AstraZeneca uses PyTorch and Azure ML to speed up drug discovery

In June, Microsoft unveiled new Azure Machine Learning (ML) courses on Udacity, providing students with open source tools and frameworks such as PyTorch to help them build complex ML solutions. Last week, in an interview with us at Ignite, Microsoft Power BI CVP Arun Ulag also commented on the importance of Azure ML services in the building of end-to-end systems.

Today, research-based biopharmaceutical firm AstraZeneca has released information on how its collaboration between Azure ML and PyTorch is being utilized to speed up research in the development of new medicines.

The team behind this research believes that machine learning is the key as far as analyzing data to find relevant connections is concerned, which is why it has made use of a knowledge graphs-based approach to comprehend the relationship between networks of contextualized scientific data facts. The natural language processing (NLP) members belonging to the team preferred PyTorch for the construction of various models efficiently, adhering to the latest researches.

In combination with Azure ML, recommendation systems are created - sporting trained embeddings that can be used to map nodes in the knowledge graph to low-dimensional numeric data that can meaningfully represent the original data. These systems are then used to train use case-specific models in a coherent manner, helped through the advanced compute capabilities of Azure ML. A recent paper from AstraZeneca comparing model performance under different circumstances also utilized both Azure ML and PyTorch. The firm makes use of Azure Blob storage to handle the massive amounts of required data.

*Machine Learning model creation stages within Azure ML*

Similarly, end-to-end lifecycle management for the whole machine learning process is also made easier through Azure ML, as shown in the figure above, speeding up iterations and quickening the process of model development. The models built using these approaches are eventually used to discover and recommend "potential new and novel drug targets" in a faster and more accurate manner.

Moving forward, the firm plans to continue to expand its knowledge graph - applying machine learning through the aforementioned platforms -, with the eventual goal of delivering new medicines to the healthcare industry in a more efficient manner.