|
|
Strojové učení v úloze predikce vlivu aminokyselinových mutací na stabilitu proteinu
Malinka, František ; Martínek, Tomáš (oponent) ; Bendl, Jaroslav (vedoucí práce)
Tato práce popisuje nový přístup k predikci vlivu aminokyselinových mutací na změnu stability proteinu. Cílem je vytvořit nový meta-nástroj, který kombinuje výstupy osmi vybraných nástrojů, díky čemuž je schopen svoji predikční schopnost zlepšit. Pro nalezení optimálního konsenzu mezi těmito nástroji je použito různých metod strojového učení. Ze všech testovaných metod strojového učení dosahuje KStar nejvyšší úspěšnosti predikce na trénovacím datasetu tvořeného experimentálně ověřenými mutacemi z databáze ProTherm. Právě z tohoto důvodu je KStar vybrán jako optimální predikční technika. Pro prokázání korektnosti výsledků tohoto meta-nástroje je použito testovacího datasetu vytvořeného ojedinělým způsobem, a to z vícebodových mutací extrahovaných taktéž z databáze ProTherm. Jelikož nebyly vícebodové mutace použity pro natrénování žádného z integrovaných nástrojů, předpokládá se, že takovéto porovnání je objektivní. Ve výsledku se tímto přístupem podařilo pomocí metody strojového učení KStar zvýšit korelační koeficient na trénovacím datasetu o 0,130, respektive o 0,239 na datasetu testovacím oproti nejúspěšnějšímu integrovanému nástroji. Na základě zjištěných údajů je možné říci, že metody strojového učení jsou vhodnými technikami pro problémy z oblasti proteinových predikcí.
|
|
|
Computational Workflow for the Prediction and Design of Stable Proteins
Kadleček, Josef ; Martínek, Tomáš (oponent) ; Musil, Miloš (vedoucí práce)
This thesis focuses on enriching computational tools for the prediction of protein mutations for the purpose of enriching their stability. Stable mutants of proteins are necessary for the development of the new drugs. Unfortunately, experimental validation of the stabilization effect of given mutations is costly and time demanding. Therefore, the FireProt tool was developed. FireProt utilizes a combination of both evolutionary and energy-based approaches for identifying potentially stabilizing mutations. This thesis enriches the FireProt tool with the possibility of entering custom mutations for the analysis. It also integrates other prediction software, FireProt ASR, and provides user with an option to select multiple mutational strategies for the detection of stabilizing mutations. Furthermore, it enriches the FireProt tool with another information about proteins residues (for instance relative B-factor) and makes it possible to utilize homology modeling to model the protein's structure based on its sequence. Lastly, it introduces a novel approach for the design of multiple-point mutants.
|
|
|
Computational Design of Stable Proteins
Musil, Miloš ; Lexa, Matej (oponent) ; Vinař, Tomáš (oponent) ; Zendulka, Jaroslav (vedoucí práce)
Stable proteins are utilized in a vast number of medical and biotechnological applications. However, the native proteins have mostly evolved to function under mild conditions inside the living cells. As a result, there is a great interest in increasing protein stability to enhance their utility in the harsh industrial conditions. In recent years, the field of protein engineering has matured to the point that enables tailoring of native proteins for specific practical applications. However, the identification of stable mutations is still burdened by costly and laborious experimental work. Computational methods offer attractive alternatives that allow a rapid search of the pool of potentially stabilizing mutations to prioritize them for further experimental validation. A plethora of the computational strategies was developed: i) force-field-based energy calculations, ii) evolution-based techniques, iii) machine learning, or iv) the combination of several approaches. Those strategies are usually limited in their predictions to less impactful single-point mutations, while some more sophisticated methods for prediction of multiple-point mutations require more complex inputs from the side of the user. The main aim of this Thesis is to provide users with a fully automated workflow that would allow for the prediction of the highly stable multiple-point mutants without the requirement of the extensive knowledge of the bioinformatics tools and the protein of interest. FireProt is a fully automated workflow for the design of the highly stable multiple-point mutants. It is a hybrid method that combines both energy- and evolution-based approaches in its calculation core, utilizing sequence information as a filter for robust force-field calculations. FireProt workflow not only detects a pool of potentially stabilizing mutations but also tries to combine them together while reducing the risk of antagonistic effects. FireProt-ASR is a fully automated workflow for ancestral sequence reconstruction, allowing users to utilize this protein engineering strategy without the need for the laborious manual work and the knowledge of the system of interest. It resolves all the steps required during the process of ancestral sequence reconstruction, including the collection of the biologically relevant homologs, construction of the rooted tree, and the reconstruction of the ancestral sequences and ancestral gaps.HotSpotWizard is a workflow for the automated design of mutations and smart libraries for the engineering of protein function and stability. It allows for a wider analysis of the protein of interest by utilizing four different protein engineering strategies: i) identification of the highly mutable residues located in the catalytic pockets and tunnels, ii) identification of the flexible regions, iii) calculation of the sequence consensus, and iv) identification of the correlated residues.FireProt-DB is a database of the known experimental data quantifying a protein stability. The main aim of this database is to standardize protein stability data, provide users with well-manageable storage, and allow them to construct protein stability datasets to use them as training sets for various machine learning applications.
|
|
|
Fully Automated Method for the Design of Ancestral Proteins
Štěpánek, Martin ; Martínek, Tomáš (oponent) ; Musil, Miloš (vedoucí práce)
Protein stability is a crucial feature for the industrial applications of proteins. This thesis focuses on the enhancements of FireProtASR, an automated tool for the design of stable proteins via ancestral sequence reconstruction. A novel technique of the successor prediction was developed and implemented into FireProtASR. Successors represent the evolutionary future of protein sequences and are supposed to have higher activity than current day proteins. They are also expected to have higher ability to target specific molecules. Furthermore, a new web user interface of FireProtASR was developed. It provides a fast and intuitive way to control the tool.
|
|
|
Fully Automated Method for the Design of Ancestral Proteins
Štěpánek, Martin ; Martínek, Tomáš (oponent) ; Musil, Miloš (vedoucí práce)
Protein stability is a crucial feature for the industrial applications of proteins. This thesis focuses on the enhancements of FireProtASR, an automated tool for the design of stable proteins via ancestral sequence reconstruction. A novel technique of the successor prediction was developed and implemented into FireProtASR. Successors represent the evolutionary future of protein sequences and are supposed to have higher activity than current day proteins. They are also expected to have higher ability to target specific molecules. Furthermore, a new web user interface of FireProtASR was developed. It provides a fast and intuitive way to control the tool.
|
|
|
Computational Workflow for the Prediction and Design of Stable Proteins
Kadleček, Josef ; Martínek, Tomáš (oponent) ; Musil, Miloš (vedoucí práce)
This thesis focuses on enriching computational tools for the prediction of protein mutations for the purpose of enriching their stability. Stable mutants of proteins are necessary for the development of the new drugs. Unfortunately, experimental validation of the stabilization effect of given mutations is costly and time demanding. Therefore, the FireProt tool was developed. FireProt utilizes a combination of both evolutionary and energy-based approaches for identifying potentially stabilizing mutations. This thesis enriches the FireProt tool with the possibility of entering custom mutations for the analysis. It also integrates other prediction software, FireProt ASR, and provides user with an option to select multiple mutational strategies for the detection of stabilizing mutations. Furthermore, it enriches the FireProt tool with another information about proteins residues (for instance relative B-factor) and makes it possible to utilize homology modeling to model the protein's structure based on its sequence. Lastly, it introduces a novel approach for the design of multiple-point mutants.
|
|
|
Computational Design of Stable Proteins
Musil, Miloš ; Lexa, Matej (oponent) ; Vinař, Tomáš (oponent) ; Zendulka, Jaroslav (vedoucí práce)
Stable proteins are utilized in a vast number of medical and biotechnological applications. However, the native proteins have mostly evolved to function under mild conditions inside the living cells. As a result, there is a great interest in increasing protein stability to enhance their utility in the harsh industrial conditions. In recent years, the field of protein engineering has matured to the point that enables tailoring of native proteins for specific practical applications. However, the identification of stable mutations is still burdened by costly and laborious experimental work. Computational methods offer attractive alternatives that allow a rapid search of the pool of potentially stabilizing mutations to prioritize them for further experimental validation. A plethora of the computational strategies was developed: i) force-field-based energy calculations, ii) evolution-based techniques, iii) machine learning, or iv) the combination of several approaches. Those strategies are usually limited in their predictions to less impactful single-point mutations, while some more sophisticated methods for prediction of multiple-point mutations require more complex inputs from the side of the user. The main aim of this Thesis is to provide users with a fully automated workflow that would allow for the prediction of the highly stable multiple-point mutants without the requirement of the extensive knowledge of the bioinformatics tools and the protein of interest. FireProt is a fully automated workflow for the design of the highly stable multiple-point mutants. It is a hybrid method that combines both energy- and evolution-based approaches in its calculation core, utilizing sequence information as a filter for robust force-field calculations. FireProt workflow not only detects a pool of potentially stabilizing mutations but also tries to combine them together while reducing the risk of antagonistic effects. FireProt-ASR is a fully automated workflow for ancestral sequence reconstruction, allowing users to utilize this protein engineering strategy without the need for the laborious manual work and the knowledge of the system of interest. It resolves all the steps required during the process of ancestral sequence reconstruction, including the collection of the biologically relevant homologs, construction of the rooted tree, and the reconstruction of the ancestral sequences and ancestral gaps.HotSpotWizard is a workflow for the automated design of mutations and smart libraries for the engineering of protein function and stability. It allows for a wider analysis of the protein of interest by utilizing four different protein engineering strategies: i) identification of the highly mutable residues located in the catalytic pockets and tunnels, ii) identification of the flexible regions, iii) calculation of the sequence consensus, and iv) identification of the correlated residues.FireProt-DB is a database of the known experimental data quantifying a protein stability. The main aim of this database is to standardize protein stability data, provide users with well-manageable storage, and allow them to construct protein stability datasets to use them as training sets for various machine learning applications.
|
|
|
Strojové učení v úloze predikce vlivu aminokyselinových mutací na stabilitu proteinu
Malinka, František ; Martínek, Tomáš (oponent) ; Bendl, Jaroslav (vedoucí práce)
Tato práce popisuje nový přístup k predikci vlivu aminokyselinových mutací na změnu stability proteinu. Cílem je vytvořit nový meta-nástroj, který kombinuje výstupy osmi vybraných nástrojů, díky čemuž je schopen svoji predikční schopnost zlepšit. Pro nalezení optimálního konsenzu mezi těmito nástroji je použito různých metod strojového učení. Ze všech testovaných metod strojového učení dosahuje KStar nejvyšší úspěšnosti predikce na trénovacím datasetu tvořeného experimentálně ověřenými mutacemi z databáze ProTherm. Právě z tohoto důvodu je KStar vybrán jako optimální predikční technika. Pro prokázání korektnosti výsledků tohoto meta-nástroje je použito testovacího datasetu vytvořeného ojedinělým způsobem, a to z vícebodových mutací extrahovaných taktéž z databáze ProTherm. Jelikož nebyly vícebodové mutace použity pro natrénování žádného z integrovaných nástrojů, předpokládá se, že takovéto porovnání je objektivní. Ve výsledku se tímto přístupem podařilo pomocí metody strojového učení KStar zvýšit korelační koeficient na trénovacím datasetu o 0,130, respektive o 0,239 na datasetu testovacím oproti nejúspěšnějšímu integrovanému nástroji. Na základě zjištěných údajů je možné říci, že metody strojového učení jsou vhodnými technikami pro problémy z oblasti proteinových predikcí.
|
host ::
RSS.