De novo načrtovanje ciklično simetričnih homo-oligomerov

Ribnikar, Neža (author), Ljubetič, Ajasja (mentor), Novinec, Marko (comentor)

Metode strojnega učenja so se nedavno pojavile kot strategija za učinkovito raziskovanje funkcionalnega prostora proteinov, ki med drugim omogočajo tudi načrtovanje povsem novih proteinskih struktur. Nedavno razvit difuzijski model RFdiffusion poleg nevodenega načrtovanja proteinskih struktur (tj. brez uporabe pogojnih informacij) omogoča tudi načrtovanje simetričnih oligomernih struktur s poljubno točkovno simetrijo. V okviru magistrske naloge smo želeli načrtovati ciklično simetrične homo-oligomere s C10 simetrijo z uporabo programov RFdiffusion ter ProteinMPNN. Načrtovanje dekamerov namreč še vedno predstavlja zahtevno nalogo, saj so razlike v vmesnih površinah med C9 in C10 oligomeri zelo majhne. Do danes je bil edini uspešno načrtovan in eksperimentalno potrjen primer dekamera opisan v delu Huddyja in sodelavcev. Z uporabo omenjenih programov smo načrtovali tri serije de novo proteinov, za katere je bilo z bioinformacijskimi orodji napovedano, da tvorijo homo-oligomerne dekamere. Z uporabo algoritma za napovedovanje strukture AlphaFold2 in AlphaFold3 smo pridobili in silico parametre za ocenjevanje načrtov, dodatne fizikalne parametre pa smo izračunali še z uporabo programa Rosetta. Na podlagi pridobljenih rezultatov smo nato filtrirali načrte in za laboratorijsko testiranje izbrali tiste z najugodnejšimi karakteristikami za posamezen parameter. Namen eksperimentalnega dela je bil preveriti, ali se načrtovani proteini v laboratorijskih pogojih izražajo in tvorijo dekamere. Za izbrane načrte smo zato naročili sintetične gene ter jih izrazili v bakterijskih celicah. Po čiščenju z nikelj-afinitetno kromatografijo smo tvorbo dekamernih struktur preverili s kromatografijo z ločevanjem po velikosti (angl. Size Exclusion Chromatography, SEC). Najobetavnejše proteine smo nato analizirali še s pomočjo masne fotometrije. Dobljeni rezultati nudijo vpogled v zmožnosti načrtovanja simetričnih višještevnih homo-oligomerov z metodami globokega učenja. V sklopu magistrskega dela smo uspeli načrtovati več obetavnih proteinov, ki so med drugim tvorili tudi dekamerne strukture, zato to delo predstavlja temelj za nadaljnje raziskave na področju načrtovanja večjih de novo oligomerov.

• Language: Slovenian
• Keywords: RFdiffusion, de novo proteini, C10 simetrija, SEC, masna fotometrija
• Work type: Master's thesis/paper
• Organization: FKKT - Faculty of Chemistry and Chemical Technology
• Year: 2026
• PID: 20.500.12556/RUL-190897
• Publication date in RUL: 06.11.2026

Secondary language

• Language: English
• Title: De novo design of cyclically symmetric homo-oligomers

Abstract

Machine learning methods have recently been established as a powerful tool for exploring the functional space of proteins, including the design of entirely novel protein structures. The recently developed diffusion model RFdiffusion has enabled unconditional design of novel protein structures (i.e. without conditional information) as well as the design of symmetric protein structures with a chosen point group symmetry. In this master's thesis, we have focused on the design of C10 symmetric cyclic homo-oligomers using RFdiffusion and ProteinMPNN. The design of decamers still remains a challenging task mainly due to small differences in interfaces between C10 and C9 oligomers. To date, the only successfully designed and experimentally confirmed decameric protein has been described by Huddy and colleagues. With the use of the aforementioned programs, we have designed three rounds of de novo proteins for which the bioinformatics tools have predicted decameric structure. The algorithms for protein structure prediction AlphaFold2 and AlphaFold3 have enabled us to obtain in silico parameters for evaluating the designs as well as the additionally calculated physical parameters with Rosetta. Based on the obtained results we selected those designs with the most favorable characteristics for different parameters for experimental evaluation. The aim of the experimental work was to determine whether the designed proteins express well in bacterial cells and if they assemble into decamers. For the selected designs we therefore ordered synthetic genes and expressed them in bacterial cells. After nickel-affinity chromatography purification, we looked into the formation of decamer structures using size-exclusion chromatography (SEC). The most promising proteins were further analyzed using mass photometry. The results provide an insight into the current abilities for the design of higher-order symmetric homo-oligomers. We have successfully designed a few proteins with promising characteristics, including those which also formed decameric structures, establishing a foundation for further research in the field of de novo oligomer design.

• Keywords: RFdiffusion, de novo proteins, C10 symmetry, SEC, mass photometry