TL;DR:
- Deep Apple, a startup, aims to revolutionize small-molecule therapeutics through AI-generated virtual library screening.
- Backed by $52 million from Apple Tree Partners, the company combines cryo-EM, deep learning, and molecular docking for rapid lead optimization.
- Deep Apple’s team comprises experts in computational chemistry, machine learning, and biology.
- Their innovative approach involves deep machine learning to enhance 3D mapping and identify pockets for screening.
- The in-house virtual library, Orchard.ai, prioritizes compounds using a proprietary scoring algorithm.
- Deep Apple focuses on integral membrane proteins, particularly GPCRs, with applications in various medical fields.
- They plan to unveil their first clinical candidate for inflammation in Q2 2024, with further targets in metabolic diseases.
- Deep Apple emphasizes rational deep learning models for contemporary drug discovery challenges.
Main AI News:
In the world of cutting-edge therapeutics, a unique collaboration has given rise to a groundbreaking startup. Picture this: a cryo-EM and GPCR expert, a pioneer in virtual screening, and a virtuoso in creating virtual chemical libraries converging with a common goal. This scenario isn’t the opening line of jest but the genesis of Deep Apple, a startup poised to revolutionize the discovery of novel small-molecule therapeutics through the virtual screening of AI-generated libraries.
Backed by a substantial Series A investment of $52 million from the renowned life sciences venture capital firm Apple Tree Partners, Deep Apple sets out to create a discovery engine. This engine ingeniously combines ensemble cryo-EM, deep learning, and molecular docking screens, resulting in ultra-large libraries. Their ambitious claim: lead optimization in under a year, alongside the ability to delve into biological target signaling previously unreachable by conventional discovery methods.
Spiros Liras, PhD, Founding CEO of Deep Apple and a Venture Partner at Apple Tree Partners, highlighted a recent [Journal of Medicinal Chemistry] publication revealing that only 1% of clinical candidates originate from virtual screens. Rather than perceiving this as a limitation, Deep Apple views it as a profound opportunity for a paradigm shift and a clarion call for innovation.
Deep Apple’s drug discovery engine leverages the knowledge and technologies of its three esteemed academic co-founders: Georgios Skiniotis, PhD from Stanford University, an expert in cryo-EM and GPCR structural biology; Brian Shoichet, PhD from the University of California, San Francisco (UCSF), a pioneer in virtual screening; and his UCSF colleague John Irwin, PhD, a maestro in computer library curation responsible for the popular ZINC free virtual library boasting more than 10 billion drugs.
The Deep Apple team comprises over 20 experts, including a cadre of computational chemists and machine learning engineers well-versed in molecular dynamics and simulations. Notably, the head of drug discovery is a medicinal chemist, and the team is actively assembling biologists for in-house designed in vitro pharmacology assays.
Going straight to the core of their innovation, Liras explained that raw two-dimensional cryo-EM data is often filtered by the human eye, resulting in data loss. Deep Apple harnesses deep machine learning to scrutinize all raw two-dimensional cryo-EM data comprehensively. This process accelerates the creation of higher-quality 3D maps and extracts dynamic information, such as stability or transience of conformational states.
This treasure trove of data aids Deep Apple in identifying cryptic or fleeting pockets that can stabilize target proteins for screening. Their in-house virtual library collection, Orchard.ai, primarily features novel compounds derived from GPCR subfamily models. To select and prioritize compounds, Deep Apple employs a proprietary large-scale docking-based scoring algorithm.
Orchard.ai was developed for several reasons, one being the inadequacy of pre-existing virtual libraries, like the ZINC virtual library, for specific needs. Deep Apple only produces compounds for physical screening when their computational approach convinces them of intrinsic biological value.
While Deep Apple’s platform is versatile, it finds its niche in targeting integral membrane proteins, including receptors, transporters, and ion channels, with a primary focus on GPCRs. The company is actively advancing multiple programs centered on GPCR modulators, a well-established target class with applications spanning metabolic disorders, inflammation, immunology, and endocrine diseases.
Deep Apple is in the nascent stages of its drug discovery process, boasting a portfolio of seven targets. Their aim is to unveil their first clinical candidate for an inflammation target in the second quarter of 2024. Additionally, they have plans to introduce more clinical candidates in early 2025, with a particular emphasis on targets related to metabolic diseases and weight loss management.
Avoiding the label of “just another AI company,” Liras emphasizes that Deep Apple applies rational deep learning models to tackle contemporary drug discovery challenges. He sees the genesis of Deep Apple as a response to the realization that biological signaling offers a unique avenue for drug discovery.
In Liras’ words, “We are aware that deep learning can accelerate a lot of the computation. It’s a very rational way to exploit all the data that we get from structures that seem capable of inducing a particular signaling cascade.”
Conclusion:
Deep Apple’s approach to AI-driven therapeutics signifies a significant shift in drug discovery, promising faster and more effective solutions. By harnessing deep learning and innovative screening methods, they are poised to make a substantial impact on the pharmaceutical market, potentially accelerating the development of novel therapies for a range of medical conditions.
