Natural products (NPs) are well known for their ability to interact with therapeutically important protein targets. Their structural variety and biological activity continue to be inspirations for the creation of small molecules and macrocyclic medicines. In the mid-1970s, NPs dominated the sources of innovative human therapies in the pharmaceutical drug pipeline. Two-thirds of the medicines were derived from unmodified natural products, analogues, or NP pharmacophores. Natural product research has fallen at most major pharmaceutical companies, despite being a proven source for current small-molecule drug discovery. The primary objections include the time-consuming dereplication procedure, complicated syntheses, and extracts that are unsuitable for high-throughput screening. Furthermore, many NPs have ADME and physicochemical features that go beyond the existing drug-like chemical space, such as high degrees of stereochemistry, fused ring systems, or rotatable links.
Chemoinformatics, bioinformatics, and other informatics-related fields have made significant contributions to NP-based medicine throughout the years. Artificial intelligence and machine learning algorithms have steadily made their way into natural product research, a recognised source of current small molecule medication discovery. In the early 2000s, AI applications mostly comprised the digitalization of organic molecules and dimensionality reduction techniques using principal component analysis to map the NP chemical space. The next decade saw the development of ML binary classifiers for predicting biological functions. Scientists have recently begun to utilise neural network topologies for genome mining and molecular design.
Using Deep Learning and Text Mining, a scientist could monitor the roughly 20,000 new compounds published in medicinal chemistry journals every year. Additionally, text mining has helped to curate, preserve, and even shed new light on the ancient pharmacopoeia written about medicinal plants from various parts of the world. Traditional medicine formulas from Chinese, Indian, Egyptian and other sources could be verified via comparison with modern studies reported in modern medicinal journals. Another technique called Natural Language Processing can help to construct a knowledge graph to represent complex relations of protein-chemical-disease found in many NP scientific studies.
Chemical and biomolecular databases are prevalent in informatics-related areas and are important to many AI applications. However, only about half of these databases allowed for substructure searches using at least one of the aforementioned molecular representations, and many lacked stereochemical information. As a result, the Steinbeck group in Germany created COCONUT (https://coconut.naturalproducts.net/), the world’s biggest repository of open-access natural products, which combined the structures and relevant information of over 400 000 non-redundant NPs. One can enter a newly created chemical SMILES structure to search it in the natural products space.
A novel chemical created via the drug-design process can be compared with closely similar molecules found in the NP domain. To do this, one can use the Natural Product Likeness Score (Naples, https://naples.naturalproducts.net/) which is also published by the Steinbeck group. Furthermore, the biological activity, including pharmacological effects, mechanisms of action, and toxic and adverse effects of a new chemical structural formula can be predicted using the Prediction of Activity Spectra for Substances (PASS). The interaction with the metabolic system and specific toxicity for drug-like molecules can also be predicted using PASS.