AI in Drug Discovery

Artificial intelligence (AI) has started to be used more extensively in society recently, with the pharmaceutical industry setting the standard for its advantages. In many respects, artificial intelligence (AI) has changed the pharmaceutical sector. The pharmaceutical business benefits from AI help across the whole life cycle of the product. AI has several applications in the pharmaceutical industry, from developing new medications to managing existing ones. The application of artificial intelligence (AI) is growing across various domains in society, chief among them being the pharmaceutical industry. This paper emphasizes the use of AI in the pharmaceutical business, with applications spanning from clinical trials, medication repurposing, and development and discovery to increased pharmaceutical production. These apps reduce the amount of labour that needs to be done by humans while also accelerating objective attainment [1]. Because computer-assisted design of drugs has a chance to expedite and reduce the expenses of the process of developing drugs, there is a great deal to explore in this emerging discipline. The process of finding new medicines is expensive and time-consuming; it usually takes ten to fifteen years for a drug to be commercialized. This field of study has been significantly impacted by CADD. Additionally, these technologies are used to handle enormous amounts of biological data has reduced the time and cost associated with the drug development process.

AI in Pharmaceutical Products

It is appropriate to imagine artificial intelligence (AI) having an effect in the creation of a medicine from the laboratory bench to the hospital because AI may assist with rational formulate drugs, support making choices, identify the most appropriate course of therapy for a patient, which includes customized pharmaceuticals, and manage the produced medical information to use it to guide subsequent drug development. Marketing professionals can get help from the quantitative and the making of choices AI platform E-VAI in deciding where to invest, turning around underperforming revenue, and assigning resources for the greatest increase in market share. The method creates predictive directions based on rivals, important stakeholders, and holds market share to forecast major trends in pharmaceutical sales. It does this by utilizing technologies in conjunction with a user-friendly interface [2].

Process of Drug Discovery

For a molecule to possess any kind of therapeutic value, it needs to be "druggable." The focus of drug development in the post-genomic age has switched, in order to develop newer drugs in the future, fundamental principles of design to molecules or new methods to bind, control, or destroy challenging biological targets. Historically, the pharmaceutical industry has concentrated on creating tiny compounds that are orally accessible and have predetermined targets, sometimes known as druggable targets. In accordance with Ro5, if the molecular weight is greater than 500 Da, there are a minimum of five hydrogen-bond donors (HBD > 5), as well as greater over ten hydrogen-bond acceptors, there is a greater chance of poor absorption or penetration. Since then, while developing developable compounds for drug discovery, Ro5 has been used as guidance. Although there has been some success in finding small molecule Ro5 drugs that interact with known "druggable" targets, there is a growing need for innovation to engage additional targets for transformational treatments. Therefore, from the beginning in the drug-discovery process, discovering and validating novel biological targets has taken centre stage. Historical data supports the application of AI and DL in this domain. Additionally, cutting-edge methods for data mining, curation, and management offer crucial backing for freshly created modeling algorithms. Identification of medicinal product targets, confirmation of targets, production of hits to leads, and optimization of the lead, both preclinical and clinical research are all steps in the drug research and development process [3][4].

Drug Designing

Drug discovery is the process of applying computational aspects, experimental in nature & clinical frameworks to identify potential new medicinal entities. The search for novel therapies remains a very difficult, time-consuming, expensive, and inefficient process with few new therapeutic discoveries, despite advances in technologies and the comprehension of biological mechanisms. Drug creation is the process of developing novel pharmaceuticals by utilizing a target in biology. In the simplest sense, drug design is the creation of molecules with complementary shapes and charges to the molecular target with which they interact and bind. In the big data era, bioinformatics methods and computational modeling techniques are often, but not always, used in drug design [5]. Preclinical research on animal and cell models, as well as human clinical trials, are all part of the process of developing and discovering new drugs. Afterward, the drug must receive regulatory approval before being put on the market. Finding screening hits, improving medicinal chemistry, and boosting the drug's effectiveness, stability in metabolic processes, oral accessibility, affinity, selectivity, and efficiency are every requirement in the process of developing new medications. Before conducting clinical trials, a molecule that satisfies all of these criteria will be found, and the drug development process will start [6].

Fig 1: Pipeline of Drug Discovery [15]

Types of Drug Designing

It is expected that computational techniques will be crucial in comprehending the distinct molecular recognition events of the target macromolecule, as candidate hits will aid in the creation of better leads for the target. Computer-aided pharmaceutical design (in silico) approaches have been widely used in the lead discovery and optimized lead discovery stages of the development of medicines against many targets over the years. Using rational drug design methodologies instead of traditional drug discovery procedures shortens the time and costs involved in the medication development process. The process can be applied to de novo identification and to create new inhibitors or to optimize the distribution, metabolism, excretion, absorption, and toxicity profile of compounds that have been found from different sources [7]. Developments in technology and computational approaches have made it easier to use in silico methods for discovery. These are two types: Ligand-based drug design (LBDD) and Structure-based drug design (SBDD) [7].

Fig 2: Basic principles and types of drug design [8].

Structure – based drug design

This method known as SBDD uses the drug receptor's knowledge to create an inhibitor of the target. The most popular methods for determining the structure of a receptor are experiments like NMR or X-ray crystallography. If one does not know the structure of the protein therapeutic target, one can utilize computational approaches like homology simulation and threading to anticipate the protein structure. Technique termed threading, or fold, is used to simulate proteins in the absence of protein structures. The steps involved in homology modeling proteins are as follows: Identifying homologous proteins with known three-dimensional structures that could serve as templates; aligning the sequences of the target and template proteins; building a target model based on the template's three-dimensional arrangement and alignment; and validating and improving the model. When crystal structures are unavailable, modeling homology has become the main alternative to get a three-dimensional image of the target over time [9].

De Novo Pharmaceutical Design

The Latin word "de novo" means "from the beginning". The structural characterization of a drug target's active site will reveal information about its binding characteristics. It is possible to create ligands that are specifically tailored to a target by using the makeup orientation. For de novo design techniques, computational tools that can identify possible chemicals and study the active site of proteins are widely used. These are the following steps:

Techniques for locating fragments:  To pinpoint the ideal positions of atoms or tiny pieces inside the activeregion.
Methods for connecting site points: Identifying specific places, or "site points," inserting fragments into the active site to occupy those spots with the right atoms.
Methods for connecting pieces:  After placing the fragments, "linkers” “scaffolds”, join them and maintain their intended alignment.
Methods for sequential buildup: Assemble a ligand piece by piece or atom by atom.
Whole molecule techniques: Different conformations of compounds are inserted into the active site to evaluate shape and/or electrostatic complementarily.  A unique class of approaches known as randomized connection methods combines elements of sequential building and fragment connection techniques with bond disconnection tactics and randomness-inserting procedures [10]. Many de novo techniques, particularly whole molecule techniques like docking, have been incorporated throughout time into computer modeling, chemistry, pharmacology, and molecular biology. Calculating electrostatic and salvation terms, which are essential for determining accurate binding energies, can be challenging and time-consuming. The complexity of algorithms is increasing, yielding ever-better estimates for certain parameters. Lastly, it is evident from the most recent research that the drug design process is now a crucial component of drug development initiatives [10].

Virtual Screening using Structure

One popular strategy for the lead identification stage is virtual screening, which is a supplement to test screening at high rates in a research study in order to increase the productivity and speed of the drug discovery and development process. This entails scoring and detailed molecular docking of each ligand to the target binding site. The chemicals in the screened databases are ordered in order to identify and test experimentally a limited subset for biological activity deemed suitable for a particular receptor. Numerous effective uses of molecular docking-based virtual screening have been documented. With the aim of enhancing the faster and efficacy of the process of finding and developing drugs, virtual screening is being employed more frequently in conjunction with high-throughput screening [11].

Fig 3: A basic drug discovery workflow using CADD [12].

Ligand – Based Drug Design

In the absence of 3D models of possible therapeutic targets, widely used method for lead selection and to find possible drug. These techniques can offer valuable information about the type of interactions that occur between the ligand and drug target as well as predictive models that are appropriate for lead compound optimization. With CADD methodologies, drug design can be achieved on both ligand- and structure-based approaches. When there isn’t an experimental three-dimensional structure available. Since there is presently no empirical structure associated with these ligands, the known ligand particles that bound to the therapeutic target are analyzed in order to understand the structural and physical features of the ligands that match with the anticipated therapeutic properties of those ligands. For ligand-based approaches, natural products or substrate equivalents that bind with the molecule being studied and deliver the intended pharmaceutical effect can also be used instead of well-known ligand molecules [13].

QSAR

The QSAR method modelling technique is used for ligand-based drug design. The association between a certain biological action is determined using a computer method known as QSAR. Similar structural or physicochemical features result in similar activity, according to the fundamental hypothesis of the QSAR technique. First, a group of substances known as lead molecules are found to possess the necessary biological capacity of interest. The developed QSAR model is then used to maximise the active compounds in order to achieve the greatest appropriate biological properties.

The intended activity of the anticipated chemicals is next investigated experimentally [13].

The foundation of the generic QSAR technique consists of the following steps:
1. Determine which ligands have the desired biological activity at values obtained from experiments.

2. Find and identify the molecular designations associated with the various structural and properties of the molecules under investigation.

3.  Determine the connections between descriptors of molecules and activity in biology that explain the difference in activity shown in the information set.

4. Analyze the prediction effectiveness and analytical reliability of the QSAR techniques model.

Fig 4: Workflow of QSAR [13]

In conclusion, developments in AI and DL offer a great chance for a logical approach to medication design and discovery that will eventually benefit humanity.

AI in Drug Designing

Computers to assist drug development has been propelled by artificial intelligence. This pattern is being filled in by advances in computer software and hardware as well as the growing application of machine learning, especially deep learning, in many scientific domains. The initial doubt that started to dissipate about the application of AI in the development of drugs has been advantageous. Deep learning has seen a relatively late comeback, but it has already resulted in an unparalleled proliferation of new modeling techniques and applications. Deep learning's constant advancements have already proven beneficial to many branches of the chemical sciences. Through examples, this opinion piece outlines some of the factors that have facilitated the growth of deep learning methodologies and, in certain situations, have even outperformed the state-of-the-art cheminformatics approaches. Specifically, structure-based modelling, de novo molecular design, synthesis prediction, and ligand-based quantitative structure-activity/property relationship (QSAR/QSPR) are covered.

Novel deep learning approaches, which are the foundation of ligand-based drug development, may make it easier to find new chemical entities.
Deep learning approaches that include chemical knowledge and incorporate multidimensional patterns into their creation show promise for conformation-aware and structure-based modeling. Integration based on rules and rule-free approaches enhance artificial intelligence's capacity to produce synthesizable bioactive compounds and discover uncharted chemical space. Multitasking, meta-learning, and explainable AI will open the door for a new class of prediction models that perform well in low-data regimes and are easier to understand.
Models of natural language processing will be widely used for forward synthesis prediction as well as retro synthesis. Further work will be focused on associated issues including reaction condition prediction. Some basic issues in drug development have only recently started to be addressed by deep learning-based methods. Some of the most difficult challenges will probably be solved with the help of methodological advancements such as hybridized new-generation design, other methodologies. The development of artificial intelligence-powered drug discovery will depend heavily on the development of models and the free sharing of data. [14].

Fig 5: Comparison of Classical and Modern method [14]

Fig 6: Comparison of Synthesis planning of classical and modern [14]