China
Africa
Explore chapters and articles related to this topic
Integrating CADD and Herbal Informatics Approach to Explore Potential Drug Candidates Against HPV E6 Associated With Cervical Cancer
Published in Shazia Rashid, Ankur Saxena, Sabia Rashid, Latest Advances in Diagnosis and Treatment of Women-Associated Cancers, 2022
Arushi Verma, Jyoti Bala, Navkiran Kaur, Anupama Avasthi
For ligand analysis, two probable hits were short-listed after a thorough literature search: luteolin (an anti-cancerous, anti-allergenic, antioxidant, anti-inflammatory agent) [25] and daphnoretin (an anti-apoptotic, anti-cancerous [26], and antiviral (CHEBI: 4324) agent). Physicochemical properties (structure and formula, source, class of compound, molecular weight, SMILES id, H-bond acceptors and donors, topological surface area, etc.) were analysed and are tabulated in Table 8.1. PubChem’s SMILES notation was used in Chimera to create 3D structures of phytochemicals and appropriately processed and viewed on Chimera (Figures 8.6, 8.7).
AI/ML in Medical Research and Drug Development
Published in Wei Zhang, Fangrong Yan, Feng Chen, Shein-Chung Chow, Advanced Statistics in Regulatory Critical Clinical Initiatives, 2022
The Simplified Molecular-Input Line-Entry System (SMILES), which represents molecular graphs as strings of characters [Weininger, 1988] has facilitated the use of recurrent neural networks (RNN) and other deep learning methods in de novo drug design. For instance, [Arús-Pous et al., 2020] use a deep learning SMILES-based generative architecture that first exploits RNN that generates scaffolds and then a model to generate suitable decorations for each attachment point in the scaffold. Along similar lines one can refer to [Lim et al., 2020] and [Li et al., 2019] where graph generative neural networks (GGNN) are utilized molecular generation from scaffolds. Moreover, [Ma et al., 2015] show that deep neural networks are a better practical method for quantitative structure-activity relationship (QSAR) problems for predicting on-target and off-target activities in the drug discovery process.
In Silico approach of soursop leaf for prediction of anticancer molecular target therapy
Published in Ade Gafar Abdullah, Isma Widiaty, Cep Ubad Abdullah, Medical Technology and Environmental Health, 2020
M.K. Dewi, Y. Kharisma, L. Yuniarti
SMILES is able to predict the target protein of interaction of a compound based on structure-based similarity between the structure we want to predict (query structure) with the structure of FDA-approved drugs. It is also able to predict the interaction on non-drug compounds that have been analyzed in vitro and in vivo (Gfeller et al. 2014, Dunken et al. 2008). Based on the findings, the target proteins selected were the ones above 70%. The next step is the pathway analysis of these proteins on protein targets that play a role in the melanogenesis process is done using a STRING and cystoscope.
Quantitative structure–toxicity relationship models for predication of toxicity of ionic liquids toward leukemia rat cell line IPC-81 based on index of ideality of correlation
Published in Toxicology Mechanisms and Methods, 2022
Shahin Ahmadi, Shahram Lotfi, Parvin Kumar
To develop the QSTR model, the cytotoxicity data (log EC50) toward Leukemia rat cell line IPC-81 for a large set of various 304 ILs containing eight cations and 12 types of anions was used and retrieved from the literature (Sosnowska et al. 2017). The data set mainly contains 35 ammonium, 104 imidazolium, 33 morpholinium, 4 phosphonium, 17 piperidinium, 67 pyridinium, 28 pyrrolidinium, 5 quinolinium, 2 sulfonium, and 9 protic ILs based on substituted amines (mono-, di-, and tri-ethanolamine). The logEC50 (µM) of ILs against Leukemia rat cell line IPC-81 was used as an endpoint. The range of logEC50 was from −0.24 to 4.90 µM. Using free software BIOVIA draw 2019, the molecular structure of all ILs was depicted. After drawing the molecular structure, transformed into SMILES notations for the modeling process. The IDs of ILs, SMILES code, as well as the corresponding experimental and predicted logEC50 values of ILs provided in Table S1. The dataset was split into four splits (Table 1). The individual split was randomly distributed into four sets, i.e. the training (≈34% of the total data set), invisible training (≈23% of the total data set), calibration (CAL; ≈17% of the total data set), and validation (≈25% of the total data set) sets. The task of each set is well defined in the literature (Toropov and Toropova 2019; Lotfi et al. 2021).
The power of deep learning to ligand-based novel drug discovery
Published in Expert Opinion on Drug Discovery, 2020
Deep learning provides very powerful tools that can be used in ligand-based novel drug discovery both to conduct virtual screening of the already prepared libraries of chemical compounds and to generate new compounds with desired properties. The deep architecture of neural networks with multiple hidden layers makes an internal representation of molecules more suitable for predicting target properties, as well as for transferring knowledge between different tasks. CNNs can work directly with chemical structures without the use of descriptor sets previously invented by humans. Using RNNs, one can apply text processing methods to chemical structures represented using SMILES strings. Chemical structures can be generated either through SMILES strings or directly working with molecular graphs. Autoencoders can be used to produce ‘invertible’ descriptors from which chemical structures can be recovered. This allows one to effectively design new compounds, both by generating new chemical structures for a given property and by optimizing chemical structures in the latent space of autoencoders. Different types of VAEs, GANs, and RL are powerful tools that can be very useful for ligand-based drug discovery. Several success stories that have already been reported in the literature.
An overview of neural networks for drug discovery and the inputs used
Published in Expert Opinion on Drug Discovery, 2018
Yinqiu Xu, Hequan Yao, Kejiang Lin
The main purpose of using SMILES as inputs is to generate new SMILES sequences so that automatic de novo drug design can be realized. RNNs can be trained for predicting next tokens according to current inputs and the states of the networks, with the states having information about previous contents. With cells like GRU cells and long short-term memory (LSTM) cells, the propagation of information is controlled by gates so that the cells can learn to predict according to important inputs. After being trained with molecules that have the same property, RNNs can reproduce them or create new molecules according to corresponding probability distributions. However, it is difficult for RNNs to produce valid SMILES with limited molecules for training. This problem was solved by fine-tuning [69], a method of TL. Meanwhile, sampling temperatures were applied by Gupta et al. [70] to control the novelty of the produced SMILES. The method was applied in practice, and five new compounds targeting retinoid X receptors (RXRs) and peroxisome proliferator-activated receptors (PPARs) were designed, of which four show activities [71].