Revolutionizing Drug Discovery with AI: A Case Study with Benevolent AI

Table of Contents

1. Introduction
2. The Challenges of Pharmaceutical R&D
3. The Benevolent AI Platform
4. Using Large Language Models to Discover Novel Drug Targets
5. Addressing Transparency and Reliability Concerns
6. Tackling Data Diversity Issues
7. Successes with the Benevolent AI Platform
8. Conclusion
9. FAQ

Using AI to Discover Novel Drug Targets: A Case Study with Benevolent AI

Drug discovery is a complex and challenging process that can take up to 10 years and cost billions of dollars. Despite this, many drugs fail to make it to market, and even those that do may not work effectively for all patients. One of the key reasons for this is a lack of understanding of the underlying disease biology and a poor choice of drug target.

Benevolent AI is a clinical-stage AI drug discovery company that uses AI to decipher complex disease biology, discover novel drugs, and tackle some of the most significant challenges facing pharmaceutical R&D. In this article, we will explore how the Benevolent AI platform uses large language models to discover novel drug targets and address some of the challenges facing drug discovery.

The Challenges of Pharmaceutical R&D

Pharmaceutical R&D is a challenging and expensive process. It can take up to 10 years and cost billions of dollars to bring a drug to market. For every 100 molecules that enter clinical development, around 90 of those will fail, and over half of those failures happen in phase two when testing the molecule in the patient for the very first time. Furthermore, for the top-selling drugs, they don't actually work very effectively for a significant minority of patients taking them. A key reason for this failure is a lack of understanding of the underlying disease biology, and this often translates to a poor choice of drug target.

The Benevolent AI Platform

Benevolent AI has created a comprehensive view of disease biology with its data foundations comprising over 85 diverse data sources, including millions of Peer-reviewed papers that they interrogate with their in-house suite of AI-Based tools. The platform is incredibly versatile and can work on any disease with any drug modality. It can identify novel disease targets that support both internal programs but also external collaborations.

Using Large Language Models to Discover Novel Drug Targets

One of the methodologies that Benevolent AI uses in its suite of models is large language models. They use a Transformer architecture to systematically mask out or blank out the particular target and train the model to guess the target. The model learns a semantically rich representation of biology, which means it learns machine-readable vectors that they can use in downstream tasks such as ranking novel targets.

Addressing Transparency and Reliability Concerns

One of the drawbacks of large language models is a lack of transparency, and this is a key concern for drug discovery colleagues. Benevolent AI uses a model to surface evidence to support a particular target, and they embed a sentence alongside a particular target, thereby providing support and mitigating some of the concerns about lack of transparency. They also recognize that models are not 100% reliable, and they have the human expert firmly in the loop, augmenting the human expert, not replacing them.

Tackling Data Diversity Issues

There is a huge diversity problem with the data that Benevolent AI typically trains on. They have built a tool that allows them to quantify diversity in data and set up their data diversity initiative to highlight and tackle some of these key challenges.

Successes with the Benevolent AI Platform

Benevolent AI's entire pipeline, which currently consists of 15 named platform programs and over 10 exploratory stage programs, is 100% platform-generated. They have assets that are even in clinical development, and they have a thriving collaboration with AstraZeneca that's into its fourth year now. The platform has delivered five targets for this collaboration to date in a variety of different diseases. During the Height of the COVID-19 pandemic, they were able to use their platform to identify an existing FDA-approved drug that could be repurposed for COVID-19, and this drug was approved to treat COVID-19 patients in ICU that had a very poor prognosis and actually reduced mortality by 40%.

Conclusion

Benevolent AI's platform is a unique and differentiated approach to drug discovery that harnesses AI to decipher complex disease biology, discover novel drugs, and tackle some of the most significant challenges facing pharmaceutical R&D. By using large language models, they can systematically identify novel therapeutic targets and address some of the challenges facing drug discovery. They have had many successes with their platform, and they Continue to push the boundaries of what is possible in drug discovery.

FAQ

Q: What is Benevolent AI? A: Benevolent AI is a clinical-stage AI drug discovery company that uses AI to decipher complex disease biology, discover novel drugs, and tackle some of the most significant challenges facing pharmaceutical R&D.

Q: What are the challenges of pharmaceutical R&D? A: Pharmaceutical R&D is a challenging and expensive process. It can take up to 10 years and cost billions of dollars to bring a drug to market. For every 100 molecules that enter clinical development, around 90 of those will fail, and over half of those failures happen in phase two when testing the molecule in the patient for the very first time.

Q: How does Benevolent AI use large language models? A: Benevolent AI uses a Transformer architecture to systematically mask out or blank out the particular target and train the model to guess the target. The model learns a semantically rich representation of biology, which means it learns machine-readable vectors that they can use in downstream tasks such as ranking novel targets.

Q: What successes has Benevolent AI had with its platform? A: Benevolent AI's entire pipeline, which currently consists of 15 named platform programs and over 10 exploratory stage programs, is 100% platform-generated. They have assets that are even in clinical development, and they have a thriving collaboration with AstraZeneca that's into its fourth year now. The platform has delivered five targets for this collaboration to date in a variety of different diseases. During the height of the COVID-19 pandemic, they were able to use their platform to identify an existing FDA-approved drug that could be repurposed for COVID-19, and this drug was approved to treat COVID-19 patients in ICU that had a very poor prognosis and actually reduced mortality by 40%.