Development of bioisostere database and a program for ligand-based virtual screening

Vedno naprednejše in hitrejše računalniške metode imajo pomembno vlogo v različnih fazah sodobnega načrtovanju zdravilnih učinkovin. Virtualno rešetanje je postalo nepogrešljiv del iskanja novih aktivnih spojin (zadetkov) v začetnih stopnjah razvoja zdravil. Virtualno rešetanje lahko delimo na virtualno rešetanje na osnovi tarče in virtualno rešetanje na osnovi liganda. Bistvo obeh pristopov je, da iz velikih knjižnic molekul s hitrimi računalniškimi algoritmi na podlagi izbranih opisovalnikov (deskriptorjev) izluščimo tiste molekule, za katere obstaja večja verjetnost, da bodo izkazovale delovanje na določeno tarčo. Napredek računalniške tehnologije je omogočil nove pristope odkrivanja in uporabe sicer že dolgo znanih konceptov bioizosterije in skeletnega preskoka, ki sta bistvena v fazi optimizacije spojin vodnic. Bioizosterne zamenjave predstavljajo skupine atomov (fragmente), s katerimi lahko zamenjamo fragment bioaktivne spojine, da ostane biološko delovanje učinkovine ohranjeno oziroma začne spojina izkazovati diametralno različno biološko delovanje. Skeletni preskok pa predstavlja specifično obliko bioizosternih zamenjav, pri katerih spojini zamenjamo večji del podstrukture oziroma njen osrednji skelet. Bioizosterne zamenjave so pomembne, saj omogočajo izboljšanje farmakokinetičnih lastnosti, selektivnosti ali patentibilnosti spojine. V okviru doktorske naloge smo razvili algoritem LiSiCA za virtualno rešetanje na osnovi liganda in bazo BoBER za računalniško podprto iskanje in uporabo bioizosternih in skeletnih zamenjav. Oba razvita pristopa temeljita na matematičnem konceptu teorije grafov in s tem povezanim hitrim in učinkovitim algoritmom za iskanje maksimalne klike MaxCliqueDyn. Algoritem LiSiCA, ki omogoča paralelno dvodimenzionalno in tridimenzionalno virtualno rešetanje, smo validirali na bazi Directory of Useful Decoys – Enhanced. Izkaže se, da je program v mnogih primerih zmožen razlikovati med aktivnimi in neaktivnimi spojinami bolje kot nekateri drugi široko uporabljani programi za virtualno rešetanje na osnovi liganda. Program LiSiCA smo nato uporabili za iskanje novih inhibitorjev encima butirilholin esteraze, spojin, ki bi se lahko uporabljale za zdravljenje Alzheimerjeve bolezni. Pri tem smo odkrili pet novih struktur, ki so izkazovale nanomolarno inhibicijo encima, dodatno pa so imele značilno drugačno osnovno strukturo od referenčne spojine, s čimer smo pokazali sposobnost programa LiSiCA za skeletni preskok. Razvili smo tudi vtičnik LiSiCA v široko uporabljanem programu PyMOL, ki omogoča uporabniku prijazno in hitro izvedbo celotnega postopka virtualnega rešetanja na osnovi liganda, prav tako pa učinkovito vizualizacijo in primerjavo dobljenih rezultatov. Razvili smo pristop BoBER, ki sestoji iz baze bioizosterno oziroma skeletno zamenljivih fragmentov malih molekul in spletnega vmesnika, ki omogoča preprosto iskanje omenjenih zamenjav za izbrano referenčno spojino. Pridobivanje izmenljivih fragmentov temelji na algoritmu ProBiS, s katerim smo vezavna mesta vsakega holo proteina (proteina z vezanim ligandom) iz baze Protein Data Bank prilegali z vsemi ostalimi holo proteini. Hkrati smo dosegli tudi prileganje ligandov znotraj proteinskih vezavnih mest, kar nam omogoči prepoznavo prostorsko prekrivajočih se fragmentov teh ligandov. Pare fragmentov, za katere se izkaže, da se dovolj dobro prekrivajo v prostoru, smo dodali v bazo možnih bioizosternih zamenjav. Pristop BoBER je torej osredotočen na že objavljenih proteinskih strukturah oziroma ligandih vezanih na te strukture. Pristop smo uporabili za iskanje bioizosternih zamenjav inhibitorja butirilholin esteraze in inhibitorja monoamin oksidaze B, pri čemer se inhibitorji slednjega encima uporabljajo večinoma kot podpora pri zdravljenju Parkinsonove bolezni. Na študiji primera inhibitorja butirilholin esteraze smo pokazali možne bioizosterne zamenjave treh cikličnih fragmentov tega inhibitorja. V primeru inhibitorja monoamin oksidaze B pa smo iskali bioizosterne zamenjave za njegov fluorofenilni fragment z namenom povečanja njegove topnosti in zmanjšanja porazdelitvenega koeficienta (LogP). V obeh primerih se je izkazalo, da moramo za splošno uporabnost programa definirati pet pravil, ki omogočajo razširitev kemijskega prostora, znotraj katerega iščemo bioizosterne zamenjave. Ta pravila so nam omogočila, da smo za vsak referenčni fragment našli vsaj nekaj potencialnih bioizosternih zamenjav. Dodatno je del pristopa BoBER prosto dostopni spletni strežnik, ki omogoča uporabniku prijazno in hitro iskanje bioizosternih zamenjav prek intuitivnega vizualizacijskega vmesnika. Strežnik bioizosterne zamenjave črpa iz baze izmenljivih bioizosternih parov fragmentov BoBER. V sklop strežnika smo vključili tudi program LiSiCA, ki je nadgrajen tako, da poleg dvodimenzionalnih in tridimenzionalnih podobnosti na osnovi atomske resolucije, išče tudi podobnosti med molekulami na osnovi izmenljivih bioizosternih parov. Razvita računalniška pristopa prispevata k začetnim fazam razvoja zdravilnih učinkovin, in sicer pri iskanju zadetkov in optimizaciji spojin vodnic. Programi, navodila in primeri uporabe so prosto dostopni na spletnem naslovu http://insilab.org/.Advanced and fast computational methods have an essential role in modern drug design. Virtual screening has become an indispensable part in searching for new active compounds (hits) in initial stages of drug discovery. The virtual screening process can be ligand-based or target-based. The essence of both is in using efficient computer algorithms, which are based on selected descriptors, with the goal of obtaining structures that have a greater likelihood of binding to a certain target. Computational methods also allow for new approaches of detection and implementation of the familiar concepts of bioisosterism and scaffold hopping, which are crucial in the lead optimization phase of drug design. Bioisosteric replacements are atom groups (fragments) that can replace drug fragments, while retaining or reversing the biological activity of the selected drug. The concept of scaffold hopping represents a specific case of bioisosteric replacements, where a larger part of a substructure or respectively the main scaffold of a molecule is replaced. Bioisosteric replacements are of great importance in drug design as they allow for the improvement of the compound\u27s pharmacokinetic properties, selectivity and/or patentability. As part of this doctoral thesis we have developed a ligand-based virtual screening approach LiSiCA and a computer-aided knowledge based method for acquiring and implementing bioisosteric and scaffold replacements named BoBER. Both methods are based on the mathematical concept of graph theory and subsequently on the fast and efficient MaxCliqueDyn maximum clique algorithm. The validation of LiSiCA, which enables parallel 2D and 3D ligand-based virtual screening was performed using the Directory of Useful Decoys – Enhanced database. The program demonstrated its ability to differentiate between active and inactive compounds while in a number of cases surpassing commonly used virtual screening methods. Additionally LiSiCA was implemented in searching for novel butyrylcholinesterase enzyme inhibitors, which can be used in the treatment of Alzheimer\u27s disease. In doing so we discovered five new molecules exhibiting nanomolar inhibition of the enzyme, which also had a different core structure compared to the reference one. We thus proved LiSiCA\u27s ability for scaffold hopping. Moreover, we developed LiSiCA-PyMOL plugin, which allows for a user-friendly and fast execution of a complete ligand-based virtual screening procedure with subsequent visualization and comparison of results. The BoBER method combines a database containing fragments which can be used in bioisosteric or scaffold replacements in small molecules, and a web interface which implements this database and enables an easy-to-use query searching for bioisosteric replacements of a specific reference structure. The acquisition of replaceable fragments is based on the ProBiS algorithm, which is used to superimpose every holo protein\u27s (a protein containing a bound ligand) binding site from the Protein Data Base to every other holo protein. This binding site superimposition also leads to the alignment of ligands which are part of the binding sites. This allows for the identification of spatially overlapping ligand fragments. A pair of fragments that show a good enough spatial overlap is added to the mentioned database of bioisosteric replacements. BoBER was tested by searching for possible bioisosteric replacements of butyrylcholinesterase and monoamine oxidase B inhibitors, where the latter are used primarily as supporting treatment of Parkinson\u27s disease. Bioisosteric replacements of butyrylcholinesterase inhibitor ring fragments were shown as a general example of the usage of BoBER. Meanwhile, in the monoamine oxidase B inhibitor case, ioisosteric replacements for a specific fluorophenyl fragment were sought in order to increase the solubility and decrease the partition coefficient (LogP) of the compound. Both cases suggested that in order to increase the universality of the method, we had to define five rules, which enabled us to expand the chemical space from which we are acquiring bioisosteric replacements. The implementation of these rules made possible that for each reference fragment at least a few potential bioisosteric replacements were obtained. In addition, the BoBER web server is part of the BoBER method, which allows for a user-friendly and fast search of bioisosteric and scaffold replacements. The freely accessible web server implements the aforementioned database of replaceable pairs, which a user can access and employ via an intuitive visualization-enabling interface. The server also includes the LiSiCA program, which now allows for bioisoster-based similarity searching. The newly developed computational-based approaches can contribute in the early stages of drug design, especially in searching for hits and optimization of lead compounds. The software, instructions and usage examples are available at http://insilab.org/