AI in Drug Discovery

Artificial Intelligence
• Artificial intelligence (AI) plays an important role in daily life.
Significant achievements in numerous different areas such as
image and speech recognition, natural language processing etc.
have emerged.

• In pharmaceutical research, novel artificial intelligence

technologies received wide interest, when deep learning
architectures demonstrated superior results in property prediction.

• In the meantime, the scope of AI applications for early drug

discovery has been widely increased, for example to de novo
design of chemical compounds and peptides as well as to
synthesis planning.
 Drug Discovery
• In drug discovery, clinical candidate molecules must
meet a set of different criteria. Next to the right potency
for the biological target, the compound should be rather
selective against undesired targets and also exhibit good
physicochemical as well as ADMET properties
(absorption, distribution, metabolism, excretion and
toxicity properties).
Applications of AI and ML in Drug Discovery include:
1. In High Throughput screening.
In the pursuit of reducing the cost of HTS while increasing its efficiency
and predictability, virtual screening has been employed successfully to
optimize the hit finding workflow. Virtual screening can be divided into
two broad categories:
structure-based virtual screening (SBVS) and ligand-based virtual
screening(LBVS).
2. Prediction of physiochemical properties of the drug including
ADMET properties, Lipophilicity, aqueous solubility and intrinsic
permeability.
3. Drug repurposing.
4. Precision Medicine and Big Data-Driven Drug discovery.
• A good drug target needs to be relevant to the disease phenotype
and should be amenable to therapeutic modulation. At the same
time, you need to have a good therapeutic window to assure that
any therapeutic modality aimed at the target will not cause side
effects by disrupting the physiological function of the target in
healthy tissue.

• A drug target is a molecule in the body, usually proteins, peptides

and nucleic acids, that are associated with a particular disease
process and that could be addressed by a drug to produce a
desired therapeutic effect.

• In drug discovery, clinical candidate molecules must meet a set of

different criteria. Next to the right potency for the biological target,
the compound should be rather selective against undesired targets
and also exhibit good physicochemical as well as ADMET properties
(absorption, distribution, metabolism, excretion and toxicity
properties).
Properties of a promising drug target include:

• The target has a confirmed role in the pathophysiology of a disease and/or is

disease-modifying.

• Target expression is not evenly distributed throughout the body.

• The target’s 3D-structure is available to assess druggability.

• The target is easily ‘assayable’ enabling high-throughput screening.

• The target possesses a promising toxicity profile, potential adverse effects can
be predicted using phenotypic data.
ADMET properties:

1. Absorption is the crucial first step in which drug molecules are absorbed
into the bloodstream from the absorption site through one or more
membranes.
Human intestinal absorption (HIA) is one the most influential ADMET
properties and directly affects oral bioavailability. A number of
computational classification and regression models have been introduced
to predict HIA based on data from in vivo and in vitro experimental assays.

2. Metabolism-The chemical transformation of xenobiotics in the liver (as

well as other tissues and organs) is usually an enzyme mediated process,
and considering the versatility of metabolic enzymes, the diversity of the
substrates and inhibitors, and the complexity of metabolic processes, has
proven challenging to predict. Metabolism determines the fate of a drug in
the body and consequently influences its safety profile and therapeutic
effect.
In a broad sense, the applications of AI in the field of predicting
metabolism are divided into three main categories:
• predicting the location of sites of metabolism
• predicting specific isoforms responsible for metabolism
• predicting metabolic pharmacokinetics.

3. Distribution is the process by which drug molecules

diffuse or are actively transported from the bloodstream to body tissues,
particularly the tissue(s) where the action is expected to occur.
Of the various factors that determine the drug distribution, “physical
barrier” tissues are often investigated (e.g., the blood brain-barrier and
Placental barrier). The blood brain barrier plays key roles in protecting the
brain from changes in Blood composition (e.g., hormones,
neurotransmitters, and other xenobiotics) and in brain homeostasis. In
recent years, machine learning methods have been utilized to minimize
the number of high-cost, time consuming, and experimentally challenging
blood-brain barrier permeation experiments required in development
efforts.
4. Excretion
Drug excretion is the process of removing drug molecules from the
body, which directly affects the plasma concentrations of drugs and
their metabolites, and plays an important role in the design of drug
regimens.
Excretion is a complex process that involves several elimination
pathways, such as billiary and renal excretion, each of which is
composed of a number of different processes.

5. Drug toxicity refers to the adverse effect on an organism or a

substructure of the organism (e.g., cells and organs) due to the action or
metabolism of a compound. Measurements of toxicity are one of the
most important and challenging steps involved in the drug discovery and
development cycle.
Reliable, high-throughput assays are expensive; therefore,
computational models that provide a rapid, inexpensive, and reliable
alternative to large-scale in vivo and in vitro bioassays are in high
demand. Advanced AI algorithms have opened up new avenues for
predicting toxicity, with quite a few publications discussing the potential
roles of AI in this field, such as in predicting toxicological end points or
classifying toxic compounds. Most of these algorithms are roughly
divided into similarity and feature-based approaches.
Dealing with the dataset
Possible types of inputs and outputs for the model:
input X :
1. sequence (e.g., primary structures of biomacromolecules, biomedical texts for
extracting drug−drug interactions; SMILES strings, time-series data)
2. fixed-size vector (e.g., bitstrings, vector of real-valued numbers)
3. molecular structure graph
output Y :
1. binary values (binary classification tasks)
2. integer values (multiclass classification tasks)
3. real-valued numbers (regression tasks)
4. fixed-size vector (e.g., a generated fingerprint vector)
5. single data column corresponds to single-task learning; multiple
6. data columns correspond to multitask learning
sequential data (e.g., SMILES string, amino acid sequence)
In most deep learning applications the encoding of a particular molecule is
represented by a fixed-length vector, and thus the success of deep learning
methods strongly depends on good encoding functions for mapping molecular
structural information into vectors.

The input data types used include molecular descriptors and fingerprints in the
form of fixed length vectors.

The problem is that these vectors can be large and this leads to many
parameters that have to be learnt but the input size is small.

One method proposed is that with the help of differentiable neural networks
whose inputs are molecular structure graphs and the outputs are various
types of graph finger prints. The graph finger prints serve as inputs for the
main algorithm.

Another problem is that the reverse of the above process, i.e., the construction of
a molecular structure from input vectors is even more difficult because one
finger print representation can correspond to multiple chemical molecules.
This is overcome by using GAN or some deep generative models along with
the combinations of SMILES strings as molecular representations.
 Drug Repurposing
• In 2004, Ashburn and Thor wrote their landmark article ‘Drug repositioning:
identifying and developing new uses for existing drugs’, in which they outlined the
opportunities for drug repositioning. They stated that: ‘the process of finding new
uses outside the scope of the original medical indication for existing drugs is also
known as redirecting, repurposing, repositioning and reprofiling’.
• Ashburn and Thor explained that the development risks would be reduced,
because drug repositioning candidates could be developed quicker owing to the
use of existing knowledge about the drug.
• Serendipity is one of the many factors that may contribute to drug discovery.
One of the examples include the disaster of Thalidomide where it was initially
used by Pregnant woman for treating morning sickness which let to it’s ban
because of skeletal deformities and underdeveloped body parts of infants
(known as Phocomelia). Later it was discovered that it can be used for the
treatment of leprosy. Some other drugs that were discovered serendipitously
include aniline purple, penicillin, lysergic acid diethylamide, meprobamate,
chlorpromazine, and imipramine.
 Use Of Drug Repurposing
• Currently, pharmaceutical companies face a challenging economical and
societal environment that requires them to continuously look for strategies to
improve their capacities to develop original drugs at reduced cost.
• Within this context, the pharmaceutical community considers that finding novel
indications and targets for already existing drugs, a method called ‘drug
repurposing’, can compensate for the lack of technical efficiency of the
traditional drug discovery approaches that results in a high failure rate and
continual decline in the number of new approved small-molecular entities
released by pharmaceutical industry pipelines.
• The major advantages of a drug-repurposing approach are that the preclinical,
pharmacokinetic, pharmacodynamic and toxicity profiles of the drug are
already known, reducing the risk of compound development. Thus, the
compound can rapidly translate into Phase II and III clinical studies, resulting in
a decreased development cost [6], a better return on investment and an
accelerated development time.
• drug-repurposing methods could help to find cures for orphan diseases .
Indeed, there are 400 million people worldwide affected by such diseases,
but with current research and development costs, it is impossible to
develop de novo therapies for each of the 5000–8000 orphan diseases
identified so far.
• The development of a drug for the treatment of new diseases might
involve new patient populations, dosage forms or routes of
administration. However, wording like a new ‘application’ does not
necessarily mean the treatment of a new disease. It can also relate to the
development of a drug for new patient populations, new dosage forms,
routes of administration or line of treatment. For example, fentanyl was
approved in the 1980s as solution for infusion and nowadays is authorised
as nasal spray, transdermal patch, buccal tablet and lozenge for
oromucosal use.
 AI In Drug Repurposing
• Drug development and trials in animals and humans is a time‐consuming and
expensive process. In general, the whole process for developing a new FDA‐approved
drug requires 10–17 years of period and the tremendous cost of $2.6 billion.301
• we classify the ML applications for drug repositioning into the following three
categories: (i) Similarity‐based methods that employ different types of classifiers like
logistic regression SVM RF KNN and CNN (ii) feature vector‐based methods that
utilize supervised and semi supervised learning algorithms, and (iii) network‐based
methods that mainly use semi supervised learning algorithms (e.g., Laplacian
regularized least square, label propagation, random walk,).
• Drug Repurposing under the Unsupervised Learning Paradigm Unsupervised
computational tools for drug repurposing have several distinct advantages over the
supervised tools. They do not require any a priori knowledge about drug labels for
training data. By discovering hidden patterns in data, these tools can group and
identify compounds with similar therapeutic effect. A variety of unsupervised
methodologies exist for discovering potential clinical applications of drugs, the most
popular of which is the clustering algorithm.
The difficulty in computational drug repurposing lies in the
rarity of known associations between entities (drug, target, and
disease) and the myriad of unknown associations that remain
to be predicted. The number of known associations from
public databases is still quite small compared to the entire
pharmacological space. In response to this challenge, semi-
supervised learning has been proposed as a novel approach
for drug repurposing due to its reported efficacy in working
with small labeled training sets in other disciplines. It tackles
this issue by utilizing both the small amount of available
labeled information and the abundant unlabeled information to
create models with a balance of accuracy and generalizability to
new instances.
 AI/ML APPLICATIONS IN CNS DRUG DISCOVERY
• CNS diseases are a group of neurological disorders that impose a significant
economic and social impact. Development of new drugs for CNS diseases poses
unique challenges compared to other diseases, including the complexity of brain
anatomy and function, incomplete understanding of the biology of the complex
nature of CNS diseases and the presence of BBB.
• In addition to the difficulties in CNS target identification, designing new
molecules with the ability to penetrate the BBB is also a major obstacle. The
role of the BBB is to protect the brain from variations in blood composition
(e.g., hormones, amino acids, and potassium) and circulating pathogens. It
consists of capillary endothelial cells that are lined by the basal lamina made
from structural proteins (i.e., extracellular matrix proteins collagen and
laminin), pericytes, astrocytic endfeet, and microglial cells.327 This biologic
membrane allows the uptake of water, glucose, and essential amino acids, the
efflux of small molecules and nonessential amino acids from the brain to the
blood and the passage of some molecules by passive diffusion.
• construction of BBB permeability predictive models, researchers have
employed various supervised learning approaches, such as SVMrecursive
partitioning (RP), Gaussian process, DT, KNN, linear discriminant analysis,339
consensus classifier, and ANN. All of these methods were developed to
process physical and chemical features, which mainly include molecular
weight, hydrophilicity (ClogP), lipophilicity (ClogD),topological polar surface
area, acidic and basic atoms numbers, hydrogen bond donors and acceptors,
water‐accessible volume, flexibility (rotatable bonds), van der Waals volume,
and ionization potential.
 AI/ML IN CLINICAL PHARMACOLOGY
CHALLENGES AND OUTLOOK
• Currently, the key challenges in drug discovery are related to chemical
structures (e.g., toxicity, side effects, or even intellectual property), choosing
the right drug target or the appropriate dosage for a specific patient
subpopulation.

• The pharmaceutical industry does not typically share pharmacokinetic and

pharmacodynamic measurements for most of the candidate drugs or their
combinations, unless the drugs are approved for human use. Indeed, unlike
other areas of research and development, only a small fraction of the drug
discovery data overall is available for training AI/ML models.
We can safely state that recent AI/ML developments are likely to impact clinical
pharmacologists at all levels in the next decade:
• In teaching, the use of data-intensive methodologies for example, literature
searching and processing, as well as interactions with on-line predictive AI/ML
models, are likely to increase.
• In framing policy, AI/ML methods are anticipated to influence healthcare in most
countries.
• For the proper use of medicines, most “digital natives” (i.e., patients and
scientists who started to use computers/tablets/smartphones from an early age)
are seeking healthcare and drug information via social media and web-based
platforms. Predictive analytics (e.g., anticipating query terms and personalized
searches) for drug-related information are already available. Advanced AI/ML
systems, tailored to an individual’s medical and genomic history are around the
corner.
• Although most AI/ML developments are expected to benefit patients and
scientists alike, we do anticipate that these trends will not replace clinical
pharmacologists in the next decade, and clinical pharmacology in general will
be disrupted slower than the other fields of drug discovery and development.
• There is, however, a strong incentive for clinical pharmacologists to adapt, to
monitor such trends, and to become cognizant on the usage and perils of
AI/ML systems, particularly for those that influence their daily practice. There
needs to be a community and regulatory effort for open data sharing to make
larger number of preclinical and clinical pharmacology data available for the
ML community to use in order to accelerate the AI/ML progress in the area.