Where AI Is Heading and Why X‑Chem Is Positioned to Lead
The use of AI to increase the efficiency of drug discovery has increased considerably over the past few years. Drug discovery can now benefit from tools and methods that can accelerate the discovery of novel chemical entities, aid the process of optimizing chemical properties and even predict lab experiments before taking the compounds to the lab. These advanced tools ultimately contribute to significant reduction in the cost and time associated with running a drug discovery campaign.
Despite the remarkable contributions AI has made to drug discovery, the process remains challenging. Success stories around AI-designed drugs remain very limited. Here are four primary reasons for AI’s slower adoption cycle and success rate:
AI for drug discovery is a multidisciplinary endeavor. While AI is a strong tool, successfully applying it to drug discovery is a fine art that requires careful integration of three disciplines: AI, chemistry and engineering. Models and methods need hiqh-quality chemistry data to produce meaningful output, and the process of building an end-to-end drug discovery pipeline requires a scalable infrastructure supported by high-performance computing.
Data quality and quantity matters. AI is no panacea; the quality of the output is defined by the quality of the data put into the model. This is perhaps more critical in chemistry than in any other domain due to what’s called activity cliffs — the strong correlation between small modifications in chemical structures and compound activities. This makes it crucial that the models see diverse, clean datasets that are well annotated and cover many possible variations. A lot of the data available in the public domain lack at least one of these aspects. They’re either clean but small or large, but far from machine learning (ML)-grade.
Negative data is crucial. In machine learning, we talk a lot about data balancing and how important it is to have a representative sample of negative as well as positive samples to train a good ML model. This naturally extends to the application of ML in chemistry. Unfortunately, negative data, in chemistry, especially in the early hit identification stage, is hard to come by. Published literature only discloses positive results which makes it challenging to build a good model without investing in new methods and/or paying for lab testing.
Research does not necessarily translate. For many of the above reasons, most published work, especially around model accuracy, does not translate when put in production. Publicly available datasets are relatively small and lack diversity, and while they form the main testbed for AI innovation and serve as a benchmark for comparison, the results are not necessarily conclusive. Internal benchmarking with industrial-grade data is mostly required.
Building an Integrated Data Factory
The above factors are central to how we think about what’s next for AI in drug discovery and where the main hurdles lie. They also form the foundation for how we think about the capabilities and the values behind GlamorousAI’s technology. We focus our efforts on solving data challenges that hinder AI applicability, and we have made great strides in developing methods that are robust, data efficient and scalable.
Our approach to building AI for drug discovery, our vision for the field and how AI can accelerate the path to cures form the guiding campus that brought X-Chem and GlamorousAI together. The synergies between the two companies and the capabilities the merger brings to bear are immense.
As an integrated company, we’re a massive data factory providing an unprecedented depth and breadth of clean and diverse data to fuel our AI drug discovery engine. Our combined forces allow us to train better, more robust models that we use to transform every step of preclinical drug discovery from hit identification to preclinical candidate.
Our mission is to share our innovation with the world. With our platform, ArtemisAI, we are building a seamless, intuitive interface that makes our best models and innovative technologies accessible by every drug hunter. Our ultimate aim is to cure disease, and we believe that our technology, put in the hands of drug experts, will enable a step change in drug discovery.
In 1992, the late professors Brenner and Lerner were the first to publish the concept of encoding synthetic chemistry with...
Over the last 5 years, biopharma has fully embraced the use of DNA-encoded libraries (DELs) for hit identification for all...