SARS-CoV-2 virus is still emerging around the world. The number of infected people continues to overgrow, and still no definitive therapy that have been approved for effective treatment. Finding broad-spectrum inhibitors that may reduce the effects of human coronavirus infection remains a challenging research focus. Given the time-consuming nature of anti-viral drug development and registration, drug repurposing is one of shortcut to cure the disease. For most of these drugs that have been prepared, the medication has sufficient experience and dosage, and their safety and ADME situation are well known.
Despite still continuing the research of conventional medicine, Indonesia that has mega biodiversity, has potential herbal compounds as SARS-CoV-2 inhibitor for an alternative. In order to get the potential herbal compounds by using computational approach, we have to be careful about the research methods. We should not use our self-preferential in certain herbal, because it would lead to subjective decision on the research results. Especially, when the computational approach only use molecular docking method. While molecular docking is a powerful tool for pharmaceutical research after decades of development, there is a limitation about docking accuracy due to relatively simple scoring functions. Additionally, entropic factors are generally not captured well by scoring based on a single structure. As a result, structure-based ligand screening by docking often generates a large number of false positive hits [74].
To minimize the false positive hits by conducted research with molecular docking only, we tried to use two different approaches in generating prediction model before we did the virtual screening on HerbalDB compounds. In this study, we used machine learning and pharmacophore modeling methods that are complementary to each other to generate more accurate prediction model. The machine learning approach was used to perform big data analysis DTI dataset that collected from literature and public domain database. This approach used pharmacological features that obtained by integrating both the chemical space of compounds and the genomic space of target proteins. Compared with the ligand-based pharmacophore methods, the machine learning approach can be used to predict the DTI with insufficient known ligands [75]. Thus, using these two approaches in the methodology will lead us to increase the confidence level of the predicted compounds candidates.
Based on the virtual screening on HerbalDB using two prediction approaches, we got 14 compounds that overlap from two method results. We continued to analyse with molecular docking to confirm each compound interaction with 3CLpro protein of SARS-CoV-2. From molecular docking analysis, we got six potential compounds, i.e Hesperidin, Kaempferol-3,4'-di-O-methyl ether (Ermanin); Myricetin-3-glucoside, Peonidine 3-(4’-arabinosylglucoside); Quercetin 3-(2G-rhamnosylrutinoside); and Rhamnetin 3-mannosyl-(1–2)-alloside, that predicted could inhibit the 3CLpro protein of SARS-CoV-2.
After we got this result, we further checked the previous studies to find the biological activities of each compound. So that this research can be useful for community, we also tried to find from commodity crops. One of the commodity crops in Indonesia is Guava (Psidium guajava) that can be harvested continuously in one year. In Indonesia, production of guava in year 2018 is 230697 tons, with growth rate from year 2017 to 2018 is 15.06% [76]. Guava is consumed not only as food but also as folk medicine in subtropical areas all over the world due to its pharmacologic activities. Guava is well known has several flavonoids compounds i.e myricetin, quercetin, luteolin, kaempferol, isorhamnetin [77], and Hesperidin [57]. These compounds were also shown in our result, although without the aglycones.
Luteolin is known as furin protein inhibitor [78] which is predicted to be one of the enzymes that break down Corona virus S (spike) protein as in MERS into units S1 and S2 [79]. In the S1 unit, there is a Receptor Binding Domain (RBD) where the ACE2 peptidase binds so that the virus can bind to the host [79]. Hesperidin / Hesperitin compounds in the in silico study are known to inhibit RBD domain binding of the SARS-COV-2 Spike protein with ACE2 receptors in humans so that it is predicted to potentially inhibit the entry of the SARS-COV-2 [5]. It is also known that luteolin is a neuramidase inhibitor as well as oseltamivir which is currently one of the drugs used in the CDC protocol. Hesperitin (the form of hesperidin aglycone) and Quercetin are known to also act as inhibitors of 3Clpro [80, 81]. Other compounds in guava such as myricetin are known to act as SARS coronavirus helicase inhibitors [82]. The kaempferol has the potential to be a non-competitive inhibitor of 3CLPro and PLpro as well as quercetin [83]. Another interesting thing is kaempferol acts as a modulator of autophagy, both as an inducer and inhibitor, both of which can be utilized in strategies to inhibit the SARS-COV-2 virus.