The pharmaceutical industry is one of the biggest industries in the modern world. This industry aims to increase people’s lifespan and improve people’s health. Its revenue has nearly tripled since 2001 and exceeded 1.25 trillion dollars in 2019 and will keep growing.¹ Drug designing and production have a considerable contribution to that revenue, however, conventional drug designing methods are getting outdated and more costly. Computer-Aided Drug Design(CADD) is a multidisciplinary approach that aims to revolutionize the conventional drug discovery techniques and optimize the drug discovery mechanism in terms of cost, energy, and time. This approach highly depends on machine learning and artificial intelligence, so it minimizes the need for biological and chemical testings and gives the chance of modeling drug-protein interactions as accurately as possible.
In this post, an in-depth CADD mechanism analysis with several important examples of already existing drugs that are designed using computational methods will be introduced, followed by current approaches and a discussion of speculations on potential uses of the CADD and its current challenges.
Although CADD is a relatively new approach to drug production there are numerous methods used for different purposes and conditions. The most famous of them are Structure-Based Drug Design (SBDD) and Ligand-Based Drug Design(LBDD), the former uses the “binding site identification”, “docking and scoring” techniques while the latter uses “QSAR” and “pharmacophore” technologies. Both of them will be introduced with their relative strengths and weaknesses in the following paragraphs, and they will be followed by the rest of the drug discovery steps.
Structure-Based Drug Design (SBDD)
SBDD is a target based method, it is used when the receptor structure and its mechanism is known while the ligand is not necessarily known. The first step to SBDD is to identify the binding site by evaluating the chemical properties of the molecule using computational tools to predict the interactions between the drug ligand and the protein. Macalino lists the protein structure revelation tools as X-ray crystallography, nuclear magnetic resonance (NMR), cryo-electron microscopy (EM), homology modeling, and molecular dynamic simulations.² There are several websites such as Protein Data Bank(rcsb.org) that share the three-dimensional structure of the proteins along with the computer modal drawings. After using the protein libraries and other tools to find a proper site for binding, the next step is docking and scoring. Molecular docking “predicts possible binding modes of a compound in a particular target binding site and estimates affinity based on its conformation,”.² This step is to find the type and orientation of bonds that will form between the target and the ligand molecule. For this purpose, there are several screening functions such as force field functions and empirical-based functions. Force field-based functions use the relations derived by the laws of physics (Figure 1)³ to elucidate the interactions between the target-ligand complex, while the empirical-based functions use experimental knowledge of the chemical molecular affinity.
Ligand Based Drug Design
Unlike the SBDD method, Ligand Based Drug Design is independent of the knowledge of receptor structure and its mechanism. However, it relies on the knowledge of molecules that bind to the target. This method depends on the assumption that similar compounds regarding chemical properties show similar biological effects. Quantitative Structure-Activity Relationship(QSAR) technology works on this principle and analyzes the relationship between similar compounds and their biological activities using computational statistics, training on data, and machine learning. According to Cleber C Melo-Filho, there are restrictions to QSAR listed as such: there have to be at least 20 comparable compounds with known bioactivities, proper selection of compounds for training and test sets.⁴ QSAR technology is developing by adding new quantitative features with the rise of computational power and the advancements in machine learning technology. It now is capable of interpreting the orientation in the target site, flexibility, and solvation of the ligand-target complex alongside the 3D strategies. After scanning through the compounds the elected ones are modeled using “Pharmacohore modeling” and some of them get restricted due to the fact of impossible 3D geometry or molecular limitations.
The next vital step is to choose the fittest compound and using the Docking and Modeling methods number of candidate molecules decreases and their eligibilities are ranked by some filters. Using automatic analysis tools like AuPosSOM and evaluating the manufacturing process the compound is chosen regarding the biological activity and cost-time efficiency.
Important Drugs That Designed with CADD
Since Computer-Aided Drug Design (CADD) technique was developed, an important number of drugs have been designed and used for therapeutic aims. This technique is quite useful to develop drugs for intractable diseases. In the following paragraphs, we will focus on the drugs that developed with CADD, such as HIV drugs, antihypertensive drugs, anticancer drugs, etc. Some of the clinically approved drugs designed by CADD has expressed in Figure 2.⁵
Captopril (Figure 3)⁵ is a competitive inhibitor, which is binding to the active site of the enzymes and inhibits the activity, of angiotensin-converting enzyme (ACE) which is a carboxypeptidase.⁵ This enzyme converts the angiotensin I (ATI) to angiotensin II (ATII). ATII regulates blood pressure. Captopril is used as a cure for hypertension. At first succinyl proline (Figure 4)⁵ was proposed as an ACE inhibitor. Later than, Cushman & Ondetti proposed that due to ACE contains zinc cofactor; if the mercapto functional group is replaced with an acid group of succinyl group, it will coordinate better with zinc and consequently, it will inhibit better. After little more modifications with computational drug design, captopril has been discovered and pioneered as a medical treatment.
Reverse transcriptase is an important enzyme for the human immunodeficiency virus (HIV)-1 life cycle. It produces the DNA from the virus’ RNA. So, if this enzyme’s interaction with its substrates is intervented, it can stop the reproducing of the virus. These drugs are called the “antiretroviral drugs”. Currently, there are 26 antiretroviral drugs that are approved by FDA. Long-term applications of retroviral therapies cause disorders such as insulin resistance, cardiac disorders, etc. and develop drug resistance in the virus when viral suppression is not maintained. Computational drug design is used for improving drug activity and minimizing the issues described above. There are four types of HIV drugs that are categorized by their inhibiting sites. Figure 5 shows these categories and drugs.⁵
Nolatrexed dihydrochloride (Thymitaq®, Agouron)
Dr. Webber and his group proposed that if the thymidylate synthase (TS), the enzyme that catalyzes the conversion of deoxyuridine monophosphate (dUMP) to deoxythymidylate (dTMP), is inhibited, cancer cells will die. If the dTMP formation is prevented, TS enzyme activity will be stopped, and cells will die due to thyminless circumstances. X-Ray structure data and molecular modeling methods lead to the design of four quinazolinones.CB3717 (Figure 6)⁵ has remarkable inhibitory activity on human and E. coli. After calculating the docking sites, interactions between amino acids and the drug, etc. caused the modifications of the drug. The final drug was nolatrexed dihydrochloride (Thymitaq). This drug had important results against cancer in E.Coli and human. This drug is using in the treatment of liver cancer. Nolatrexed dihydrochloride is showed in Figure 7.⁵
Dorzolamide is a medication used for glaucoma. Glaucoma is the disorder of high pressure in the eye. Dorzolamide inhibits the carbonic anhydrase enzyme. Carbonic anhydrase enzyme converts H2CO3 to HCO3- and H+. H+ is exchanged for sodium in the eye cells and causes the secretion of aqueous humor. In other words, dorzolamide inhibits carbonic anhydrase in order to decrease the amount of aqueous humor. Dorzolamide has shown in Figure 8.⁵
Luminespib is a drug candidate for the treatment of cancer. Hsp90 (heat shock protein 90) is a chaperone protein that assists other proteins to fold properly, stabilizes proteins against heat stress, and aids in protein degradation. It also stabilizes several proteins required for tumor growth. So, in order to prevent tumor growth, The Institute of Cancer Research and drug companies collaborated to design an inhibitor for the Hsp90. Luminespib is shown near in Figure 9.⁵
Current Approaches and Future Predictions of CADD
Computer-aided drug discovery is “a promising attempt” which makes it possible to develop new, target specific drugs.⁶ Since the world is struggling with the Covid-19 pandemic, using computer-aided methods will save lots of time during the drug discovery process. Also, it could help scientists to find feasible compounds more easily. Moreover, the best potential drugs would be chosen from the approved drug list with computational techniques for drug repurposing in such a limited time.⁶ As Bashir Akhlaq Akhoon and his colleagues (2019) underline that drug repurposing could be a major development during sudden outbreaks of infections and for the treatment of new pathogens. Additionally, pharmacokinetic data could be used to pass preclinical and phase-1 studies for approved drugs. For instance, the antimalaria drug chloroquine was repurposed by Shiryaev et al. (2017) for Zika-virus treatment and prophylaxis.⁷
Another aim of the computer-aided drug design can be found in the rare-disease domain. A lot of companies aren’t doing research in the field of rare-diseases due to its expensive costs. Therefore, the majority of the patients are not able to afford the therapies. However, drug repurposing will solve this problem by decreasing the cost of research and increasing the chance of success as compared to classical drug studies.⁷
Other than rare-diseases, a recent study demonstrates that computational methods could be utilized for a novel treatment to cure cancer. Targeted cancer therapy, a novel treatment, will be a less harmful treatment choice for the patients who are suffering from cancer. In this research, computational techniques, which are homology modeling, molecular docking and refinement, are used for comprehending the linker-drug interaction and conceiving a better potential fit between antibodies and antigens.⁸
In years to come, personalized medicine will significantly change the treatment methods with the aid of computational drug design. Symptomatic therapy will be altered by genome-based therapy. Genetics molecular classification will substitute the common knowledge of clinical diseases.⁹ Treatments will be oriented to cure the root causes of illnesses rather than just curing the symptoms. By a pharmacogenomic test’s prediction from the patient’s genomic profile will help us assess the therapy response and guide the drug discoveries.⁹
According to Gisbert Schneider, it is inevitable that automated computer-assisted de novo design of new molecular entities (NMEs) will be considered important for future drug discovery. For instance, in reference to novel information, improved structure-activity relationship models can be developed that help to generate the next generation of molecular structures.¹⁰ It can be concluded that radical changes both transform the medicine and the pharmaceutical industry. With this transformation’s help, the time and cost of clinical trials and overall estimated cost of healthcare could be decreased. Additionally, the risk of adverse drug reactions could be minimalized and determining the ideal therapy would be easier.¹¹
Current Challenges of CADD
Despite all of the advantages and promises of CADD, computational methods including virtual screening and molecular docking still lack some precision and accuracy in some cases. Lack of precision and accuracy may occur because of multiple reasons. Some of these reasons are the miscalculation of binding affinity, the classification difference between binder/non-binder the incorporation of the water molecule, the lack of knowledge of interaction region or anchoring point and trying to dock without this knowledge, the existence of non-natural amino acid or building blocks in peptide‐like molecules, and the complication while transformation to smaller molecules.
The prediction of peptide binding flexibility to a target protein is important. The mobility of ligands is less after they are bonded, and therefore the power of binding forces also decreases. Nevertheless, recent results show that rotamer-counting approaches display a higher number of accessible stable conformation by ligands, because of the common assumption that the different ligands lose the same amount of entropy, which proved to be wrong.¹²
Different molecular docking methods (such as soft docking, ensemble docking, etc.) target different rates of flexibility which broadens the range of the applications. Virtual Screening and Geometry Prediction respond to different rates of flexibility which concludes that while Virtual Screening methods benefit from ensemble docking and selective docking, Geometry Prediction concentrates on more complex systems and realistic binding predictions, which are different approaches.¹²
Even though with molecular dynamics (MD) method, if enough information such as conformational flexibility, incorporation of salts, etc. is provided, natural environment can be simulated, and the limitation of docking result can be minimized; because MD demands supercomputing facilities, the conclusion is determined within a few microseconds, protein folding requires milliseconds to seconds to complete. Thus conformation samples are not adequate.¹³ Water molecules have been incorporated into ligands in order to secure more enhanced binding affinity. Although in a class of HIV protease inhibitors, this approach was successful, in some cases, the placement change of interfacial water does not enhance binding affinity.¹⁴ Also, in the reactions between protein and peptide, there are extended surfaces, that are considered in order to pinpoint the possible interaction areas, which is considered a challenge.¹³ When a peptide-like molecule is modified, it often includes elements such as non-natural amino acids, building blocks, etc., hence they are not considered as short proteins. Because of that, sequence-based tools or knowledge-based protein force fields cannot be used.¹³ A pharmacophore model of the protein-peptide complex that shows hydrogen bonding, electrostatic and hydrophobic interaction, etc. is considered as a good model, which can be used to design smaller molecules that represent peptides. The tools to develop such models are up to be developed.¹³
Currently, these challenges and more limit the capabilities of computational drug design. However, promising approaches are being developed and within the near future, the computational drug design method promises very significant outcomes. As the popularity of computational drug design method rises among scientists, the drug developing process accelerates and accordingly some illness-related loss of lives are prevented. CADD’s future might be brighter than our expectations.
Arda Görkem Tokur, Hakancan Öztürk, Derin Aktaş and Ece Türksever.
1-Global pharmaceutical market size 2001–2019, Statista online. https://www.statista.com/statistics/263102/pharmaceutical-market-worldwide-revenue-since-2001/. (accessed: 15-Dec-2020).
2-S. J. Y. Macalino, V. Gosu, S. Hong, and S. Choi(2015), Role of computer-aided drug design in modern drug discovery, Arch. Pharm. Res., vol. 38, no. 9, pp. 1686–1701.
3-NAFlex: Nucleic acids flexibility, Irbbarcelona online http://mmb.irbbarcelona.org/NAFlex/help.php?id=md.(accessed: 14-Dec-2020).
4-C. C. Melo-Filho, R. C. Braga, and C. H. Andrade(2014), 3D-QSAR approaches in drug design: perspectives to generate reliable CoMFA models, Curr. Comput. Aided Drug Des., vol. 10, no. 2, pp. 148–159.
5-Talele, T., Khedkar, S., and Rigby, A. (2010) Successful applications of computer aided drug discovery: Moving drugs from concept to the clinic. Current Topics in Medicinal Chemistry 10, 127–141.
6- Kawthar Mohamed, Niloufar Yazdanpanah, Amene Saghazadeh, Nima Rezaei,Computational drug discovery and repurposing for the treatment of COVID-19: A systematic review,Bioorganic Chemistry,2020,104490,ISSN 0045–2068,
7- Bashir Akhlaq Akhoon, Harshita Tiwari, Amit Nargotra, Chapter 3 — In Silico Drug Design Methods for Drug Repurposing, Editor(s): Kunal Roy, In Silico Drug Design,
Academic Press, 2019, Pages 47–84, ISBN 9780128161258, https://doi.org/10.1016/B978-0-12-816125-8.00003-1
8-Melo, R., Lemos, A., Preto, A. J., Almeida, J. G., Correia, J., Sensoy, O., & Moreira, I. S. (2018). Computational Approaches in Antibody-drug Conjugate Optimization for Targeted Cancer Therapy. Current topics in medicinal chemistry, 18(13), 1091–1109. https://doi.org/10.2174/1568026618666180731165222
9-Ginsburg, G. S., & McCarthy, J. J. (2001). Personalized medicine: revolutionizing drug discovery and patient care. Trends in biotechnology, 19(12), 491–496. https://doi.org/10.1016/s0167-7799(01)01814-5
10-Schneider, Gisbert. (2018). Future Perspectives of Computational Drug Design: Achievements and Future Opportunities. 10.1002/9783527806539.ch6m.
11-Valeska, Margareta & Adisurja, Gabriella & Bernard, Stefanus & Wijaya, Renadya & Hafidzhah, Muhammad & Parikesit, Arli Aditya. (2019). The Role of Bioinformatics in Personalized Medicine: Your Future Medical Treatment. 46. 785–788.
12- Dinler A Antunes , Didier Devaurs & Lydia E Kavraki (2015) Understanding the challenges of protein flexibility in drug design, Expert Opinion on Drug Discovery, 10:12,1301–1313, DOI: 10.1517/17460441.2015.1094458
13- Rubinstein, M., and Niv, M. Y. (2009) Peptidic modulators of protein-protein interactions: Progress and challenges in computational design., Biopolymers 91, 505–513.
14- Akhoon, Bashir & Tiwari, Harshita & Nargotra, Amit. (2019). In Silico Drug Design Methods for Drug Repurposing. 10.1016/B978–0–12–816125–8.00003–1.