DeepMirror is on a mission to make AI powered drug design as simple as using a spreadsheet and empower every biopharma team to find better drugs faster. Ultimately, advancing the century of biology and elevating health for all. AI can now accelerate preclinical drug discovery by 2x. But less than 10% of all biopharma teams have adopted AI due to limitations in skills and trust. DeepMirror’s software helps teams deploy AI within minutes and can be used by everyone, empowering companies in their journey from discovery to clinic in a seamless user experience. Founded in 2019, DeepMirror is now used by academic institutions, biotech, and pharma businesses across the globe.
We recently caught up with Dr. Cecilia Cabrera, Application Scientist at DeepMirror, to discuss how the company uses ClearML’s open source, end-to-end platform for unleashing AI & ML.
Dr. Cecilia Cabrera
Project Manager | DeepMirror
Previous experience developing AI solutions had exposed the team at DeepMirror to the difficulty of keeping track of data, environments, and models from machine learning experiments. When DeepMirror was founded, the team decided they never wanted to experience that difficulty again. So, the team sought first to operate in an efficient and organized manner and secondly, they wanted to ensure they could re-deploy models in an easy and streamlined way.
“It was of vital importance for us to track Machine Learning experiments so we didn’t have to repeat work and therefore we could move faster and more efficiently as a company,” said Dr. Cabrera. “From the very start, we decided to log models, data, and experiments with ClearML.”
“We discovered ClearML when we were looking for easy-to-use MLOps providers. A main requirement for us when deciding on an MLOps solution was to ensure we didn’t end up with a difficult-to-use tool, such as KubeFlow, that requires a lot more knowledge and training to get started,” said Dr. Cabrera. “That’s why our main evaluation criteria was ease of use, while not compromising on features, and that’s why we ultimately chose ClearML.”
“Getting started was simple,” says Dr. Cabrera. “ClearML was incredibly easy to use. Within a few weeks we were training and deploying models to endpoints.”
The team at DeepMirror tests a variety of ML model architectures on several types of biological and chemical data, such as proteins, RNA/DNA, and small molecules. The team uses ClearML for experiment tracking and data logging for these projects. ClearML enables the team to retrieve the results from past ML experiments and compare them to current experiments when new data is generated.
Dr. Cabrera explains, “We use ClearML as an experiment tracker for our environment, data, and models, all in one easy-to-access place. Experiment tracking helps us understand how and why AI trainings go wrong and to compare experiments. Another vital feature for us is that we can run all experiments and store all the data on our own AWS infrastructure so that data never leaves our premises. As well, we have enjoyed interacting with a very active and helpful ClearML Slack community.”
With ClearML, organizations can easily develop, integrate, ship, and improve AI/ML models at any scale with only two lines of code. ClearML delivers a unified, open source platform for continuous AI. Customers can use all of our modules for a complete end-to-end ecosystem, or swap out any module with tools they already have for a custom experience. ClearML is available as a unified platform or a modular offering:
Figure 1: ClearML’s open source, end-to-end MLOps platform.
The ability to quickly retrieve and compare experiments has significantly sped up the team’s work, enhanced efficiency and reliability of their work, ultimately contributing to the company’s fast growth. The variety of datasets and approaches the team works with means a high volume of model performance metrics, biological and chemical property datasets and experimental metadata to keep track of, and ClearML has enabled them to do so effectively.
“Before ClearML we could not go back in time to past experiments and retrieve metrics or data,” Dr. Cabrera noted. “Now all of our internal experiments and data are tracked, helping us become more efficient and reliable, without compromising on security or features.”
“We have a structured way to track ML/AI performance, enabling us to run experiments at scale and to identify optimal models for our purposes in an efficient manner,” she concluded.
Get started with ClearML by using our free tier servers or by hosting your own. Read our documentation. You’ll find more in-depth tutorials about ClearML on our YouTube channel and we also have a very active Slack channel for anyone that needs help.
If you need to scale your ML pipelines and data abstraction, need unmatched performance and control, or want access to Hyper-Datasets, Role-Based Access Control, SSO and LDAP integration and other features, please request a demo. To learn more about ClearML, please visit: https://clear.ml/.