Analysis-of-Pharmacological-Data-to-Predict-the-Activity-of-Enyzme-Compounds
This is my diploma thesis with the title "Analysis of Pharmacological Data for the Prediction of the Effectiveness of Chemical Compounds with Cardio-Protective Activity" under the supervision of Dr. George Manis.
Problem Definition : Non-steroidal anti-inflammatory drugs are the most useful and commonly prescribed treatment for inflammatory conditions. Nevertheless, their wide use has been linked to many adverse effects, either minor (e.g., gastrointestinal irritation) or major (e.g., increased risk for cardiovascular disease). An interesting approach to overcome these risks includes the design of antioxidants compounds able to scavenge DPPH and lipoxygenase (LOX) inhibitors in parallel. The implication of reactive oxygen species in inflammatory conditions is well proven, mainly via the pro-inflammatory properties of the generated superoxide anion. The latter is involved in the deterioration of a plethora of inflammatory conditions. The problem we are called to face, using machine learning as well as data analysis is, the creation of capable machine learning models to discriminate pharmacological selectivity and discover compounds with dual activity prior to synthesis. Knowledge from already in vitro tested compounds can establish an accurate filter algorithm to separate compounds with high, moderate or low affinity. We constructed a classification protocol to check whether the compounds designed could be LOX inhibitors and/or DPPH scavengers as an indication of in vitro anti-inflammatory and/or antioxidant activity, respectively. Drugs with dual activity could assist in the proper functioning of the heart and specifically in its proper perfusion. Some produced compounds of LOX enzyme oxidize the fatty acids that one needs to help with an inflammation, with the compounds belonging to the category of anti-inflammatory drugs. Additionally, when an organism is stressed it produces some free radicals therefore there are some compounds that block this radicals preventing the organism from destroying itself. This compounds belong to the category of antioxidant drugs and we tried to categorize correctly the new compounds that have a good effect on their pharmacological target. Eventually, we wanted to observe the way in which these newly synthesized compounds are separated from each other, regarding a specific pharmacological goal, using machine learning models for classification as well as regression. Finally, the properties of each compound are being described by some molecular descriptors. This descriptors are the features of our databases, consequently we had to single out and pick the best out of them for better model training as well as using them in future studies.
Contribution : The contribution of this diploma thesis is that a systematic study and research was conducted in order to design, implement and evaluate different types of machine learning models with application to pharmaceutical data. According to the current literature, we innovated with the research and analysis conducted on the lipoxygenase (LOX) enzyme with the use of computer science. We successfully created models capable of categorizing correctly the type of activity for new synthesized compounds as well as predict the exact value of effectiveness of different proteins. Furthermore we managed to extract valuable features from the data sets in order to create new and more reliable models as well as use this features for future studies. Finally we confirmed given the results we have produced, that computer science and particularly the branch of machine learning, can play a key role in the field of health care as well as in pharmacology.