Making ML more interpretable for drug discovery
Reticular, a startup working at the intersection of AI and drug discovery, set out to make machine learning models more interpretable for scientists. Their goal was to help pharma teams discover safer compounds by unpacking how models arrive at their conclusions. By using advanced ML techniques, Reticular pursued projects that focused on unlocking and understanding the thought process of models during drug discovery. “We’re trying to open the black box,” said co-founder Nithin Parsan. “Interpretability in ML isn’t just a technical problem, it’s a scientific one. If researchers can’t understand what a model is doing, they can’t trust it.”
Fragmented systems and remote collaboration
Initially, Reticular faced significant challenges typical for a nascent startup. Reticular’s team was distributed across time zones and institutions. But collaboration on their research was crucial, and setting up a shared environment often got in the way of actual research.
“Without a unified environment, we faced a lot of hurdles in setup time and effective collaboration,” said Sophie Wang, a researcher on the team. “We were spending more time configuring systems than interpreting models.”
Their workflow, built around traditional tools like AWS SageMaker, wasn’t designed for fast, collaborative iteration. Running experiments, sharing results, and debugging code all took longer than they should have.
One platform, shared context
Having used PyTorch Lightning in earlier work, the team turned to Lightning AI to simplify their infrastructure and accelerate collaboration.
“With Lightning AI, setting up a consistent working environment was no longer a challenge,” said Parsan. “Everyone could see experiments, rerun them, debug issues, and contribute without friction.”
The shared platform let them move quickly across projects, parallelizing jobs, adjusting configs, and adapting ideas in real time. They weren’t just unblocking infrastructure. They were changing how research happened.
One thousand experiments in three months
With shared access and fewer blockers, the distributed team moved fast. They ran over a thousand experiments in just three months, an effort that helped them hit their submission deadline for ICML.
The ability to launch, tweak, and iterate on models without bottlenecks transformed their workflow from sequential to parallel.
“We couldn’t have done that volume of work in that time without the infrastructure being this smooth,” Parsan said.
50% Faster project timeline
By simplifying how they worked together, Reticular effectively cut their research timeline in half. A process that might have taken six months was done in three. Researchers were able to run experiments in parallel, contribute simultaneously, and stay unblocked, no matter where they were working from. “New collaborators were contributing within days,” Parsan said. “That kind of onboarding speed used to be unthinkable.”
The speed boost didn’t just help them hit deadlines, it gave them more room to explore, test ideas, and iterate on their models with confidence.
Reticular’s work was featured at the GEM Bio Workshop at ICLR 2025, highlighting their progress at the intersection of machine learning and drug discovery. Read their paper here.
Faster drugs means more lives saved
Reticular isn’t just trying to publish papers. They’re working to make machine learning more interpretable, so that scientists can better understand how drug discovery decisions are made, and ultimately, bring safer, more effective drugs to patients faster.
In drug discovery, where timelines stretch into years and a single insight can change the trajectory of a treatment, being able to move quickly, and with clarity, makes a tangible difference. For Reticular, tools like Lightning AI don’t just accelerate research. They enable collaboration, reduce friction, and help teams stay focused on the scientific questions that matter most.
Read more case studies and blogs:
• Sentinal 4D Advances Cancer Research with Lightning AI, Achieving 2x Faster Development
• How Columbia Labs Accelerates Neuroscience Research and Democratizes AI with Lightning AI
• View all