LEE- Hesaplamalı Bilim ve Mühendislik-Yüksek Lisans
Bu koleksiyon için kalıcı URI
Gözat
Başlık ile LEE- Hesaplamalı Bilim ve Mühendislik-Yüksek Lisans'a göz atma
Sayfa başına sonuç
Sıralama Seçenekleri
-
ÖgeIdentification of novel inhibitors targeting putative dantrolene binding site for ryanodine receptor 2(Graduate School, 2022-06-13) Saylan, Cemil Can ; Baday, Sefer ; 702191003 ; Computational Science and EngineeringRyanodine receptors (RyRs) are large (around 20 MDa) homotetrameric intracellular ion channels. RyRs are located in the membrane of the sarcoplasmic reticulum (SR). RyRs play a central role in the excitation-contraction coupling by regulating Ca2+ release from the SR to the cytosol. Three isoforms are sharing 70% sequence similarity; RyR1, RyR2 and RyR3 are predominantly expressed in skeletal, cardiac muscles and neurons, respectively. In all isoforms, the open and close state transition occurs by rigid-body shifts of domains that provide the overall breathing motion of the cytoplasmic region. Dysregulation of RyRs leads to abnormal cellular activity. More than 300 mutations have been associated with muscle and neuronal diseases. Today, there have been identified several heart diseases caused by aberrant RyR2 activity such as catecholaminergic polymorphic ventricular tachycardia, cardiomyopathies, and cardiac arrhythmias. This unwanted RyR2 activity can be modulated by drugs. Dantrolene is an approved muscle relaxant used for the treatment of malignant hyperthermia that occurs by dysregulation of RyR1. Previously, a dantrolene binding sequence was suggested, and this sequence is conserved for all RyRs. However, dantrolene has poor water solubility and dantrolene exerts its effect on RyR2 only in the presence of regulators such as Mg2+ and calmodulin. In addition to this, although the amino acids 590-609 in RyR1 (601-620 RyR2 equivalent) were identified as dantrolene binding sequences, dantrolene bound complex structure has not been elucidated yet. Here, we aimed three things; 1) modelling of full atom structure of RyR2 with membrane, 2) predicting dantrolene binding orientation and 3) identifying novel inhibitors targeted to the putative binding sequence of dantrolene to regulate RyR2 function. While most of the structure of RyR2 has been solved recently, some of the regions are still missing. To predict missing regions, we used trRosetta and AlphaFold2 which is a state-of-art deep-learning-based method for protein modelling. Each missing segment was modelled separately and combined at the end. The final model, then, was optimized by 35ns MD simulation. Subsequently, to predict the dantrolene binding pose, the putative dantrolene binding site was searched using three different docking programs; Vina, LeDock and Glide. There was a distinct difference around the dantrolene binding site between the AF2-based model and Cryo-EM based model. Thus, the docking was performed for either structure. The dantrolene population was obtained predominantly in a particular cavity formed by 6 domains. Among docking results, five binding poses (3 of cryo-em, 2 for AF2 model) were selected regarding the affinity scores and pose similarity. These poses were used in 200ns MD simulation (298K, NPT) to address the pose binding behaviour. In the simulations truncated system was used, and restraints were introduced at regions where we split the structure. After 200ns, four orientations of dantrolene (2 for each) remained in its interaction with the dantrolene binding sequence. To calculate the binding free energy of the binding poses, MMPBSA analysis was performed using MD trajectories. Besides this, we also investigated the FKBP12.6 binding effect on dantrolene binding using the docking structures. Here, all proceeded MD runs with dantrolene were replicated with FKBP12.6 bound conditions. Structural clustering of MD simulations together with MMPBSA results showed that particular dantrolene orientation showed the highest binding capability to around R606, E1649, and L1650 residues. However, we could not identify a significant effect of FKBP12.6 on dantrolene binding. Next, for the identification of novel inhibitors, we focused on the dantrolene binding region and the high-throughput screening of 3.5 million molecules retrieved from the ZINC15 database was applied. The molecules were selected based on molecular weights (<450 Da) and logP values (<3.5). Virtual screening has proceeded with 3-step gradual filtering. The initial step includes 3.5 million molecule screening using AutoDock Vina with 8 exhaustiveness. This step was followed by two screening procedures in which the top-ranked 200K molecules selected from the previous step were filtered with LeDock and Vina (with 24 exhaustiveness). Molecules shared in top-10K for both Vina and LeDock results were used for the third and last screening with GlideXP. Subsequently, among the top-100 molecules, top-20 was selected as the final candidate list. According to >%70 human oral absorption conditions, the best 11 molecules have proceeded for MD simulations. 200ns MD simulation was carried out using Desmond. The seven molecules remained in their interaction with the dantrolene binding sequence. These were suggested as candidates that might regulate RyR2 activity. Particularly, two molecules showed significant stability at the binding site at around 1Å. These molecules will be tested experimentally by our experimental collaborators.
-
ÖgePredicting the bandgap of hole-transport materials by deep learning(Graduate School, 2023-01-30) Aydın, Miraç ; Tekin, Adem ; 702191021 ; Computational Science and EngineeringÇalışmada öncelikle OMDB veri seti yapay çoklamaya maruz bırakılmadan kullanıldı. SchNetPack ve ALIGNN modellerinin makalelerinde yer verilen varsayılan parametrelerle eğitim süreci gerçekleştirildi. Eğitim süreci NVIDIA Tesla A100 (40GB), 2 adet NVIDIA RTX A4500 (20GB) ve NVIDIA RTX A4000 (16GB) grafik kartları üzerinde gerçekleştirildi. Yapılan eğitim sonucunda SchNetPack ve ALIGNN modelleri için MAE değeri sırasıyla 0.43 eV ve 0.25 eV bulundu. Bu değerlere sahip modellerle Spiro-OMeTAD molekülünün bant aralığı tahmini yapıldı. Literatür değeri 3.05 eV olan bant aralığı, SchNetPack ve ALIGNN modelleri tarafından sırasıyla 2.73 eV ve 2.52 olarak tahmin edildi. Daha sonra OMDB veri setine Kristalografi Açık Veritabanı'nda (Crystallography Open Database, COD) bulunan 10 adet delik geçiş malzemesi yapısı ve her bir yapı için 10 adet konformer olacak şekilde toplamda 100 adet yapı eklendi. Eklenen bu yapılarla beraber veri setine AugLiChem kütüphanesi ile yapay çoklama yöntemi uygulandı. Model performansını artırabilmek için yeni yapıların araştırılmasına devam edildi ve makale taramalarından 79 yeni yapı daha bulundu. Bu yapıların bant genişliği değerleri DFT metodu ile hesaplandıktan sonra veri setine eklendi. Yapılan bu işlemler neticesinde veri setinde toplamda 52835 yapı elde edildi. SchNetPack ve ALIGNN modelleri, yapay çoklanmış OMDB veri seti ve farklı parametrelerle birçok kez eğitildi. Bu eğitimler sonucunda en düşük MAE değerleri SchNetPack ve ALIGNN modelleri için MAE sırasıyla 0.23 eV ve 0.25 eV olarak bulundu. Bu değerlere sahip modellerle Spiro-OMeTAD molekülünün bant aralığı tahmini yapıldı. Literatür değeri 3.05 eV olan bant aralığı, SchNetPack ve ALIGNN modelleri tarafından sırasıyla 2.97 eV ve 2.82 olarak tahmin edildi. Yapılan yapay çoklama ve parametre değişiklikleri sonrasında modellerin MAE değerleri ortalama %40, bant aralığı değeri tahmin performansı ise ortalama %13 oranında artırılmıştır. Bant değeri tahmini yapılan diğer moleküller çalışma içerisinde gösterilmişti