Avicenna Introduces Machine Learning-Enhanced Medicinal Chemistry Platform to Accelerate the Last Mile of Small Molecule Drug Discovery

Avicenna Biosciences today introduced an extension to its machine learning (ML) technology platform to enhance medicinal chemistry and expedite clinical-stage drug discovery. The company has raised $14.5 million in funding to date, with DCVC Bio leading its 2022 seed round, and this month published a paper in the peer-reviewed Journal of Chemical Information and Modeling.

Co-authored with researchers from Schrödinger and Microsoft Research AI4Science, the paper outlines how combining Schrödinger's physics-based methods with Avicenna's novel ML methods can make the lead-to-drug optimization phase of small molecule drug discovery faster, less expensive and more successful -- particularly when it comes to engineering potency and selectivity against a potential biological target.

"We're accustomed to hearing scientific success stories, but the countless failures that happen along the way often get overlooked. In medicinal chemistry especially, failure is prevalent. It can require hundreds of millions of dollars across many clinical attempts to bridge the complex gap from chemistry and biology to medicine, and successfully develop an approved drug," said Dr. Thomas Kaiser, co-founder and Chief Scientific Officer at Avicenna. "Avicenna is applying novel ML methods to navigate the unknowns in medicinal chemistry and design around them. Our methods allow a drug design team to learn from years of failure and optimize against validated targets more efficiently. We're making the most critical phase of drug design --- the last mile of drug discovery --- substantially faster and more cost effective."

Technology Applications and Results

Avicenna is leveraging its technology to develop its own therapeutic programs, with an initial focus on neurodegenerative diseases. For example, Rho kinase (ROCK) inhibitors demonstrate potential in neurodegeneration and metabolic diseases. However, designing an orally dosed, central nervous system-penetrant ROCK inhibitor has proved extremely difficult. Fasudil, a promising ROCK inhibitor currently in Phase 2 clinical trials, must be dosed intravenously twice daily, precluding its use in illnesses like chronic kidney disease or neurodegeneration. To overcome these challenges, Avicenna has initiated its own ROCK Inhibitor Program. Using its novel ML technology to identify drug-like compounds with desired pharmacokinetic properties, the company has already achieved:

• Faster Timelines: 9 months from idea to in vivo proof of concept
• Reduced Costs: $220,000 from concept to initiation of Investigational New Drug (IND)-enabling studies
• Better Therapeutics:__Two development candidates discovered while only synthesizing 11 total compounds

Research Paper Implications

"Creating new medicines and safely delivering them to the people who need them is incredibly difficult and full of risk. Avicenna is working to make drug discovery straightforward and fast by creating new algorithms to identify molecules with ideal drug-like properties," said Dr. John Hamer, Managing Partner at DCVC Bio. "Avicenna's collaborative research with Schrödinger and Microsoft demonstrates how a physics-based augmentation of ML requires just tens of molecules to optimize small molecules against a new drug target, as opposed to thousands. We couldn't be more excited to support this potentially game-changing biotechnology company as it scales to its next stage of growth."

Titled "FEP-Augmentation as a Means to Solve Data Paucity Problems for Machine Learning in Chemical Biology," Avicenna, Schrödinger, and Microsoft's newly published paper describes how to use physics-based methods such as Schrödinger's free energy perturbation technology, FEP+, to generate virtual data, which can be used to augment the sparse data sets commonly found in early medicinal chemistry optimization. These augmented datasets can be used for ML training where it was previously impossible to train without the disclosed approach. The augmentation is comparably informative for ML training as going through the effort and expense of making and testing the needed compounds. Ultimately, this allows an early-stage discovery team to access ML which can then quickly query millions of lead-like compounds and identify promising leads for drug development. The paper outlines the key mechanistic considerations for augmenting such data sets and lays the groundwork for implementation, demonstrating that an initial series of 10-20 related compounds, accompanied by 3D structures co-resolved with a small set of ligands, can serve as an accelerated foundation for lead optimization.

The research shows that combining FEP+ with ML can accelerate the hit-to-lead phase of drug discovery, which can generate profound implications, including:

• Shorten the time spent in hit-to-lead optimization
• Significant reduction in synthetic effort

Team and Partnership Opportunities

The Avicenna team's collective background in mathematics, chemistry and medicine brings a uniquely comprehensive perspective to their core discipline of medicinal chemistry. Co-founders Drs. Kaiser and Pieter Burger met while working within theLiotta Research Group at Emory University. The group is well known for its successes in drug development across virology, oncology and neurology, creating more than 20 FDA-approved therapeutics. As a synthetic organic chemist, Dr. Kaiser led the Liotta antivirals group, and as a structural bioinformaticist, Dr. Burger led its computational group.

In addition to developing its own therapeutic programs, Avicenna partners with clinical-stage biotech startups and major pharmaceutical organizations to optimize their drug discovery campaigns and maximize risk-return profiles, easily fitting into partners' existing chemistry workflows. For more information on Avicenna and its technology platform or to inquire about partnership opportunities, visit www.avicenna-bio.com.