Supervisors info:
Ιωάννης Λουκάς, Αναπληρωτής Καθηγητής του Τομέα Φαρμακευτικής Χημείας, Τμήμα Φαρμακευτικής, ΕΚΠΑ
Summary:
Lung cancer is one of the most deadly diseases of the 21st century. It is divided into sub-categories depending on the pathophysiology that occurs. The most widespread type of lung cancer is non-small cell cancer, which is divided into additional categories, one of the most basic being adenocarcinoma. For the diagnosis of the disease there are currently many alternatives, while a method that gains ground is the lab testing with microarrays. The available treatment options vary depending on the molecular profile of each cancer.
In the present thesis, we analyzed the use of artificial intelligence and computational methods on results from Affymetrix microarray experiments, derived from the GEO and Array Express databases, to find specific predictor genes that can help predict clinical outcome of lung cancer (and in particular of adenocarcinoma) and could also be used to generate genetic signatures.
It was also trained, among many, a particular model-learner, who was able to classify patients with adenocarcinoma in patients with good or poor survival probability. The processing and filtering of the data, so that they can be analyzed more quickly and easily as well as the other calculations were done mainly by means of the programming language R and the statistical packets developed for it. The most important predictors predicted belong to groups of genes involved in lung cancer except for one who is likely to be a new biological marker.
Predictors can help generate genetic signatures, but also open up new horizons in searching for target molecules for treatment. Finally, the identified model adds one more tool in the search of the best treatment for an adenocarcinoma patient.