Mixed-integer optimization approach for fragment-based design of drug candidates

Abstract

Ilac tasarm icin bilgisayar metotlarnn gelistirilmesi, son yirmi yldr en cok calslanalanlardan biri olmustur. Bu calsmalardaki asl amac, bilinen ilaclardaki topoloji veetkilesim enerji degerleri goz onunde bulundurularak baglanma ve kenetlenme enerjilerien dusuk olacak sekilde yeni ilaclar kesfetmektir. Buyuk molekullerce olusturulankutuphaneleri kullanmak yerine bilinen ilaclarn parcalarn kullanarak yeni ilaclardizayn etmek, daha cok kombinasyon olusumu saglamaktadr. Bilinen ilaclarn parcalarkullanlarak olusturulan bu kombinasyonlar ile, var olan molekullerden daha dusukenerjili ilaclarn tasarlanlmas mumkundur.Deneysel metotlar, parca bazl ilac tasarmn etkili bir sekilde uygulamak adnadusuk molekuler agrlkl molekullerin hedef proteine baglanmalarn test etmek icinbinlerce urun taramaktadrlar. Parca bazl ilac tasarmnda basarl sonuclar ise sadecebilgisayar metotlarnn yenilikci NMR deneyleri ve X-Ray kristalograsi ile etkilesimidurumunda elde edilmektedir.Varolan bircok bilgisayar teknigi, hangi parcann proteinyuzeyine etkili bir sekilde baglanacagn duzgun olarak tahmin edememektedir.Bu calsmada, ilac tasarmnda ogrenme ve dizayn basamaklarndan olusan bilgisayarlbir yontem gelistirilmistir. _Ilk olarak, yaps bilinen molekullerin kenetlenme vebaglanma enerjileri kullanlarak, yaplardaki parcalara enerji degerlerine katklarnagore ayr ayr agrlk katsaylar belirleyen bir ogrenme algoritmas uygulanr. Dahasonra dizayn ksmna gecilir ve ilk admda elde edilen agrlk katsaylar kullanlaraken dusuk kenetlenme ve baglanma enerjilerini veren parcalar kullanlarak tamamenyeni ilac tasarmlar yaplr.

Developing computational methods for drug design has become one of the mostwidely studied areas of life sciences in the past two decades. The main objective is todiscover new drugs using the information regarding topology and interaction energiesof known drugs, that minimizes binding energy and docking energy. Designing newdrugs from the fragments of known drugs rather than libraries of larger moleculesprovides more combinations in products. From these many possible combinations itis possible to nd new drugs that have lower energy values than existing molecularstructures.Experimental methods were applied to screen thousands of low-molecular-weight compoundsfor testing their binding to the target protein in using fragment based drugdesign effectively. Successful results in fragment based drug design were only obtainedwith the strong integration of computational techniques, innovative NMRexperiments and X-Ray crystallography. Many existing computational algorithmscannot accurately predict the affinity with which fragments might or might not bindto the protein surface alone. In this work, we developed a computational modelwhich includes both learning and design parts. First, we apply a learning algorithmon known results of docking and binding energies to determine weights for the eachfragment reliably. Then, in the design part, the regression model is formed based onthe results obtained in the first step, to predict the contribution of each group andindividual fragment in docking energy and binding energy and finally designs entirelynew molecules based on the fragments available.