB SL

Segon tutor extern: Eduardo Dominguez Sala

Paraules clau:
Models animals, Model lipoproteic; Models estadístics

Descripció del contingut i pla d'activitats:
Obtención de resultados metabolómicos a partir de espectros de H-
RMN.
Creación, manejo y análisis de bases de datos.
Estadística univariante y análisis de regresión lineal (STOCSY).
Creación de modelos supervisados de regresión (partial least
squares) para la predicción de variables a partir de espectros de
RMN.

Overview (resum en anglès): Nowadays cardiovascular diseases have a high influence on the world's population and this trend is expected to increase over the years to come. Therefore, there is a general interest to develop new methods that enable professionals to advise reliably depending on the patient's cardiovascular profile. Thanks to metabolomics, new techniques have been achieved such as the nuclear magnetic resonance (NMR) which has been applied to predict standar lipids highly related with cardiovascular diseases. The present project evaluates and develops PLS (Partial Least Square) prediction models based on nuclear magnetic resonance of protons with the aim to achieve a quantification of the total cholesterol and triglycerides of three sets of animal plasma samples (pigs, N = 230 where N is the number of samples). First basic descriptive statistics were performed in order to better understand the behavior of the samples. Afterwards, a PCA (Principal Components Analysis) was performed showing how the samples spread according to different variables. Then, it was decided to exclude the samples belonging to the set number 3 since they were not comparable to the samples of the other two sets. Next STOCSY (Statistical TOtal Correlation SpectroscopY) plots were created to observe the correlation between the NMR spectra and the studied prediction variable, lastly, from all the obtained data, the PLS predictions models were created. For this purpose, two methods were used, the first one used 70% of the values as a training set and the remaining 30% as a testing set, the second one consisted on selecting the 10% of the highest values, the 10% of the lowest values, 50% of the values which were randomly selected as a training set and the remaining 30% were used as a testing set. Two types of preprocessing were applied to the data in order to identify which was the one returning the best results, when applied, the Autoscale preprocessing got promising results achieving models with a 0.97 (R=0,97) of goodness of fit to cholesterol. Regarding triglycerides, better results were given when the Mean centering preprocessing was applied giving a maximum fitness of 0.86 (R=0.86). After the completion of this project, the creation of PLS prediction models to quantify lipoproteins in pig plasma samples reliably is clearly possible.

© CBLTIC Campus del Baix Llobregat - UPC