Conventional methods of high throughput screening (HTS) can only evaluate a small fraction of the vast number of commercially available molecules, limiting the potential for discovering new therapeutic applications. However, with the emergence of artificial intelligence (AI) and machine learning (ML) tools, there is hope for a more efficient and effective approach to drug discovery. A recent study from the Atomwise AIMS (Artificial Intelligence Molecular Screen) initiative demonstrates the potential of computational screening as a viable alternative to physical HTS in the early stages of small molecule drug discovery.
Atomwise, a technology-enabled pharmaceutical company, employs deep learning techniques for structure-based drug design. Their proprietary AI/ML platform, AtomNet, takes a virtual HTS approach by searching a chemical library containing an immense number of synthesizable compounds, surpassing 15 quadrillion in total. This approach allows AtomNet to explore new chemical space and identify potential hits.
The study, titled “AI is a viable alternative to high throughput screening: a 318-target study,” showcases AtomNet’s abilities by applying it to 318 different targets. These targets were identified through collaborations with over 250 academic labs across 30 countries. AtomNet successfully identified structurally novel hits for 235 out of the 318 targets, resulting in a remarkable success rate of 74%. This surpasses the success rates typically achieved through conventional HTS methods, which are estimated to be around 50%.
Abraham Heifets, the CEO of Atomwise, emphasizes the importance of venturing into novel areas of chemical space in order to develop meaningful treatments. The ability to differentiate therapies in the clinic is crucial for providing effective solutions to patients. AtomNet’s versatility allows it to handle a wide range of targets, making it a powerful tool for drug discovery. The hits it identified span across various protein classes and major therapeutic areas, including oncology, infectious disease, neurology, immunology, and cardiovascular disease. Enzymes accounted for the largest proportion (59%) of the target protein classes, followed by GPCRs, transporters, ion channels, and DNA/RNA-binding proteins.
Noteworthy breakthroughs from the AIMS initiative include the identification of the first reducer for Miro1, a new target for Parkinson’s disease treatment. In addition, AtomNet successfully discovered the first inhibitors for challenging deubiquitinase targets (OTUD7A and OTUD7B) related to solid and hematological tumors. Furthermore, AtomNet has identified small molecule inhibitors for CTLA-4, a well-established target in oncology.
Gregory Bowman, a professor at the University of Pennsylvania, highlights the significance of AtomNet’s success in finding hits for biologically challenging targets. He states that typical virtual screening platforms often have limited predictive power, particularly for allosteric or protein-protein interactions. However, the AIMS study demonstrates AtomNet’s high success rate in these difficult areas.
AtomNet’s success can be attributed to a unique paradigm shift from a per target model to a global model. Instead of building separate ML models for each protein target, AtomNet is pre-trained on a wide range of molecular data from the proteome. This approach allows for greater generalizability across different targets, regardless of available training data.
Looking ahead, Atomwise is entering the inflammatory disease market using AtomNet. The company aims to file an IND (Investigational New Drug) application this year for its lead candidate, a novel allosteric TYK2 inhibitor discovered through AtomNet. This breakthrough showcases the potential of AI in accelerating the development of new and innovative therapies.
FAQ
What is high throughput screening (HTS)?
High throughput screening is a method used in drug discovery to rapidly test large numbers of compounds for potential therapeutic activity. It allows scientists to evaluate the properties of thousands or even millions of molecules and identify potential candidates for further development.
What is artificial intelligence (AI) in drug discovery?
Artificial intelligence in drug discovery refers to the use of computer algorithms and machine learning techniques to analyze large amounts of data and identify potential drug candidates. AI can help researchers in various aspects of the drug discovery process, including target identification, virtual screening, and optimization of lead compounds.
What is AtomNet?
AtomNet is an AI/ML drug discovery platform developed by Atomwise. It uses deep learning algorithms to analyze molecular structures and predict their potential activity against specific protein targets. AtomNet has been trained on a vast amount of molecular data and demonstrates a high success rate in identifying hits for a wide range of targets.
How does AtomNet differ from traditional high throughput screening?
AtomNet differs from traditional high throughput screening methods in that it is a computational approach rather than a physical one. While conventional HTS relies on physically testing compounds in the lab, AtomNet uses AI and ML algorithms to virtually screen a massive chemical library and identify potential hits. This approach allows for a broader exploration of chemical space and can potentially uncover novel therapeutic candidates.
The pharmaceutical industry is constantly looking for more efficient and effective methods of drug discovery. Conventional high throughput screening (HTS) methods have limitations, as they can only evaluate a small fraction of commercially available molecules. However, with the emergence of artificial intelligence (AI) and machine learning (ML) tools, there is hope for a more innovative approach.
Atomwise, a technology-enabled pharmaceutical company, is at the forefront of using AI and ML for drug discovery. Their proprietary platform, AtomNet, employs deep learning techniques for structure-based drug design. AtomNet takes a virtual HTS approach by searching a chemical library containing a staggering number of synthesizable compounds, surpassing 15 quadrillion in total. This allows AtomNet to explore new chemical space and identify potential hits.
A recent study from the Atomwise AIMS (Artificial Intelligence Molecular Screen) initiative showcases the potential of computational screening as a viable alternative to physical HTS. The study applied AtomNet to 318 different targets, identified through collaborations with over 250 academic labs across 30 countries. AtomNet successfully identified structurally novel hits for 235 out of the 318 targets, resulting in a remarkable success rate of 74%. This surpasses the success rates typically achieved through conventional HTS methods, estimated to be around 50%.
The success of AtomNet in identifying hits for a wide range of targets is significant. Hits were found across various protein classes and major therapeutic areas, including oncology, infectious disease, neurology, immunology, and cardiovascular disease. Noteworthy breakthroughs include the identification of the first reducer for Miro1, a new target for Parkinson’s disease treatment, as well as the first inhibitors for challenging deubiquitinase targets related to solid and hematological tumors. AtomNet has also successfully identified small molecule inhibitors for CTLA-4, a well-established target in oncology.
One of the reasons for AtomNet’s success is its unique approach of using a global model instead of a per target model. Rather than building separate ML models for each protein target, AtomNet is pre-trained on a wide range of molecular data from the proteome. This allows for greater generalizability across different targets, regardless of available training data.
The success of AtomNet has led Atomwise to enter the inflammatory disease market using their AI/ML platform. Atomwise aims to file an IND (Investigational New Drug) application this year for its lead candidate, a novel allosteric TYK2 inhibitor discovered through AtomNet. This shows the potential of AI in accelerating the development of new and innovative therapies.
In the pharmaceutical industry, the use of AI and ML for drug discovery is expected to continue growing. According to a report by Market Research Future, the global AI in healthcare market, which includes drug discovery, is projected to reach a value of $19.25 billion by 2024. This growth is driven by factors such as the need for more efficient drug discovery processes, the increasing availability of big data, and advancements in AI and ML technologies.
However, there are also challenges and concerns associated with the use of AI in drug discovery. One issue is the lack of transparency and interpretability of AI models. Many AI algorithms, including those used in drug discovery, are considered black boxes, meaning they produce results without providing insights into how those results are generated. This can make it difficult for researchers to trust and validate the predictions made by AI models.
Additionally, there are ethical considerations surrounding the use of AI in drug discovery. For example, there may be concerns about bias in the data used to train AI models, which can lead to biased predictions and potentially unequal access to healthcare. Ensuring diversity and representativeness in the data used to train AI models is crucial to avoid these biases.
Despite these challenges, the use of AI and ML in drug discovery holds great promise. With continued advancements in technology and a focus on addressing the associated issues, AI has the potential to revolutionize the way new drugs are discovered, bringing hope for improved treatments for a wide range of diseases.
Related links:
Atomwise