National and Kapodistrian University of Athens
Computational design for drugging protein membrane interfaces
Abstract
dc:descriptionAbnormal protein-membrane attachment is involved in deregulated cellular pathways and in disease. Therefore, the possibility to modulate protein-membrane interactions represents a new promising therapeutic strategy for peripheral membrane proteins that have been considered so far undruggable. Currently, the lack of a suitable theoretical background, experimental data, and simulation technology capable to implement this drug design strategy leads to high development costs of small molecules targeting this interface. The aim of this dissertation is to implement a novel drug design workflow for targeting protein-membrane interfaces by developing new algorithms used complementary to existing methods using the following steps: 1. Development of a fast and efficient algorithm predicting the protein-membrane interfaces of peripheral membrane proteins using machine learning. 2. Development of a drug design workflow for targeting the protein-membrane interfaces and implementing it in a web-server. 3. Identification of allosteric networks connecting membrane-binding domains using existing methodologies using PI3Kα and K-Ras4B as test cases. 4. Development of a new algorithm for the prediction of allosteric networks using deep learning. To design a new algorithm for the prediction of the protein-membrane interfaces using machine learning (step 1), first a dataset of peripheral membrane proteins with experimentally known membrane-penetrating amino acids was assembled. Physicochemical features related to the penetration of amino acids into the membrane were used to train 21 different classifiers and voting classifiers. Evaluation of the ensemble classifier accuracy produced a macro-averaged F1 score = 0.92 and an MCC = 0.84 for predicting correctly membrane-penetrating amino acids on unknown proteins of a validation set. Predictions were further verified in independent test sets. The python code is available at https://github.com/zoecournia/DREAMM. This algorithm was integrated into our workflow for determining binding sites in the predicted protein-membrane interfaces (step 2). In this workflow, small molecule binding sites in multiple protein structural conformations are identified and, only the sites located adjacent to the predicted membrane-penetrating amino acids are selected. The workflow was integrated into a web-server, which was onboarded in the European Open Science Cloud at https://dreamm.ni4os.eu/. In order for the identified binding sites to be functional, they need to communicate with the active site of the protein through allosteric networks. We thus applied existing allosteric network prediction methodologies for the study of two membrane oncoproteins PI3Kα and K-Ras4B as test cases (step 3). Finally, we aimed at creating a novel tool for predicting and evaluating the allosteric cross-linking of protein amino acids (step 4). To this end, we trained Graph Neural Networks (GNNs) in molecular dynamics simulations of 147 proteins. Structural and evolutionary-based characteristics were calculated, and GNNs were trained with the aim of predicting the allosteric cross-linking of amino acids. This tool is expected to predict allosteric networks in novel proteins without the need of molecular dynamics simulations.
Degree
thesis:*- Grantor dc:publisher
- National and Kapodistrian University of Athens
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Chatzigoulas, Alexios
- Χατζηγούλας, Αλέξιος
Subjects
dc:subject × 16- Μηχανική μάθηση
- Νευρωνικά δίκτυα γράφων
- Περιφερικές μεμβρανικές πρωτεΐνες
- Αλληλεπιδράσεις πρωτεΐνης-μεμβράνης
- Αλλοστερικός σχεδιασμός φαρμάκων
- Machine learning
- Graph neural networks
- Peripheral membrane proteins
- Protein-membrane interactions
- Allosteric drug design
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Τεχνητή νοημοσύνη
- Computer and Information Sciences
- Natural Sciences
- Artificial Intelligence
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/51412