For the first time, the mathematical and theoretical foundations of a complex approach to the creation of a computer program for the virtual screening of NO scavengers in a number of newly synthesized compounds were developed. Approaches to the implementation of the software complex are described.
In our work, for the first time, the antioxidant activity of 532 xanthine derivatives was evaluated in vitro for NO inhibition. For the first time, with the help of semi-empirical quantum chemical methods, the main descriptors of the frontier molecular orbitals of xanthine derivatives have been substantiated
by their influence on the ability of these compounds to bind NO. This research aims to assess the in vitro antioxidant properties of 532 xanthine derivatives with regard to NO inhibition. The dependence of antioxidant activity on the quantum chemical parameters of xanthine derivatives was analyzed using machine learning algorithms using the following models: Linear Regression, Support Vector Machine Regression, Random Forest Regression, Gradient Boosting Regression, K-Nearest Neighbor Regression. As a result of our analysis, we tested several models for solving regression problems. The best models without optimization turned out to be the "Support Vector Machine Regression" and "K-Nearest Neighbors Regression" models. When optimizing the studied models, the Gradient Boosting Regression model showed the best generalizing ability with an error within 16%. This model can be used for the prediction of antioxidant activity based on quantum chemical parameters. The model's quality can be further improved by increasing the training and test samples, as well as expanding the features to deepen the model and improve the generalization ability. A program of virtual screening of substances with the properties of NO scavengers has been developed and created. In the process of testing the new synthesized xanthine derivatives, a computer program made it possible to predict the most pronounced properties of the NO scavenger in 8- enzylaminotheophilinyl-7-acetic acid hydrazide (C-3). In vitro experiments confirmed the prediction of the properties of the NO scavenger in C-3 (267.3%). Addition of C-3 (10 -5 M) to the incubation mixture leads to a decrease in nitrotyrosine by 45% and oxidized glutathione by 53.2% concomitantly with an increase in the concentration of reduced glutathione by 43.8% and increase in the activity of GSH-dependent enzymes - GPR by 337% and GR by 195% (p < 0.05). It should be noted that the antioxidant effect of C-3 is accompanied by an increase in concentration of HSP 70 by 34.7%. By regulating the level of NO and its cytotoxic forms, C- 3 is able to reduce the suppression of GSH, which determines the concentration of HSP 70 . In terms of potency, C-3 is significantly superior to Mexidol (10 -5 M). The obtained results in vitro confirm the results of the NO scavenger’s C-3 compound obtained as a result of the virtual screening.
What challenges or limitations did you encounter in the development of the computer program for virtual screening, and how were they overcome?
Thank you for your interest in our work!
While developing the virtual screening program, we encountered several challenges that required careful consideration. Firstly, the selection of appropriate models for machine learning algorithms posed a significant hurdle. Secondly, the task of identifying and validating the most specific in-vitro method for binding chemical molecules to NO added complexity to our research. Additionally, optimizing machine learning models to minimize result errors emerged as a crucial aspect.
Moreover, we recognized the necessity for preliminary studies involving a substantial number of chemical compounds at various molar dilutions. This step was essential to augment both the training and test samples, enhancing the robustness of the models. Furthermore, we focused on expanding the features to deepen the model's understanding and improve its generalization ability.
Addressing these challenges was pivotal in refining our virtual screening program, ensuring its accuracy, and bolstering its potential applications in the field.
We have a question for you, you can read and answer bellow.
Question for Authors:
Have you already considered the posibility of implementing and online web server for online use?
What are the SWOTs of using regression vs. classification techniques?
What are better candidate species for such studies according to your experience?
REVIEWWWERS'23 participation:
We also invite you to participate in the REVIEWWWERS Workshop, which is now open,
making questions to other authors. The steps are very easy. instructions:
Step(1), Sign in/Login here to Sciforum platform https://login.mdpi.com/login.
Step(2), Go to presetations list [MOL2NET'23 Papers List] https://mol2net-09.sciforum.net/presentations/view.
Step(3), Scroll down papers list and click on one title of the communication you selected.
Step(4), Scroll down and click on Commenting button, post your comment, and click submit.
Step(5), Repeat review process for other papers including across comments in othe conference congresses.
Step (6), Check your email for responses from the authors and counter-argue/thank them for it.
Step (7), Remember to check your email if you have had questions about your own work(s) and answer them.
Step(8), Request your attendance certificate at Email: mol2net.chair@gmail.com.
Sincerely yours
MOL2NET Team
Thank you for your questions:
Question1: Have you already considered the posibility of implementing and online web server for online use?
We plan to develop an information system in the form of a web application of a computer program with a built-in database of obtained experimental data and the ability to be updated via the Internet. The computer program can be used for wireless Medscape devices (for iPhone, iPad, as well as for other cell phones and tablets).
Q2: What are the SWOTs of using regression vs. classification techniques?
We used a regression model using the method of Support Vectors. The result of this model is an error of 18.97% on the test and 17.47% on the training sample, respectively.
Q3: What are better candidate species for such studies according to your experience?
We can draw conclusions from our research:
As a result of our analysis, we have tested several models for to solve regression problems.
2. The best models without the use of optimization were the following Support Vectors and K-nearest neighbors models.
3. When optimizing the studied models, the best generalizing ability was shown by the Gradient Boosting model with an error within 16%. This model can be used to predict antioxidant activity based on quantum-physical parameters.
4. Further improvement of the model quality is possible by increasing the training and test samples, as well as expanding the features to deepen the the model and improve its generalizability.
Sincerely, the authors
