The release of ChatGPT 3.5 by OpenAI in November 2023 has precipitated a surge of interest in the capabilities of generative API in many domains, including drug discovery. However, whilst GPT 3.5 was trained on 175 billion parameters (and GPT 4 a whopping 100 trillion – effectively the whole of the internet), in contrast, AI in drug discovery world is suffering from a data famine. The 19,000 FDA approved drugs (which are typically published with comprehensive pharmacokinetic, pharmacodynamic and other datapoints) make up a tiny subsection of pharmaceutical research, whilst the internal data generated by large pharmaceutical companies may historically not have been published to FAIR standards (requiring extensive processing to be used), and may be difficult to locate and access, even for their own employees! By far the largest publicly available dataset of experimental data, is therefore to be found within the patent literature, yet this too is not without significant challenges.

 

In this session, through case studies in antibodies and RNAi, we will explore the significant challenges - and massive opportunity - in systematically mining and extracting this information from patents, the least “FAIR” of any dataset, and leveraging it to fuel the AI drug discovery revolution.