AI in drug discovery and its clinical relevance

Abstract

The COVID-19 pandemic has emphasized the need for novel drug discovery process. However, the journey from conceptualizing a drug to its eventual implementation in clinical settings is a long, complex, and expensive process, with many potential points of failure. Over the past decade, a vast growth in medical information has coincided with advances in computational hardware (cloud computing, GPUs, and TPUs) and the rise of deep learning. Medical data generated from large molecular screening profiles, personal health or pathology records, and public health organizations could benefit from analysis by Artificial Intelligence (AI) approaches to speed up and prevent failures in the drug discovery pipeline. We present applications of AI at various stages of drug discovery pipelines, including the inherently computational approaches of de novo design and prediction of a drug's likely properties. Open-source databases and AI-based software tools that facilitate drug design are discussed along with their associated problems of molecule representation, data collection, complexity, labeling, and disparities among labels. How contemporary AI methods, such as graph neural networks, reinforcement learning, and generated models, along with structure-based methods, (i.e., molecular dynamics simulations and molecular docking) can contribute to drug discovery applications and analysis of drug responses is also explored. Finally, recent developments and investments in AI-based start-up companies for biotechnology, drug design and their current progress, hopes and promotions are discussed in this article.

Keywords: Artificial intelligence; Biotechnology; Drug discovery; Graph neural networks; Molecular dynamics simulation; Molecule representation; Reinforcement learning.

Figure 1

Applications of AI-based methods at different stages of a drug discovery pipeline. There are about 2700 known potential drug target proteins in the human body and about 9600 FDA-approved small molecule drugs , , . Machine learning can be used to identify the targeted protein, GNNs can be used for predicting drug-target interactions and binding affinity, and reinforcement learning can be used to optimize the properties of a molecule. Computer vision can determine the spatial state of the tumor microenvironment. Generative models can be employed to design new molecules, simulation-based studies can suggest properties of protein-drug complexes, such as stability and dynamics, and NLP can be used to mine the existing scientific literature for drug re-purposing, FDA review, and post-market analysis.

Figure 2

Illustration of different formats of Small Molecule Representations. Molecules can be represented as Kekule diagrams with bonds and atoms, SMILES strings (which can be converted into a one-hot encoding), and as molecular graphs, where adjacency, node, and feature matrices can be constructed.

Figure 3

Graph Neural Networks in Prediction Mode. Molecules can be represented as linear data structures, such as adjacency, node, or feature matrices. These matrices can be fed to graph neural networks to learn an embedding, which can be used to predict molecular properties.

Figure 4

Graph Neural Networks in Generation Mode. Initialization is performed to add the first atom to the empty graph G₀. A graph transition (append, connect, or terminate) is sampled and performed on the intermediate molecule structure at each step .

Figure 5

The variational auto-encoder (VAE) for de novo design of molecules with desired properties . The neural network converts the discrete input molecule into a Gaussian distribution. The latent variables are reparametrized against the mean and variance. The decoder generates a new molecule from the sampled latent space.

Figure 6

A pipeline for a molecular dynamics simulation. The MD simulation pipeline can be divided into three steps, (i) System preparation, including solvation and topology and coordinate file generation (ii) System simulation for the desired time scale (iii) System or trajectory analysis using analytical methods.

Figure 7

Molecular docking and molecular dynamics simulations can investigate the efficacy of a protein-ligand system, using binding free energy and geometrical properties.

Figure 8

Statistics of AI start-ups for drug discovery.

Figure 9

A typical learning pyramid with critical questions that must be kept in mind while developing AI applications for drug discovery.

Figure 10

Learning from various data sources can aid drug design, clinical decision support, and public health policy. The collaborative intelligence resulting from the merger of “mind and machine” is expected to improve decision-making in healthcare.