IPF is a fatal lung disease with 100% mortality, 3-5 year survival, and existing treatments costing $10,000/month with limited effectiveness.
Using Anaconda’s cross-platform AI environment, researchers built UNAGI deep learning model in PyTorch to identify nifedipine, a $10/month drug with near-equal effectiveness at 1/1000th the cost.
Dr. Jun Ding, with his team (Jingtao Wang and Yumin Zheng), has dedicated his career to an ambitious goal: to find candidate treatments for complex diseases using artificial intelligence. As an assistant professor in the Department of Medicine at McGill University and affiliate member at the Mila Québec AI Institute, Ding leads a research team that leverages AI and computational models to understand how cell states change during disease progression, with the ultimate aim of reversing those changes to restore health. Part of the team’s work has been featured in a Nature Commentary, highlighting the impact and visibility of his research in the scientific community.
When his team turned their attention to idiopathic pulmonary fibrosis (IPF), they faced a daunting challenge. This devastating lung disease causes progressive scarring of lung tissue, making it increasingly difficult for patients to breathe. With a 100% mortality rate and no effective treatments that can halt or reverse the disease progression, IPF typically gives patients just 3-5 years to live after diagnosis. Current FDA-approved drugs like Nintedanib cost patients ~ $10,000 per month without insurance and provide limited effectiveness. For the thousands of patients diagnosed with this progressive disease each year, time and affordable treatment options are both in severely short supply.
What the team discovered using the Anaconda AI Platform would prove revolutionary: a commonly prescribed hypertension medication that could potentially treat IPF with nearly the same effectiveness as existing drugs at 1/1000th the cost.
Freedom to Focus on Science, Not Infrastructure
For Ding and his PhD candidate Jingtao Wang, Anaconda has been instrumental since their early research days. The cross-platform nature of Anaconda solved a fundamental problem that plagued their multidisciplinary work.
“We’re using Linux for our development, but our users are not using Linux—they’re using Mac or Windows. So there’s always a gap, always a dependency on the platform,” explains Ding. “Anaconda provides a cross-platform infrastructure, so no matter what framework or platform you are using, you can use the same environment setup.”
Wang emphasizes how Anaconda eliminated the weeks students previously spent configuring environments: “In the early years, students had to spend one or two weeks configuring the environment before they could work on a machine learning project. But now, with Anaconda, you can create a new environment specifically for each new project.”
“If I want to summarize it with one word or one sentence, I think it’s freedom,” reflects Ding about working with the Anaconda AI platform. “We don’t have to worry about anything beyond the science. All the engineering and technical infrastructure is handled by the platform.”
If I want to summarize it with one word or one sentence, I think it’s freedom. We don’t have to worry about anything beyond the science. All the engineering and technical infrastructure is handled by the platform.”
Dr. Jun Ding
Revolutionizing Drug Discovery with AI
The team’s breakthrough came from their development of UNAGI, a deep generative learning model designed to decipher cellular dynamics and enable computational drug discovery for complex diseases. Built using PyTorch within an Anaconda environment, this sophisticated AI system represents disease progression in computational space, allowing researchers to understand why cells change and potentially reverse those changes.
This disease progresses through distinct stages, from healthy (IPF-0) through mild (IPF-1), to moderate (IPF-2), to severe (IPF-3). Using single-cell sequencing data, the team trained their advanced AI model to understand each disease stage at the cellular level, then identify drugs that could modulate gene expression to counteract IPF progression by essentially pushing severe disease cells back toward healthy states.
A Thousand-Fold Cost Reduction
The results exceeded all expectations. While existing FDA-approved drugs showed only modest effects in pushing disease cells back to healthy states, the AI model identified several superior candidates. The most promising was nifedipine, a calcium channel blocker commonly prescribed for hypertension.
“This drug was usually prescribed for hypertension, and the advantage of this drug is, first of all, it’s very safe because people have been using this for a long time,” explains Ding. “And then another advantage: it’s very, very affordable. The monthly cost is ~$10, while the cost of Nintedanib is $10,000 per month without insurance.”
Experimental validation confirmed the AI model’s predictions. “We also verified experimentally. We have done laboratory testing on living tissue samples with this new drug. The effectiveness of the new drug is almost the same as Nintedanib, at least in these laboratory tests,” Ding notes.
This groundbreaking work was published in Nature Biomedical Engineering, representing a significant advancement in computational drug discovery.
Generating High-Resolution Data at Low Cost
The team’s second major innovation addresses another critical challenge: the prohibitive cost of single-cell sequencing. While this technology provides unprecedented insights into individual cellular behavior, it costs roughly $1.5 million for 100 samples compared to just $18,000 for equivalent bulk sequencing data as per estimation of year 2023 at McGill genome center.
Wang and Ding developed scSemiProfiler, a deep generative AI tool that creates high-resolution single-cell data from low-cost bulk measurements. The AI model uses reference single-cell data as a template, then leverages bulk data differences to generate accurate single-cell profiles for target samples.
Instead of spending millions on single-cell sequencing for an entire patient cohort, researchers can now sequence just a few representative samples and use AI to generate the rest. “We’re able to generate the data with similar quality while owning maybe 10 to 20% of the cost, depending on the application scenarios,” notes Ding.
We’re able to generate the data with similar quality while owning maybe 10 to 20% of the cost”
Dr. Jun Ding
Published in Nature Communications and selected as an Editor’s Highlight (top 50 in Biotechnology and Methods), this work enables large-scale single-cell studies that were previously cost-prohibitive for most research institutions.
Streamlining Research Workflows
The team’s success demonstrates how the Anaconda AI Platform streamlines the entire research lifecycle. Every lab member has Anaconda installed, enabling easy package installation and GPU utilization for AI development.
Their GitHub repositories consistently include Anaconda installation guidance, ensuring reproducibility across different research environments. “We always recommend users first install Anaconda so that it’s pretty easy to set the correct environment,” Wang notes. “ This ensures reproducibility by specifying all package version requirements, so users always install compatible packages across systems.”
From Days to Minutes: Transforming Setup Time
Ding recalls the dramatic transformation in environment setup time: “As a student during my PhD, like 2010, setting up an environment usually took days or weeks. But now: hours, minutes.”
Wang estimates that before Anaconda, setting up a complete machine learning environment could easily take many hours—and in some cases several days—especially when dealing with complex package dependencies and platform-specific issues. Anaconda streamlined this process to minutes, enabling far broader adoption of our research tools and freeing researchers to focus on impactful discovery rather than environment troubleshooting.. This accessibility enables much broader adoption of their research tools and an increased focus on impactful research and discovery rather than administrative tasks.
Advice for Fellow Researchers
Based on their experience, the team offers clear guidance for other research institutions considering Anaconda. “Use it early,” advises Ding. “You may not realize how much you need it until you try it.”
The team also emphasizes that Anaconda’s benefits extend beyond Python. Wang notes that Anaconda handles multi-language package dependencies exceptionally well: “In base R, installing a package with complex dependencies can frequently fail in practice due to version conflicts or missing system libraries. Anaconda solves the entire dependency graph in advance — including R packages and required system libraries — so it can install everything in one step without the back-and-forth of fixing conflicts.”
For researchers publishing their tools publicly, they emphasize the importance of comprehensive GitHub documentation: “Write very detailed instructions so that people can 100% replicate the environment that you locally tried and tested.” While Anaconda makes environment sharing seamless within teams, public tool releases still benefit from clear setup guidance to help researchers worldwide adopt their innovations without troubleshooting delays.
The Platform for Life-Saving Innovation
For Ding and Wang, Anaconda represents more than just a development platform. It’s the foundation that enables their mission to save lives through computational drug discovery. By eliminating technical barriers and enabling reproducible research, Anaconda allows them to focus entirely on the science that matters.
“Without the help of Anaconda, our work will be a lot harder,” reflects Ding. Their groundbreaking research, from identifying a $10 drug that could replace a $10,000 treatment to generating million-dollar datasets at a fraction of the cost, demonstrates the transformative potential when brilliant minds have access to the right tools.
Ready to accelerate your research breakthroughs with trusted open source? Discover how the Anaconda AI Platform can transform your computational research while ensuring reproducibility and collaboration. Contact our team to learn how leading researchers are advancing science with confidence.
Anaconda AI Platform
The Only Unified AI Platform for Open Source