From identifying novel drug targets to optimizing the manufacturing process, AI is transforming the way small molecule drugs are discovered, designed, and developed. AI-driven drug development holds remarkable potential for accelerating the availability of life-changing therapies through small molecule drug development. Read on to connect the linkages between small molecule drug chemistry, the drug discovery and development process, and the application of AI to deliver safe and effective drugs for clinical use in patients.

What is a small-molecule drug?

In the realm of pharmaceuticals, small-molecule drugs play a pivotal role in treating a diverse range of diseases. These organic compounds possess low molecular weights, usually below 900 Daltons, and can exhibit remarkable therapeutic potential. Their composition primarily consists of carbon, hydrogen, nitrogen, oxygen, and sulfur atoms, forming the backbone of their molecular structure. 

The versatility of small-molecule drugs stems from their ability to be synthesized chemically or derived from natural sources. Their organic origin allows for the exploration of a vast chemical space, unlocking a plethora of potential drug candidates. Their compact size enables them to penetrate biological membranes efficiently, reaching their intended targets within the body.

A small molecule drug is designed to inhibit or activate a disease target based on pathway, i.e. the drug’s mechanism of action (MOA). The mechanisms of action for small-molecule drugs are as varied as the diseases they combat. They can interfere with enzyme activity, effectively inhibiting specific biochemical reactions. Alternatively, they can bind to receptors, modulating cellular responses and signaling pathways. Moreover, small-molecule drugs can influence gene expression, altering the production of proteins within the body.

The therapeutic applications of small-molecule drugs span a wide spectrum of medical conditions. They are commonly employed to combat life-threatening diseases such as cancer, cardiovascular ailments, neurological disorders and infectious diseases. Their ability to target specific molecular processes with precision makes them invaluable tools in the medical armamentarium.

How are small-molecule drugs described chemically?  

Small-molecule drugs are chemically described using a variety of methods, one of the most common being the Simplified Molecular-Input Line-Entry System (SMILES). SMILES is a text-based format that represents the structure of a molecule using a series of characters and symbols. It is a powerful tool for representing and manipulating chemical structures and is widely used in cheminformatics and drug discovery.

SMILES notation provides a compact and unambiguous representation of molecular structures. It encodes information about the atoms, bonds, and branching within a molecule. Each atom is represented by a single character, and the bonds between atoms are represented by lines or dashes. For example, the SMILES notation for methane, the simplest hydrocarbon, is “C”. This notation indicates that methane has one carbon atom bonded to four hydrogen atoms.

The SMILES notation can also be used to represent more complex molecules, such as those found in small molecule drugs. For example, the SMILES notation for aspirin, a common pain reliever, is "CC(=O)OC(=O)C1=CC=C(C=C1)C(=O)O". This notation indicates that aspirin has a benzene ring with two carbon atoms bonded to oxygen atoms, a methyl group bonded to one of the carbon atoms in the benzene ring, and a carboxylic acid group bonded to the other carbon atom in the benzene ring.

SMILES notation is not only used to represent the structure of small-molecule drugs but also to facilitate drug discovery and development. By encoding the structural information of molecules in a standardized format, SMILES enables the rapid searching and comparison of compounds in large chemical databases. This allows researchers to identify potential drug candidates, design new drugs, and optimize the properties of existing drugs.

How do scientists find, design, and develop small molecule drugs to treat diseases? 

When discovering, designing, and developing small molecule drugs, scientists must carefully consider various factors, including target engagement, ADME (absorption, distribution, metabolism, and excretion) properties, toxicity profile, and PK (pharmacokinetics) to ensure the effectiveness and safety of their drug candidates.

Target engagement refers to the interaction between a small molecule drug and its intended molecular target, typically a protein or enzyme involved in the disease process. Identifying and selecting the right target is crucial for drug development, as it determines the drug's specificity and efficacy.

Once promising drug candidates are identified (“hits”), scientists develop the candidates further by evaluating their ADME properties for efficient delivery to the target site, and their toxicity profiles. This involves designing drugs with appropriate solubility, permeability, and metabolic stability. Scientists evaluate the drug's absorption from the gastrointestinal tract or other routes of administration, its distribution throughout the body, its metabolism by enzymes, and its excretion from the body. In order to quantify these properties, scientists typically experiment across a large battery of in vitro and in vivo models before advancing the drug candidates to human trials.

Causes of small molecule drug failure before reaching patients

Over 93% of small molecule drugs fail clinical trials. The “translational gap” in drug development refers to this routine failure of the bench-to-bedside process – literally the lack of successful drug candidates that make it all the way through discovery and research to clinical implementation in patients.

The winnowing of small molecule “hits” to preclinical “leads” for experimental evaluation is typically based on the strength of target engagement and chemical similarity to compounds that have known favorable ADME characteristics. But while target engagement strength and chemical similarity are useful, they cannot predict the nuanced biological interactions of a novel compound. Indeed, this approach often picks small molecule compounds that will ultimately fail even though they were determined to be the best in the initial phases: a case of optimizing too soon for a subset of properties. The resulting small molecule “leads” are then further winnowed to a few candidates optimized and formulated for first-in-human dosing, based on extensive in vitro and in vivo (animal model) studies. But as with good target engagement and chemical similarity, animal model similarity can only go so far, resulting in just 56% of Phase I candidates progressing to Phase II clinical trials, and only 28% advancing to Phase III. 

How can AI assist in the development of successful small molecule drugs?

Small molecule drug developers can now use AI to analyze vast datasets of existing drugs and drug candidates for radical new insights. By identifying patterns and relationships within these datasets, AI can help scientists design new small molecule drugs with enhanced properties, such as improved efficacy, reduced side effects, or better pharmacokinetic profiles.

AI enables the prediction of interactions between small molecule drugs and their target proteins. This predictive capability provides scientists with valuable insights into the mechanisms of action of the drugs and allows them to predict potential side effects. In addition, machine learning techniques can be combined with accurate mathematical representations of biology resulting in powerful new digital twins of patients, and animal models. This enhanced hybrid AI/biosimulation is improving the discovery and translation of potential drugs into clinically successful candidates by more accurately predicting dosing, toxicity, off-target effects, drug-drug-interactions, key biomarkers, clinically relevant patient strata, and more. 

BIOiSIM® as a hybrid AI-based drug development platform differentiates itself in that its Translational Index™ (think of it as a “credit score” system for drug candidates) can be generated from only a small molecule chemical structure and a known disease target/pathway. Generally, computational chemistry and statistical modeling can predict binding affinity and certain aspects of ADME, but with limited accuracy. More importantly, they do not anticipate richer aspects of pharmacodynamics, nor toxicity. The Translational Index (TI) score combines AI, disease biology, physiology, and drug chemistry to pull all of these components together into a holistic representation of translatability. The TI is also unique in that it provides this insight across various animal models to anticipate variation in translation when performing in vivo studies.

The fact that the Translational Index score can be generated just from SMILES structure and disease target/pathway means that the need for in vitro high throughput screening is dramatically reduced, if not eliminated; furthermore, the TI is helpful in ranking entire libraries of hits/leads, saving drug development companies time and money. 

The following case study demonstrates how the Translational Index correctly predicts the clinical success of drug candidates. VeriSIM Life blindly analyzed 20 drugs studied in clinical trials for the treatment of schizophrenia, using only the drug’s SMILES structure, resulting in Translational Index scores (1 - 10). 93% of drugs with TI scores above 6 proved worthy of FDA approval, while none of the drugs failing approval ranked in the top 2 Translational Index quintiles. 

Translational Index was able to accurately predict translational success for diabetes drugs in a similar blind, retrospective analysis: 

The figures above showcase the power of Translational Index: accurately predicting the likelihood of clinical success from just the SMILES string and knowledge of the target/pathway! Remarkably, Translational Index delivers this differentiation on drug candidates that had already undergone the rigorous development phase and entered clinical trials.

As powerful as the Translational Index score is, BIOiSIM goes even further. It also combines the power of  Translational Index with Generative Chemistry to design molecules with the highest chance of translational success. VeriSIM Life’s AtlasGEN™ Novel Drug Designer utilizes Generative Chemistry and Translational Index to uniquely combine chemical discovery with biological validation, dramatically compressing drug candidate selection and reducing costly experimental research, while increasing the likelihood of clinical success. By using highly efficient binding models across a chemical search space of more than 1 trillion compounds, researchers can quickly discover therapeutically engaging hits. But unlike other drug design solutions, AtlasGEN’s integration with VeriSIM Life’s Translational Index simultaneously ranks and filters compounds to those most likely to succeed in clinical trials, saving years of in vitro investigation.

Other AI-based methods of small molecule drug design and development such as computer-aided drug design, and structure-based drug design have their benefits, but they also come with their own unique challenges such as: 

1. Inadequate search libraries: Aggregating 3rd party chemical compound collections can be expensive, and net only 10 million (10⁷) molecules from the total possible universe of 10⁶³ compounds. Library scale strongly correlates to successful molecule design.
2. Synthesis feasibility: Molecular structures designed by an unconstrained generative AI engine can create physically improbable drugs, or drugs that are so complex they might be considered unnatural or impossible to successfully reproduce.
3. High computational time and cost bottlenecks: Predicting the binding affinity of novel structures to their intended target is a critical determinant to efficacy. But molecular docking techniques are computationally intense and impractical when applied to search spaces in the trillions of compounds. 
4. Lack of biological validation: Computational chemical synthesis may identify structurally promising molecules based on target engagement. But their translatability to biology may only be considered referentially based on similar known molecules. 
5. Explainability: How AI systems result in certain drug molecule designs can be a “black box” which makes it difficult to understand the underlying characteristics of these molecules and extend them for other purposes.

AtlasGEN™ Novel Drug Designer is the first and only platform that leverages generative AI models to rapidly generate novel drug compounds with a focus on comprehensive target engagement and biological translation. It reduces the time and cost associated with other computational hit discovery approaches, while compressing the hit-to-lead refinement and optimization process by pre-validating hits for translatability using VeriSIM Life’s groundbreaking Translational Index™ technology.