diff options
Diffstat (limited to 'data_understanding/main.tex')
| -rw-r--r-- | data_understanding/main.tex | 100 |
1 files changed, 85 insertions, 15 deletions
diff --git a/data_understanding/main.tex b/data_understanding/main.tex index 9879cb5..5e30de2 100644 --- a/data_understanding/main.tex +++ b/data_understanding/main.tex @@ -15,7 +15,7 @@ \section{Initial data collection} -\subsection{Technicalities} +\subsection{Technical description data collection} \subsubsection{MySQL database set up and data import} @@ -55,20 +55,90 @@ using \lstinline{php bin/import.php}. \subsection{Data Requirements} -The following subsections will list the attributes required from the data per -data mining goal. - -\subsubsection{Explore and describe SQL queries and corresponding csv tables.} - -\subsubsection{Model and visualise the different clinical study populations.} - -\subsubsection{Model and visualise the difference between vaccination types.} - -\subsubsection{Model and visualise repeat vaccination effects.} - -\subsubsection{Apply standard feature selection methods to the most interesting dataset.} - -\subsubsection{Fit classification models to the most interesting dataset.} +The following subsections will list the information required from the data per +data mining goals that are needed to answer the following business questions: + +\begin{itemize} + \item Which datasets in the FluPrint database are most interesting? + \item How do different clinical studies compare? + \item What are the differences in efficacy between vaccination types? + \item What is the effect of repeat vaccination on vaccine response? + \item What immunological factors correlate to a high vaccine response? +\end{itemize} + +\subsubsection{Requirements per data mining goal} + +\begin{displayquote} +"Explore and describe SQL queries and corresponding csv tables." +\end{displayquote} + +Falling under this data mining objective are the outputs and tasks related to +data collection and description. These comprise a report on the initial +collection of the data, selection of data, and description of general +properties of the data. The data in this case is in a database format, thus +here we describe the tables, keys, and attributes in the database, and also +include descriptive statistics about the data. The goal is to replicate the +description done in \cite{tomicFluPRINTDatasetMultidimensional2019} as well. +Using these descriptions we provide insight into which datasets in the database +are most interesting, and why in \cite{tomicSIMONAutomatedMachine2019} one +dataset in particular was chosen. + +\begin{displayquote} +"Model and visualise the different clinical study populations." +\end{displayquote} +\begin{displayquote} +"Model and visualise the difference between vaccination types." +\end{displayquote} +\begin{displayquote} +"Model and visualise repeat vaccination effects." +\end{displayquote} + +In order to answer the business question "How do clinical studies compare?" +subpopulations and groups of attributes need to be visualised and compared +across different clinical studies. The data required must have rows +corresponding to donors in a particular clinical study and columns that are +attributes of tables in the database, these could be biological assay results +or information about the donors. Thus we aimed to export one csv from the +database per clinical study by querying for different clinical studies. + +We aimed to generate csv files of donors corresponding to received vaccine +types to answer the business question "What are the differences in efficacy +between vaccination types?". One simple method to indicate the difference +between vaccines would be to report the proportion of high-reponders across all +donors, or to use a simple model to find the best predictor for a high +response. These comparisons require one table per vaccine type, with rows +corresponding to donors and columns that include the vaccine response +classification, in addition to other immune assay and donor attributes. + +The objective in question "What is the effect of repeat vaccination on vaccine +response?" requires data from long running clinical studies. One dataset that +is used by the database authors and was investigated to answer this question +was already available, here we aimed to describe and visualise any patterns we +could find in this dataset and other long running clinical study datasets. This +required data from a subset of clinical studies that spanned multiple years, at +this point in the project the data for these clinical studies should have been +available, and we just had to choose those that spanned multiple years. + +\begin{displayquote} +"Apply standard feature selection methods to the most interesting dataset." +\end{displayquote} + +\begin{displayquote} +"Fit classification models to the most interesting dataset." +\end{displayquote} + +These last two data mining objectives were chosen to comprise the data +preparation and modelling phases of this project. The authors of fluprint set +up an automated machine learning pipeline to investigate the longest running +clinical study dataset in the database. In this work we use a conventional data +mining modelling process to replicate these results. This dataset contains any +immunological assay results for all donors in the clinical study on their first +vist, and their classification as a high or low vaccine responder. To fulfill +the above two data mining goals, we used this dataset. + +\section{Data description} + +\subsection{Volumetric analysis} \citep{chattopadhyaySinglecellTechnologiesMonitoring2014} The complex heterogeneity of cells, and their interconnectedness with each |