Technological advancement apart from a deeper comprehension of disease biology, and an emphasis on breaking down conventional boundaries is driving this era of profound change in the drug development industry.
AI's evolving role in drug discovery & development
Traditional drug design methods have been successful in identifying drugs that currently treat Human immunodeficiency viruses (HIV) and cancer. With each drug taking an average of ten years to develop and 90% of drug candidates failing expensive clinical trials that can cost anywhere from $1 billion to over $2 billion per drug, researchers are searching for quicker and more effective ways to sort through possible drug molecules. With respect to this year, latest breakthroughs in AI-driven drug discoveries are discussed below.[1]
Artificial Intelligence (AI) is gaining traction in the field of drug discovery, and 2023 is the turning point for this technology. Using zero-shot generative AI in antibody creation, Absci Corporation announced at the beginning of this year that it has found three AI-created binders with a tighter binding affinity than the therapeutic antibody trastuzumab.[2] The release of MIT's DiffDock a diffusion generative model, which has a 38% success rate-higher than the conventional docking prediction techniques used in computer aided drug discovery (CADD) -could enable quicker, safer drug development. Although still in its infancy, this technique has demonstrated encouraging results in speeding up drug development and predicting the binding of proteins to ligands. By using this method, researchers at the University of Washington St. Louis are able to provide important insights into the potential efficacy of a novel drug candidate by characterizing its binding mechanism for aging-related disorders.[3]
A cloud service for generative AI-based drug development is called Nvidia's BioNeMo Cloud. Using proprietary data, researchers can refine models and carry out inference through the web or APIs. Customers of BioNeMo include AstraZeneca and a number of startups, such as Insilico Medicine and Evozyne. Evozyne declared that it had developed "supernatural" proteins using the service; that is, therapeutic proteins that may be more effective than naturally occurring ones for the rare metabolic disease phenylketonuria.[1]
Using its generative AI tool, InClinico, Insilico Medicine reported in August 2023 a substantial breakthrough in clinical trial outcome prediction. The algorithm was taught by the company using data from over 55,600 phase 2 clinical trials. Constructed over a span of seven years, InClinico's precision rate in predicting the results of actual phase 2 and 3 trials may establish the foundation for enhancing the efficiency of drug development with 35% return on investment in a virtual trading portfolio over a nine-month period.[4]
A deep learning (DL)-based framework is the University of Central Florida's known as BindingSite-AugmentedDTA model. Its goal is to improve drug-target affinity (DTA) predictions by finding possible protein binding sites as efficiently as possible. Through in vitro tests, the researchers have already confirmed the model's predictive ability. In order to anticipate the protein targets for 36 billion chemical compounds, Recursion is screening Enamine REAL Space, a searchable chemical library, utilizing Cyclica's MatchMaker technology, NVIDIA DGX Cloud supercomputing, and DeepMind's AlphaFold2 database. This method combines large-scale chemical databases and high-performance computing to assess if a tiny molecule will interact well with a protein binding site through machine learning. [1]
The first medication found by AI to enter phase 2 trials is INS018_055. Generative AI was employed by Insilico to find INS018_055, a tiny chemical that has just been included to a phase 2 research as a possible treatment for idiopathic pulmonary fibrosis. It's interesting to note that the busi...
(All Rights Reserved)