Luca Di Grazia

Course: Computer Engineering
Date of thesis discussion: 24/07/2019

A new protein characterization and classification method using 3D face recognition algorithms

The objective of this research is to translate biometric 3D face recognition algorithms into the field of protein biophysics with the aim of precise and fast classification of morphological features of proteins. Faces and proteins can be seen like free forms and some geometrical descriptors can be used.
The feature extraction relies on geometrical descriptors coming from Differential Geometry. The first part of this study is a pilot computational experiment to find a classification methodology that is able to recognise face micro-expressions and cluster them into emotional states. The second part is about classification of Tubulin isotypes and Tubulin vs FtSZ proteins. Different classification methodologies are used: classic approach with a Support Vector Machine classifier, unsupervised learning with a K-means algorithm and a Deep Learning method with a simple Neural Network. The results are significant and competitive with the state-of-the-art protein classification.