Iskanje ligandov za biološke tarče s pomočjo algoritmov in skupnostne znanosti

Pleško, Sebastian (avtor), Podlipnik, Črtomir (mentor)

V magistrskem delu sem preizkusil uporabo genetskega algoritma in koncepta skupnostne znanosti pri in silico molekulskem sidranju malih molekul v proteinske tarče. Preizkušeni so bili različni pristopi k problemu iskanja najboljših ligandov. Za najboljšega se je izkazal pristop kombiniranja genetskega algoritma (z uporabo ustreznih SMARTS filtrov) s konceptom igrificirane skupnostne znanosti. Pri tem pristopu so posamezniki, predvsem dijaki, lahko predlagali popolnoma nove molekule ali izboljšali molekulo drugega dijaka in hkrati tekmovali med seboj za čim boljši rezultat. Vzporedno pa je genetski algoritem najboljše predlagane molekule razvijal oz. mutiral naprej. Mutirane molekule in molekule, ki so jih predlagali posamezniki, in niso ustrezale enostavnemu filtru za biološko uporabnost, določene z uporabo enostavnih kemijskih deskriptorjev (Veberjev filter), je algoritem avtomatsko odstranil. Odstranil je tudi molekule, katerim je algoritem s pomočjo SMARTS filtrov določil, da vsebujejo znane kemijsko nestabilne, reaktivne ali toksične skupine. Na enak način je algoritem odstranil še molekule, ki so vsebovale t.i. PAINS skupine, ki napovedujejo nezaželeno splošno neselektivno vezavo molekule na različne tarče. Ta sinergija pristopov je dala daleč boljše rezultate, gledano z vidika napovedane vezavne energije sidranja, saj je našla kar 100 ligandov z napovedano boljšo vezavo, od drugega najbolje uvrščenega liganda, pridobljenega s klasičnimi strategijami s knjižnicami spojin. Kljub odličnim rezultatom pristopa pa smo med pregledovanjem literature odkrili potencialen problem uporabe tega pristopa v praksi, ki bi se znal pokazati v tem, da najdene spojine ne bi dosegale dobrih vezav v praksi oz. in vitro. Omejitev se je pokazala predvsem v tem, da so hitri računski modeli sidranja zaenkrat še neprimerni za napoved slabe vezave molekul, zaradi česar se dobljeni rezultati lahko ne bi skladali s tistimi, dobljenimi v praksi. Slednjih trditev v tem magistrskem delu nismo preverili v praksi oz. in vitro.

• Jezik: Slovenski jezik
• Ključne besede: genetski algoritem, skupnostna znanost, molekulsko sidranje
• Vrsta gradiva: Magistrsko delo/naloga
• Tipologija: 2.09 - Magistrsko delo
• Organizacija: FKKT - Fakulteta za kemijo in kemijsko tehnologijo
• Leto izida: 2023
• PID: 20.500.12556/RUL-146406
• COBISS.SI-ID: 158697219
• Datum objave v RUL: 31.05.2023

Sekundarni jezik

• Jezik: Angleški jezik
• Naslov: Finding ligands for biological targets using algorithms and the use of citizen science

Izvleček

In the master's thesis we've tested the use of a genetic algorithm in conjuction with the concept of citizen science in the in silico molecular docking of small molecules in protein targets. Out of the tested approaches for finding the best ligands, we've found out that the combination with using a genetic algorithm (with the right SMARTS filters) in conjuction of using a gamified approach of citizen science gave the best results. In this approach, individuals, who were mostly high school students, had the possibility of proposing completely new molecules or improve molecules created by other students and at the same time compete among each other in finding the molecule with the best score. In parallel to the input from indidividuals, the genetic algorithm developed (mutated) further the best molecule proposals. The algorithm automatically removed mutated molecules and the proposed molecules by individuals, that haven't passed a simple filter for biological availability that used simple chemical descriptors (Veber filter). Molecules which contained groups, determined via SMARTS filters, that are chemically unstable, reactive, toxic or were classified as PAINS molecules, which predict an unwanted unselective binding of molecules to various targets, were also removed. This sinergy of approaches gave much better results in the sense of predicted binding energy of docking, as it has resulted in 100 ligands with a better predicted binding energy, than the 2nd best ligand that has been found using with the classical strategy of using libraries of compounds. Despite excellent results of this approach, it has been found out, using thorough literature research, that there's a potential problem of using this approach in practice, as we suspected that the best found molecules using this approach, wouldn't achieve such good binding energies in practice in vitro. The limitation is in the quick computational models for molecular docking, which are for now unsuited for prediction of bad binding of molecules. Consequentially our results might not match practical measurments. In this thesis, the later claims weren't tested in practice i.e. in vitro.

• Ključne besede: genetic algorithm, citizen science, molecular docking