|
|
Interactive search in image datasets using CLIP neural network
Vopálková, Zuzana ; Lokoč, Jakub (advisor) ; Hoksza, David (referee)
With the growing importance and volume of multimedia data, interactive search sys- tems are essential to help users efficiently search for specific video sequences based on content. One common task is known scene retrieval, where users try to find a particular scene in a large collection of videos. However, the description of a known scene can be subjective, influenced by the perception and experience of individual users, as well as the differences between human and machine perception. In this paper, the effectiveness of an interactive retrieval system combined with image classification generated by a CLIP neural network is investigated to address this problem. V3C datasets and Marine Video Kit are used to verify the effectiveness of the proposed method. Software is also presen- ted that allows data collection for experiments and subsequent evaluation using a web interface. 1
|
|
|
Enzyme optimization using sequence homology and machine learning
Neitzert, Gesa-Maret ; Příhoda, David (advisor) ; Hoksza, David (referee)
In pharmaceutical research and development, enzymes play an important role in the synthesis of drugs and drug-related molecules. For higher efficiency and increased production, it is important to optimize the yield of these enzymes, a task often addressed by protein engineering and design. This process of enzyme optimization however can become tedious with the vast options of mutations for each single protein. To improve the process of enzyme optimization, sequence homology and machine learning methods can be used. These greatly reduce the manual effort of protein redesign and can assist in finding the most fit enzyme for the given task, increasing the efficiency of the overall drug development pipeline. With this aim in mind this thesis summarizes a selection of existing methods and their possible application to enzyme optimization. Testing two predictive models with varying complexity on 4 datasets in an attempt to optimize absorption, enantioselectivity, or thermostability found only a modest correlation between actual target values and their predicted values: mean Pearson's R 0.20775 and 0.5188. Comparing probability patterns of protein sequence embeddings led to a 0.815 correlation score with the BLOSUM62 substitution matrix, confirming the language model's intuition about natural frequency of...
|
|
|
Detection of protein-ligand binding sites using graph neural networks
Gamouh, Hamza ; Hoksza, David (advisor) ; Pilát, Martin (referee)
The function of most biological systems is realized by the interaction of proteins with other biological molecules. Protein-ligand binding is one of the most im- portant kind of interactions which, if studied well, can reveal a lot of hidden causal patterns behind biological functions, and can contribute to the rapid de- velopment of drug research. Protein-ligand binding sites detection is one aspect of the general study of protein-ligand interactions, where the goal is to develop computational methods that can use ligand as well as protein structure, possibly together with its sequence data, to predict regions of the protein that can bind to potential ligands. The significant growth of protein structure databases, such as, Protein Data Bank (PDB) and sequence databases like Universal Protein Re- source (UniProt), has led to the development of different machine learning and deep learning approaches that make use of this huge amount of biological data to solve this task. In this thesis, we examine a deep learning method based on a recent model architecture called Graph Convolutional Networks (GCN), which combines the traditional Convolutional Neural Network (CNN) architecture, and the more recent Graph Neural Network (GNN) architecture which has been suc- cessful in solving various chemoinformatics...
|
|
|
Framework for retrieval and analysis of proteins apo and holo forms from PDB
Král, Adam ; Hoksza, David (advisor) ; Sanchez Rocha, Alma Carolina (referee)
We developed a software framework that allows the analysis of ligand-free (apo) and ligand-bound (holo) forms of proteins that are accessible in PDB. The software downloads the current version of the PDB, divides the structures into groups of the same molecules, and these into apo and holo forms. Finally, it is possible to analyze pairs of apo and holo structures with respect to their different structural characteristics. In addition to the software work itself, we also verify results against previous work on an equivalent dataset, and obtain results for the current version of PDB. Keywords: protein; structural bioinformatics; PDB
|
|
|
Framework for retrieval and analysis of proteins apo and holo forms from PDB
Král, Adam ; Hoksza, David (advisor) ; Novotný, Marian (referee)
We developed a software framework that allows the analysis of ligand-free (apo) and ligand-bound (holo) forms of proteins that are accessible in PDB. The software downloads the current version of the PDB, divides the structures into groups of the same molecules, and these into apo and holo forms. Finally, it is possible to analyze pairs of apo and holo structures with respect to their different structural characteristics. In addition to the software work itself, we present the results of selected analyses of the current version of the data in the PDB. We also verify the results against previous work.
|
|
|
Virtual laboratory for simulation of biological processes on membranes
Beneš, Martin ; Bílý, Tomáš (advisor) ; Hoksza, David (referee)
The purpose of this thesis is the creation of a virtual computer laboratory, where it will be possible to simulate and analyze the behavior on the biological cell membrane. The membrane and its surroundings are described as a hex-net with other chemicals and components that interact with each other. Depending on this distribution, it is possible to observe well-known membrane phenomena such as the amplification of the action potential. The simulation itself should show, that for the basic membrane-activity modeling, you only need a few basic laws of physics and chemistry.
|
|
|
Requirement management tool
Jelen, Martin ; Kopecký, Michal (advisor) ; Hoksza, David (referee)
Managing requirements for software projects is a key task throughout the whole lifetime of the project. Recognizing this, it is the goal of the bachelor thesis to design and implement a tool that simplifies and automates requirements management and provides a clear view of the current state of managed projects as well as their history. It is not intented as a single project management solution but rather to be used with other specialized software. Implemented is the ability to capture the managed project's structure and multiple types of requirements (e.g. functional, performance, usability) with clearly defined relationships. The project's representation can further evolve, all previous versions are stored and changes can be reviewed. Data presentation is accomplished with multiple views of requirements and allows to trace changes through the relationships. Finally, the gathered data may be exported in either structured form for further transformation or as user-readable reports. Requirements, their atributes and types, as well as views on the requirements are highly configurable to provide support to projects independent of development methodology used - from agile approaches with minimum administrative documents to more formalized processes such as RUP.
|
|
|
MS SQL Application Development Framework
Hanes, Marek ; Kopecký, Michal (advisor) ; Hoksza, David (referee)
Title: MS SQL Application Development Framework Author: Bc. Marek Hanes Department: Department of Software Engineering Supervisor: RNDr. Michal Kopecký, Ph.D. Supervisor's e-mail address: kopecky@ksi.mff.cuni.cz Abstract: The thesis deals with a database application development and tries to find ways to optimize the most common problems encountered. The goal of this thesis is to design and develop a modular framework that simplifies the database application development and prevents inexperienced users from using unsafe SQL statements and/or expressions. The example of such a statement can be the insert statement without explicit column list, unsafe XPath expression, etc. The framework provides among others manipulation with history tables allowing versioning of data and reverting unwanted data changes asynchronous and parallel SQL execution support, error management and logging support, monitoring of schema changes as well as procedure and function debugging Together with means of data manipulation, the framework provides the simple way of publishing stored procedure as web service as well. The framework is accompanied by well-written programmers and users guide to allow its further development. Keywords: application development, framework, modular design, safe statements, constraints 1
|
|
|
Data Generator for Microsoft SQL Server
Drozen, Jan ; Hoksza, David (advisor) ; Kopecký, Michal (referee)
Testing of newly programmed components is an integral part of software development. Data Generator for Microsoft SQL Server is a tool for simple populating database tables with testing data which allows efficient database component testing for programmers, administrators and other database developers. Generated data can be structured using the parameters which are set in a simple user interface. Generated data correspond with database schema and constraints defined on columns. In following text we are going to describe the architecture and implementation details together with the necessary theoretical knowledge.
|
|
|
Modification of Pivot Tables method for persistent metric indexing
Moško, Juraj ; Skopal, Tomáš (advisor) ; Hoksza, David (referee)
The pivot tables is one of the most effective metric access method optimized for a number of distance computations in similarity search. In this work the new modification of the pivot tables method was proposed that is besides distance computations optimized also for a number of I/O operations. Proposed Clustered pivot tables method is indexing clusters of similar objects that were created by another metric access method - the M-tree. The indexing of clustered objects has a positive effect for searching within indexed database. Whereas the clusters are paged in second memory, page containing such cluster, which do not satisfy particular query, is not accessed in second memory at all. Non-relevant objects, that are out of the query range, are not loaded into memory, what has the effect of decreasing number of I/O operations and total volume of transferred data. The correctness of proposed approach was experimentally proved and experimental results of proposed method was compared to selected metric access methods.
|
guest ::
