Once upon a time, drug discovery was limited by the imagination of the scientists involved. But Artificial Intelligence (AI) / Machine Learning (ML) is pushing the old boundaries of drug discovery in exciting ways. A recent event hosted in Cambridge, UK by Johnson & Johnson Innovation – JLABS and the Innovation Forum illustrated the shape of the future for AI/ML enabled drug discovery.
The amount of biological data available is growing rapidly. This growth is partly driven by increasing use of data-intensive technologies like whole genome sequencing in healthcare and the digitisation of patient records. High throughput, low cost methods of gene expression profiling are also creating large publicly available datasets, such as the NIH backed LINCS L1000 gene expression database. At the same time, advances in AI/ML provide new tools with ever increasing potential for leveraging complex datasets. Together, these developments are leading to more accurate predictive models and increased efficiency in drug discovery.
Lifebit are one company expanding the utility of AI/ML in drug discovery. They can do so by ensuring disparate datasets are made comparable before being used by AI/ML. This involves focussing on the quality, accessibility and interoperability of data through their proprietary platform. Expanding the available input datasets for an AI/ML program can increase the reliability of its outputs. Their platform can also efficiently employ Deep Learning methods even when data are distributed by scaling from high performance computing (HPC) to Cloud computational resources.
The Milner Therapeutics Institute, University of Cambridge, are at the forefront of connecting the intellectual know-how of the academy with the drug development capabilities of the pharmaceutical industry. The computational methods and tools they develop will be used in (1) identification of new therapeutic targets in a variety of diseases, (2) stratification of patients to improve personalised medicine and (3) prediction of efficacy and safety of new and existing drugs, thus allowing the identification of drug positioning/repositioning opportunities. Collaboration is expected to be a key component of innovation in cross-sector opportunities opened up by AI/ML. The inter-disciplinary environment at the Institute will enable dynamic translation and validation of findings from in silico to in vitro models, and from one therapeutic area to another.
The trend towards collaboration and aggregating input data is reaching new heights through the “MELLODDY” (Machine Learning Ledger Orchestration for Drug Discovery) consortium. Traditional competitors in the drug discovery space are working together to unlock the maximum potential of their datasets. The consortium involves ten pharmaceutical companies. These include Amgen, AstraZeneca, GSK, Janssen Pharmaceutica NV, Merck KgaA and Novartis. NVIDIA provide AI computing expertise. MELLODDY is supported by the Innovative Medicines Institute (IMI); it is interesting to note that EU wide funding bodies are encouraging AI and data driven projects. The consortium is using a federated learning approach, giving ML the benefit of the datasets from all ten pharmaceutical companies. Key to the consortium’s viability is the blockchain technology of project coordinators Owkin. Blockchain architecture ensures the proprietary datasets themselves are never shared.
A fascinating, but potentially controversial next step for federated learning approaches to drug discovery could be the secure application of AI/ML to patient data to drive tomorrow’s healthcare advances. The addition of blockchain could be a crucial step to ensuring healthcare data privacy.
Ultimately, the pharmaceuticals industry will translate the benefits of AI/ML enhanced drug discovery into advances in public health. Patents are a powerful commercial tool. Those looking to commercialise the fruits of AI assisted drug discovery will be considering how to leverage the patent system to their advantage. But under the current approach of the European Patent Office (EPO), fundamental advances in AI itself are not patentable. European patenting strategies therefore need to focus on the implementation or purpose of an AI invention. AI inventions with an application in drug discovery may be protected using the patent system. Doing so will require patent attorneys able to draw on cross-sector experience from life sciences and chemistry to software.
Acknowledgements
I am grateful to Dr Namshik Han, Head of Computational Biology at the Milner Therapeutics Institute and Rishabh Shukla of Lifebit for reviewing drafts of this article.
If you would like to have a word with Michael about this or any of his recent articles, don’t hesitate to get in touch.