Possibilities of Artificial Intelligence (AI) in Drug Discovery

In this article, we’ll go through the possibilities of Artificial Intelligence (AI) in drug discovery. Drug discovery is a crucial aspect of the health industry. The main goal behind it is to bring forth a new compound that is proven to have a therapeutic effect to its users. Although our understanding of biological systems has progressed to a large extent, drug discovery is still a very long, inefficient and capital-intensive process. Furthermore, the rates of drug discovery are still shockingly low with barely any sign of improvement (66 of the 98 companies studied by Forbes launched only one drug this decade.)

The Process of Drug discovery

The process of drug discovery generally goes through four main phases:

• Target selection and validation
• Compound screening and lead optimization
• Preclinical studies
• Clinical trials.

Discovering a drug starts with target identification and validation. A drug target refers to a molecule in the body that can interact with a potential drug compound to produce a clinical effect. After identifying a target, researchers will have to narrow a variety of compounds to one particular compound that could potentially become a drug. 

As it is impossible to predict which chemical structures will have both the desired biological effects and the properties needed to become a viable drug, the process of developing a compound into a potential drug is extremely expensive and time-consuming. The average cost for research and development of a successful drug is estimated to be $2.6 billion. The overall probability of clinical success is estimated to be less than 12%.

Additionally, even if a new drug candidate shows potential during testing in laboratories, this drug might fail when it is moved into clinical trials. Less than 10% of drug candidates make it to market following Phase I trials.

The possibilities of AI in Drug discovery

From these numbers, it is evident that something needs to be done to improve the efficiency of the drug discovery process. This is why incorporating cutting-edge technology like Artificial Intelligence is important. According to a report from Bekryl, AI has the potential to offer over US$70 billion in savings for the drug discovery process by 2028.

This is why several pharmaceutical and biotech companies have started heavily investing in AI to improve their potential ROI. The size of the libraries being used to screen for new drug candidates is steadily increasing. It is extremely difficult for researchers to review everything on their own. In such cases, use of AI in drug discovery process can be very beneficial. However, this is not the only application of such technology in the world of drug discovery. Some other examples of where AI can be used include- 

• AI for primary drug screening

AI technology can quickly and accurately recognize images containing distinct objects or features. Recognizing images by manual visual analysis is a very hectic job and becomes very inefficient during the analysis of big data. This is why using AI-based computing technologies can be very beneficial. For cell target classification or diagnosis, the AI model needs to be trained to rapidly and automatically identify the different features of cell types.

For example, to classify breast cancer cells, the cell images are segmented from the background by varying the image contrast. Tamura texture features and wavelet-based texture features are then extracted, and principal component analysis (PCA) is used to reduce the dimensions of the extracted features. AI-based methodologies are then trained to classify different cell types. Among the tested methods, the least-square support vector machine method shows the highest classification accuracy (95.34%)

• Drug target identification and validation

The use of AI technology in identifying targets helps researchers to properly analyze all the relevant evidence so as to gain a better understanding of the disease and its underlying biology. AI can synthesise data, and then come to conclusions regarding the best targets. This allows researchers to make better decisions about which targets are most likely to succeed.

In December 2019, Gatehouse Bio announced a partnership with AstraZeneca to explore the identification of new targets for respiratory and cardiovascular diseases using its AI-powered platform. Gatehouse uses AI to identify novel small RNA (sRNA) signatures and molecular pathways correlated with and potentially driving disease.

Another example for a company that uses AI technology for drug target identification and validation is medicine startup GNS Healthcare. They partnered with Genentech to find and validate potential cancer drug targets by analyzing data from sources such as electronic medical records and next generation sequencing. 

• Processing biomedical, clinical and patient data

AI models help researchers deal with large volumes of biomedical and patient data. They can also provide intuitive intuitive insights about drug candidates and identify novel pathways, targets and biomarkers by reviewing this data. Santen Pharmaceuticals, a Japanese leader in the ophthalmic field, entered a strategic research collaboration with TwoXAR.

TwoXAR used its proprietary computational drug discovery platform to discover, screen, and prioritize novel drug candidates with potential application in ocular indications, with a specific focus on glaucoma.The company had to screen large catalogs of molecules, associated with known data, such as protein structures, binding affinities etc. The obtained data is then linked with molecular changes in glaucoma to derive unique disease-drug associations.

• Planning chemical synthesis

Retrosynthesis is an advanced method for designing organic synthesis. After a particular molecule has been virtually screened for its potential bioactivity and toxicology profile, researchers begin to search for an optimal chemical synthesis pathway to synthesize the drug candidates. This process is very difficult and inefficient. Despite possessing the knowledge of several transformation steps, it is not guaranteed that novel molecules can be efficiently synthesized because of novel structural features or conflicting reactivities

ML models trained on empirical data can now be used in the following ways-

1. To predict the probability of a transformation at a particular branching position.
2. To guide the selection of the random steps.

At  each transformation step, the molecule (or an intermediate) can be linked to specific precursors via a predefined transformation rule. AI algorithms can be trained from the literature regarding the yields and costs of these transformation rules, and can then predict the most feasible retrosynthesis pathway for a given molecule

• Drug optimization and repurposing

Drug repurposing is being used by many pharmaceutical companies as it provides a high value approach that presents less risk of unexpected toxicity or side effects in human trials and R&D spend. AI can be used to provide better insights on the polypharmacology of drugs to improve drug development success rates by identifying offset targets and unwanted toxic effects, while providing opportunities for drug repurposing. 

In 2016, Astellas Pharma signed a research deal with NuMedii to conduct drug repurposing projects using machine learning techniques. NuMedii’s big data resource comprises several human, biological, pharmacological and clinical data points, normalized and annotated. The company then uses neural network-based algorithms to find novel drug candidates, and biomarkers predictive of diseases, and repurpose existing drugs or drug candidates towards other medical indications.

•  AI for target based and phenotypic drug discovery

 A phenotypic screen doesn’t rely on a known drug target, but instead aims to identify molecules that alter a cell’s phenotype. Emerging methods using transcript quantification, public databases (chem/bioactivity profiles, ontologies, image-based screening results), combined with machine-learning tools, are providing innovative and alternative screening strategies by augmenting phenotypic screening results.

In 2017, GSK signed a drug discovery collaboration with Exscientia to identify small molecules for ten selected targets across undisclosed therapeutic areas. Using a rapid “design-make-test” cycle, Exscientia was able to design new molecules using an AI-system, employing phenotypic and high content screening data. The system could also assess the molecule’s potency, selectivity and binding affinity towards specific targets. 

• Predictive biomarkers

ML-based biomarker discovery and drug sensitivity predictive models have been proven to help boost clinical trial success rates. They can also improve the understanding of the drug action mechanism and to identify the appropriate drug for patients. Late-stage clinical trials take a lot of time and money to conduct, so it can be extremely advantageous to build, validate and apply predictive models earlier, using preclinical and early-stage clinical trial data. A translational biomarker can be predicted using ML approaches on preclinical data sets. After being validated using independent data sets (either preclinical or clinical), the model and its corresponding biomarker can be applied to stratify patients, identify potential indications and suggest the mechanisms of action of a drug

By applying AI in drug discovery process, the efficiency of the drug discovery process can be increased to a very large extent. We can see applications of AI in areas like cell sorting, cell classification, quantum mechanics, calculation of compound properties, computer-aided organic synthesis, designing new molecules, predicting the 3D structures of target proteins, and more. Businesses in the healthcare industries have also begun to recognize the benefits of implementing AI in their drug R&D process.