Carlos Jesús García Hernández
Nanoscience, Materials and Chemical Engineering
Alberto Fernández Sabater & Francesc Serratosa Casanelles
Carlos García-Hernández has studied Electronic Engineering degree in UNET University in Venezuela. As a final degree project, he worked on a self-driving robot moving through a labyrinth, learning all the different paths and finding the shortest way out. He also worked as a teacher assistant in programming at UNET for several semesters. After the engineering degree, he worked in different private companies in the field of industrial automation, database analysis, robotics, and mobile development. This work period helped him to improve his programming skills and originated a particular interest in computer science and machine learning. Moreover, he studied a Master degree in Computer Security Engineering and Artificial Intelligence at the URV. His master thesis was about "Active and Interactive Graph Matching in Computer Vision". This work helped him to continuous his research as a PhD student in the same university in the PhD programme of Nanoscience, Materials and Chemical Engineering. As a researcher, he is trying to follow his passion for science by participating in different conferences, workshops, seminars and summer schools in Europe. Additionally he works as a teacher assistant in mathematics and statistics in the same university.
Project: Molecular reactivity prediction based on structure reduction and graph edit distance
Extended reduced graphs provide summary representations of chemical structures using pharmacophore-type node descriptions to encode the relevant molecular properties. The effectiveness of extended reduced graphs along with graph edit distance methods for molecular similarity searching is investigated as a tool for ligand-based virtual screening applications, where an estimation of reactivity for a chemical is provided using experimental data available from highly similar compounds. Results obtained for this method show to be very stable and having a good performance as compared to the original low dimensional descriptor vector (2D fingerprint) used in previous works on extended reduced graphs. This is exemplified for 4 different activity classes (increasing the difficulty level) available in the ligand-based virtual screening benchmarking platform, for which our method performs better than other measuring methods using fingerprints. Overall, it is shown that extended reduced graphs along with graph edit distance is a widely applicable combination of methods, capable of identifying structurally diverse actives for a given active search query.