Analisi dei dati e controllo di qualità in uno studio di spettrometria di massa applicata alla proteomica

Despite all the scientific progress we have been through in the last few decades, cancer is still the second leading cause of death worldwide. According to the World Health Organization 1(WHO), from a total of 58 million deaths worldwide in 2005, cancer accounts for 7.6 million (or 13%) of all deaths.
If diagnosed in time, there is more of a chance to fight it. Unfortunately, the majority of deaths are due to late detection and lack of diagnostic methods.
The main purpose of this work is to study a new, accurate, fast and easy-to-use method of early diagnosis based on a proteomic approach.
Since cancer cells need blood to feed on and grow, they will inevitably release their metabolites and alter the blood content.
Ordinarily, we would look for a single biomarker (i.e. a protein, a hormone) as it is for the prostatic cancer (PSA, Prostate-Specific-Antigen). However this does not always lead to an accurate diagnosis. In fact, further investigation is often needed.
Why not take advantage of a much bigger source of information such as the Proteome?
The Proteome is the set of the many hundreds of thousands of proteins expressed in an organism or cell type at a given time.
To put it in a nutshell, instead of looking at one biomarker at a time, we can take an information archive, such as the serum proteome, and look at tens of thousands of proteins and peptides at once – a proteomic portrait, a barcode – that we generate
very rapidly from the mass spectrometer, more specifically using MALDI-TOF Mass Spectrometry.
By looking at the spectra, one cannot see any underlying hidden patterns – instead, they would look like chaotic pictures. As we will see, there is no such good statistics to find the “needle in the haystack”.
Therefore, we've taken a machine-based approach, an artificial intelligence-based algorithm, to look at hundreds of millions of combinations of protein patterns very quickly and find the optimal pattern that segregates cancerous samples from noncancerous samples in the training sets. This artificial-intelligence procedure is called Random Forests.
In this work, which has been conducted in a tight collaboration between the Department of Organic Chemistry “A. Mangini” and the Department of Gastroenterology and Internal Medicine (University of Bologna), we aim at understanding how the presence of a cancer can affect the protein pattern of an individual, in order to use it as a diagnostic tool.
The entire research has taken over a year and my contribution has been limited to devising an algorithm for the spectra preprocessing.